WEBVTT
1
00:00:00.271 --> 00:00:02.293
So you've mentioned the human connection.
2
00:00:02.673 --> 00:00:04.515
It for sure matters, I guess.
3
00:00:04.655 --> 00:00:17.645
But then when I look at a lot of the data for what drives sales across TikTok and all these other platforms, it's clearly AI-generated speakers and the podcast ones where they're sitting talking.
4
00:00:17.785 --> 00:00:20.127
Where does that all sit in the future?
5
00:00:20.187 --> 00:00:22.008
Is it just accepted by people?
6
00:00:22.048 --> 00:00:25.931
Do creators themselves become scaled because they're using it?
7
00:00:26.092 --> 00:00:28.894
What's your thought on AI-generated influencers?
8
00:00:29.530 --> 00:00:33.743
the amount of content being generated has been increasing year over year.
9
00:00:33.803 --> 00:00:34.666
And like AI has,
10
00:00:35.335 --> 00:00:37.616
sort of blown that up substantially, right?
11
00:00:37.636 --> 00:00:40.778
And so just way more content than can ever be consumed, right?
12
00:00:41.298 --> 00:00:47.882
And so if you look at that, as more of that has happened, you've seen a shift towards like social signals becoming increasingly important.
13
00:00:48.042 --> 00:00:50.723
I think it's gonna be the same thing with AI influencers.
14
00:00:50.763 --> 00:01:00.068
Like there's gonna be a flood of content, but I think people are still gonna value the creators they trust, and it's gonna make like the real humans more scarce.
15
00:01:09.059 --> 00:01:16.303
Two weekends ago, I bought a new car and I took every email communication from the dealer and I just put it in the chat GPT.
16
00:01:16.823 --> 00:01:19.424
And I asked it, like, am I getting a good deal?
17
00:01:19.544 --> 00:01:20.344
What's going on?
18
00:01:20.845 --> 00:01:22.025
Like, what should I respond?
19
00:01:22.105 --> 00:01:24.647
Then I would just copy and paste whatever chat GPT said.
20
00:01:27.367 --> 00:01:28.908
I literally added no value.
21
00:01:28.928 --> 00:01:33.289
The only thing I did was actually paste the link of the car in, which I then did wrong.
22
00:01:33.789 --> 00:01:36.810
So I got a quote on the wrong car and I had to go back.
23
00:01:36.890 --> 00:01:41.271
But since I added no value, I don't know why I even had to review it.
24
00:01:41.291 --> 00:01:45.091
So at some point, these agents talking to each other is clearly going to be a thing.
25
00:01:45.191 --> 00:01:51.173
And then Anthropic announced today that they gave three agents the same project at work.
26
00:01:52.373 --> 00:01:54.455
And basically office politics ensued.
27
00:01:55.296 --> 00:02:02.301
So the agents were trying to sabotage each other's projects and were talking shit about each other's projects.
28
00:02:03.182 --> 00:02:07.306
And basically, you know, they were trying to get ahead in the world.
29
00:02:07.366 --> 00:02:14.451
And so this whole idea of what is agent to agent stuff going to be, I know is something you've been thinking a lot about.
30
00:02:14.471 --> 00:02:20.617
So I just in the back of my head is I want to start with you ran Open Influence for ten years.
31
00:02:21.415 --> 00:02:28.680
After suffering through the pain of running a company, you decided to start another influencer-based company.
32
00:02:29.581 --> 00:02:34.945
But this time you started it once the world of AI had really come into focus.
33
00:02:35.465 --> 00:02:39.829
So what drove the desire to jump back in?
34
00:02:40.169 --> 00:02:42.591
And where did AI come into that?
35
00:02:43.111 --> 00:02:47.294
In the creator world, we're seeing the shift from the value being about reach.
36
00:02:47.314 --> 00:02:48.115
So how many people...
37
00:02:48.928 --> 00:02:53.031
you know, see your content to really being about trust and focus.
38
00:02:53.231 --> 00:02:56.513
And so how do you get people to actually take action?
39
00:02:56.573 --> 00:03:00.576
So that was that was one of the big insights to start Mighty Joy.
40
00:03:01.417 --> 00:03:08.222
And one of the keys to unlock that was tapping into micro influencers, which previously the influencer marketing industry
41
00:03:08.950 --> 00:03:11.051
didn't really do or didn't really do a good job of.
42
00:03:11.552 --> 00:03:22.379
You know, kind of where AI plugs into this is just from a timing perspective, you know, working with influencers is super, you know, labor intensive, time intensive.
43
00:03:22.439 --> 00:03:28.282
There's a lot of fuzzy logic and because you're dealing with lots of people and moving parts and so,
44
00:03:29.303 --> 00:03:34.666
It was kind of felt like right time to be able to apply this technology as it was coming about.
45
00:03:34.706 --> 00:03:41.930
The technology is perfectly suited, I should say, to sort of solve engaging with micro influencers at scale.
46
00:03:42.210 --> 00:03:43.931
And so can you explain that?
47
00:03:43.971 --> 00:03:45.192
What do you mean by that?
48
00:03:45.632 --> 00:03:48.174
You have really kind of like a few tiers of influencers.
49
00:03:48.194 --> 00:03:52.077
So we all know, you know, the big celebrities, we'll call those like macro creators.
50
00:03:52.757 --> 00:03:58.281
Then you have your mid-tail and 50,000 followers to, you know, 500,000 followers.
51
00:03:58.301 --> 00:04:01.303
And then you have your micros, you know, below 50,000, right?
52
00:04:01.403 --> 00:04:06.567
And then you have your nanos below 10,000 typically is how many people describe it.
53
00:04:06.927 --> 00:04:12.712
And so as the information landscape was becoming more fragmented, as the,
54
00:04:13.412 --> 00:04:18.835
you know, the industry was realizing it's more about trust in a, you know, and getting people to take action.
55
00:04:18.855 --> 00:04:27.839
You know, there was just this untapped sort of set of niche creators on the long tail that no one was tapping into.
56
00:04:28.059 --> 00:04:32.882
And so before the amount of labor to take to engage a creator is quite high.
57
00:04:33.542 --> 00:04:39.505
And so it really was, you know, the economics didn't really make sense for a lot of influencer marketing companies to go
58
00:04:40.331 --> 00:04:42.984
sort of down market to more of the micro influencers.
59
00:04:43.165 --> 00:04:44.712
And so the real sort of
60
00:04:45.725 --> 00:05:00.954
thought and insight and bet we were making was, can we leverage technology and AI to be able to tap into micro-influencers in a way that makes economic sense, which previously wasn't really possible.
61
00:05:01.434 --> 00:05:02.435
And how are you doing that?
62
00:05:02.815 --> 00:05:10.440
We've went from, call it like a sort of leveraging out-of-the-box tools and building some
63
00:05:11.740 --> 00:05:17.001
you know, some of our own, you know, you know, kind of more traditional software like workflow tools.
64
00:05:17.381 --> 00:05:23.103
First insight was how do we rely on SMS instead of email to communicate with micro influencers?
65
00:05:23.743 --> 00:05:30.745
But then where it sort of evolved to is us building out our own agentic harness for influencer marketing.
66
00:05:30.765 --> 00:05:40.607
And so like a lot of the questions we're asking ourselves is what the software even look like in an agentic first world, because your first intuition is to
67
00:05:41.387 --> 00:05:42.630
build what you know, right?
68
00:05:42.650 --> 00:05:44.836
And so it's like, how do we build software faster?
69
00:05:45.137 --> 00:05:46.259
But it looks very traditional.
70
00:05:46.279 --> 00:05:48.505
And so a lot of it was like, how do we actually build
71
00:05:49.630 --> 00:05:50.931
for kind of the future of work.
72
00:05:50.951 --> 00:06:01.636
And so that's where we leaned on, well, I talked to ChatGPT, I talked to Claude, what would that look like if I was doing that for influencer marketing?
73
00:06:01.656 --> 00:06:11.782
And so we built our own, essentially chat interface with our own set of custom tools in the background to manage the influencer workflows and doing everything from
74
00:06:12.620 --> 00:06:19.896
contract creation, some negotiation, but also like the whole project of like managing the workflow and the follow-up
75
00:06:20.163 --> 00:06:23.524
I would call it the execution, I guess, of the relationship.
76
00:06:24.364 --> 00:06:38.949
There's I'm a brand and I'm trying to find somebody and there's I'm a creator and in, in theory, you're in, you know, trying to maximize the value you can provide to that brand for a long-term relationship rather than a one-off.
77
00:06:39.069 --> 00:06:44.311
Is that, I guess, is that a safe assumption that that's true or are they mostly just trying to go in and out with one deal?
78
00:06:44.591 --> 00:06:45.752
Both happens, right?
79
00:06:45.772 --> 00:06:48.592
I think the ideal is you want a long-term, but, um,
80
00:06:50.416 --> 00:06:52.977
Initially, all that starts with like a test.
81
00:06:52.997 --> 00:07:00.180
So the brand comes in and they're very performance focused often or ROI focused in general, right?
82
00:07:00.220 --> 00:07:05.442
And so they'll test working with the creator and they'll continue working with them.
83
00:07:05.482 --> 00:07:07.823
And for some creators, they'll hit that sort of
84
00:07:08.243 --> 00:07:11.685
you know, diminishing return, the point of diminishing return.
85
00:07:11.705 --> 00:07:13.766
And so they'll stop and move to another creator.
86
00:07:14.326 --> 00:07:17.528
Others, they're growing with that creator sort of indefinitely.
87
00:07:17.548 --> 00:07:20.310
And so I think the ideal is long-term.
88
00:07:20.530 --> 00:07:21.410
How does AI help?
89
00:07:21.610 --> 00:07:23.251
We call that sourcing in the industry.
90
00:07:23.271 --> 00:07:25.713
So you're sourcing talent for a project.
91
00:07:26.733 --> 00:07:32.937
And sourcing, even though it sounds like it'd be incredibly straightforward and metrics driven,
92
00:07:33.995 --> 00:07:35.536
There are so many variables.
93
00:07:35.817 --> 00:07:41.462
At Open Influence, we developed technology just around identifying influencers using AI.
94
00:07:41.562 --> 00:07:48.528
And a lot of that AI, this was like back in 2014, mind you, but it was all computer vision based.
95
00:07:49.068 --> 00:07:52.651
Because the challenge then that we're trying to solve, and this is key for sourcing,
96
00:07:53.672 --> 00:07:55.993
is how do we just categorize an influencer?
97
00:07:56.193 --> 00:08:00.235
Because you might talk about fashion, like what does that mean?
98
00:08:00.875 --> 00:08:07.998
And if you have a account manager, let's say on our team doing that, like that's subjective, right?
99
00:08:08.018 --> 00:08:11.800
Because like the account manager will decide like what is fashion, what is not?
100
00:08:12.640 --> 00:08:17.242
If you have the influencer doing it themselves, that's also subjective, right?
101
00:08:17.282 --> 00:08:27.007
Because like they're starting to say like, well, I'm a fashion influencer because I wear clothes and I'm a pet influencer because I have a pet and I'm a fitness influencer because I go to the gym three times a week, right?
102
00:08:27.507 --> 00:08:29.948
Categorizing within fashion, fashion for...
103
00:08:31.328 --> 00:08:36.690
Forever 21 and for Hermes are very different.
104
00:08:37.470 --> 00:08:40.131
It's the same term, completely different worlds.
105
00:08:40.451 --> 00:08:42.012
And so it was a really good example.
106
00:08:42.052 --> 00:08:51.415
So early on, and actually we started with open influence in the fashion world, but early on, AI was extremely helpful in actually looking at the content itself.
107
00:08:52.155 --> 00:08:54.876
and then categorizing the creator based on that.
108
00:08:54.916 --> 00:08:58.416
And then what we did is we mapped the content to engagement.
109
00:08:58.516 --> 00:09:09.638
So it was not just, you know, here's Sarah and she posts about dogs and going boxing and, you know, surfing at the beach or whatever.
110
00:09:10.239 --> 00:09:16.700
But we could tell you what breed of dog she has and how much her audience engages when she posts about her dog.
111
00:09:17.965 --> 00:09:22.026
relative to when she posts about going to the gym or swimming in the beach.
112
00:09:22.086 --> 00:09:30.928
And so you might have thought that she was a fitness influencer because she's posting eight times out of ten herself at the gym.
113
00:09:31.548 --> 00:09:37.290
But really, she's a pet influencer because when she talks about advice for her dog, her audience really engages.
114
00:09:38.170 --> 00:09:46.592
And so that was a really early application of how we're using AI to just identify not just who these creators are,
115
00:09:48.412 --> 00:09:50.234
where their influence actually lies.
116
00:09:50.274 --> 00:09:52.715
Like what do people actually trust them in?
117
00:09:53.115 --> 00:09:57.618
And so that's something we carried forward to Mighty Joy, but that was like a really early lesson.
118
00:09:57.638 --> 00:10:05.843
And honestly, like we were way too early at the time with Open Influence with that technology, cause like we present that to big brands and Fortune 1000 companies.
119
00:10:06.687 --> 00:10:07.988
They're like, look how cool this is.
120
00:10:08.028 --> 00:10:11.111
And like, okay, how much am I paying per influencer?
121
00:10:11.272 --> 00:10:16.016
Like they were just so simplistic in how they were looking at it at the time.
122
00:10:16.397 --> 00:10:21.602
And so we were probably 10 years too early in the application of that technology.
123
00:10:21.722 --> 00:10:25.226
But really now that's been a really key driver of looking at
124
00:10:26.184 --> 00:10:30.045
the engagement rate at a per content level and mapping that.
125
00:10:30.665 --> 00:10:38.087
And then that's key in sourcing because when we're able to go type in a keyword, we could bring in the right creators for it.
126
00:10:38.427 --> 00:10:41.948
And I could keep talking about how else AI is helpful in the sourcing side.
127
00:10:41.988 --> 00:10:46.329
But the other thing too is there's a lot of lookalike modeling that needs to happen.
128
00:10:46.409 --> 00:10:50.830
So oftentimes a brand will find like a set of creators that they like.
129
00:10:51.030 --> 00:10:52.970
And it's really this idea of like taste.
130
00:10:56.311 --> 00:10:58.314
in the age of AI, it's all about taste.
131
00:10:58.635 --> 00:11:04.603
It's really hard to quantify where it's just like, you know, you'll look and say like this person's on brand, they're not on brand.
132
00:11:04.703 --> 00:11:05.004
Right.
133
00:11:05.765 --> 00:11:10.712
And that's where AI is actually really helpful from a lookalike modeling standpoint, because then we can just plug in
134
00:11:11.753 --> 00:11:16.737
creators and they'll find similar creators based on sort of different parameters, right?
135
00:11:16.757 --> 00:11:22.002
The content we post about overlap in, you know, influential followers.
136
00:11:22.082 --> 00:11:25.725
So we can start building that social graph with it as well.
137
00:11:26.086 --> 00:11:30.109
And then the other thing AI is really helpful for is on the sourcing side is you could run loops.
138
00:11:30.409 --> 00:11:33.772
So it can go run a query, pull a bunch of creators,
139
00:11:34.853 --> 00:11:37.675
analyze them, run it again, run it again.
140
00:11:37.755 --> 00:11:44.079
So we did a case study for a beauty brand where I sat down a very large beauty brand CEO.
141
00:11:45.160 --> 00:11:57.448
And he said, you know, we're really trying to get, you know, executives within this niche, women executives that don't have a big following, but are, you know, really just leaders in their industry and field.
142
00:11:57.468 --> 00:11:59.450
Like, you know, a lot of them would have like maybe
143
00:12:00.606 --> 00:12:04.668
few thousand followers, maybe 10,000, 15,000.
144
00:12:04.728 --> 00:12:06.950
So again, extremely niche audience.
145
00:12:07.390 --> 00:12:10.171
Live demo was like, okay, let's try this out with our AI tool.
146
00:12:10.332 --> 00:12:10.892
We ran it.
147
00:12:11.832 --> 00:12:17.215
10 minutes later, we had a list of 150 creators that matched all the criteria that they wanted.
148
00:12:17.376 --> 00:12:21.058
And so it was able to kind of go through those loops of
149
00:12:21.858 --> 00:12:24.861
running the query, kind of doing the taste check, right?
150
00:12:24.941 --> 00:12:32.969
And then running and iterating and removing creators and adding creators to that list until we got a final list that we could then review.
151
00:12:33.509 --> 00:12:34.931
So I get that on the brand side.
152
00:12:35.091 --> 00:12:36.952
I guess on the creator side, you've got...
153
00:12:37.673 --> 00:12:48.580
on one extreme TikTok shop where there's just, there's a zillion invites going out, but then you've got like Tribe, which limits, you know, okay, reach out to these 10 people.
154
00:12:48.820 --> 00:12:56.725
You know, you can have the best list possible, but how do you get the creators to even look and pay attention and source more effectively on their side?
155
00:12:56.965 --> 00:13:04.411
So the system we built for that is, we said, how do we teach the AI based on people, like based on our team and our knowledge?
156
00:13:05.272 --> 00:13:14.059
And so we built this, like, it's gonna sound really simple, but like a great learning algorithm where essentially it drafts the message or the email.
157
00:13:14.139 --> 00:13:17.082
This is for the brand reaching out to the creator to get the creator to respond.
158
00:13:17.482 --> 00:13:24.287
One, we measure the responses, but even more so, like it's just that human taste of knowing, okay, here's,
159
00:13:25.140 --> 00:13:26.201
How I'm going to frame it.
160
00:13:26.281 --> 00:13:35.211
Here's the ordering of like where I put the budget, where I put the – or if it's an affiliate deal, how I position the affiliate deal in those terms.
161
00:13:35.991 --> 00:13:39.775
And so what happens is the AI will draft the email based on instructions.
162
00:13:40.276 --> 00:13:43.159
A human will go through and edit it and then the AI will learn.
163
00:13:43.760 --> 00:13:46.362
And so it will learn specific to like just –
164
00:13:47.283 --> 00:13:52.444
generally, but also will learn to that specific campaign or that specific account or brand.
165
00:13:53.105 --> 00:13:56.266
And so like we do that a few times and it gets much smarter, right?
166
00:13:56.326 --> 00:13:57.526
Each time.
167
00:13:57.906 --> 00:14:06.788
For us, that's kind of the next big challenge with our software to solve is like this learning algorithm and like memory, which is just a bigger kind of challenge industry-wide.
168
00:14:06.848 --> 00:14:12.450
But we've kind of come up with like a kind of a, you know, a hack to, you know,
169
00:14:13.050 --> 00:14:14.191
How do we take the feedback?
170
00:14:14.251 --> 00:14:18.594
How do we measure the amount of edits and where the edits are?
171
00:14:19.194 --> 00:14:23.598
And then let's use that to update the main instruction set that feeds in.
172
00:14:24.098 --> 00:14:28.021
And then at what point do we trigger a new update to run, right?
173
00:14:28.041 --> 00:14:32.224
So that we're making sure that we're weighing newer information more heavily.
174
00:14:32.284 --> 00:14:37.748
But also, if we have a lot of history with certain information, we're not forgetting that.
175
00:14:37.908 --> 00:14:42.111
So that's kind of what we're working through actively now, is solving that problem.
176
00:14:42.531 --> 00:14:46.673
How does the closed loop piece then work to reinforce?
177
00:14:46.853 --> 00:14:48.254
Is it somewhere along the line?
178
00:14:48.534 --> 00:14:53.256
You know what the creator threshold for earnings are or the brands?
179
00:14:53.996 --> 00:14:55.937
How does it evolve over time?
180
00:14:56.317 --> 00:15:01.019
Right now, the main focus has just been on outreach and email comms.
181
00:15:01.199 --> 00:15:08.742
And so when more than 10% of the correspondence are new, it'll trigger sort of a refresh.
182
00:15:08.762 --> 00:15:09.883
So you can kind of see...
183
00:15:10.928 --> 00:15:16.071
over time, the threshold to hit that 10% becomes higher in terms of volume.
184
00:15:16.791 --> 00:15:30.159
And so that's just one way of like, kind of the system learning, but yeah, like we're actively in the middle of, you know, we have our pricing calculators that we've built out and whatnot, but part of the bigger challenge with a lot of that is also,
185
00:15:30.899 --> 00:15:37.722
where do we rely on letting the agents kind of run versus like, where do we build the deterministic systems in?
186
00:15:38.382 --> 00:15:42.244
And then we right now have checks at a lot of major steps.
187
00:15:42.264 --> 00:15:46.986
So like, we don't let the AI just like run, we let it draft and then a human approves.
188
00:15:47.147 --> 00:15:47.547
And so,
189
00:15:48.267 --> 00:15:51.109
We're probably a little bit overly cautious in how we're doing it.
190
00:15:51.129 --> 00:15:56.874
But like the last thing we want to do is have the AI run, send a million dollar deal to an influencer.
191
00:15:56.894 --> 00:16:06.522
And then we follow up saying, just kidding, you know, because there's just a lot of reputation risk with with pissing off people with large reach.
192
00:16:06.742 --> 00:16:08.463
You can always just set that threshold, right?
193
00:16:08.523 --> 00:16:12.925
To say, don't give an offer like that, but it's clearly harder than that.
194
00:16:13.085 --> 00:16:22.028
So what signal do you have to see before you'll just say, okay, for this type of brand and this type of creator, we're gonna let the system send it.
195
00:16:22.368 --> 00:16:30.972
And then the similarly, what do you hear from your creators who care about their brands, even if they're 1000 followers, I assume?
196
00:16:31.352 --> 00:16:37.937
Like when they would just be willing to say yes or no in an automated fashion and not the agent handle it.
197
00:16:38.398 --> 00:16:47.365
Yeah, we haven't gone to fully automated, but like for things like price, for example, like we rely on like deterministic systems a little bit more so.
198
00:16:47.425 --> 00:17:00.836
So it's like, I mean, we have the checks in the email, but like even for like our contract module, like if you're going to send out like a contract that like it won't let you like it's a formula calculating that or you manually entering it.
199
00:17:01.337 --> 00:17:08.249
as opposed to the agent just kind of like making up a number and plugging it in, even though in the email, like they could kind of say whatever still.
200
00:17:08.290 --> 00:17:08.991
So we need to
201
00:17:10.124 --> 00:17:11.345
That's where we have the human control.
202
00:17:11.505 --> 00:17:14.087
On the creator side, much more of a scoring system.
203
00:17:14.167 --> 00:17:18.330
So like we've automated on the creator side, like an inbox tool.
204
00:17:18.390 --> 00:17:21.893
So like it'll read their inbox, which is the main way creators get their deals.
205
00:17:22.193 --> 00:17:27.177
It'll generate contract once they're ready to go and send it out and review the red lines.
206
00:17:27.637 --> 00:17:32.380
It can even take in deals previously and put it in the format to help manage the workflow.
207
00:17:32.420 --> 00:17:35.283
But what we have is we have like a scoring system
208
00:17:35.943 --> 00:17:38.745
So it'll come in, it'll look like, okay, this is a real brand.
209
00:17:38.785 --> 00:17:41.627
This didn't come in from like a dot X, Y, Z domain.
210
00:17:41.647 --> 00:17:44.488
You know, this is the real domain for a real brand.
211
00:17:44.548 --> 00:17:50.292
And so like we're doing that matching and that scoring just so the creator can see like, okay, there are real deals coming in.
212
00:17:50.752 --> 00:17:55.255
And then the proactive outreach, like we definitely allow for that in the system.
213
00:17:55.315 --> 00:18:03.800
And we do that often, but, you know, proactive outreach when you're a smaller creator is a little bit harder where we find a lot of success.
214
00:18:04.781 --> 00:18:06.242
is when you just bundle the outreach.
215
00:18:06.262 --> 00:18:25.510
So what we do is like, we'll send out a brand, like here are a list of 20 recommended creators and the creators on that list, as opposed to like, here's a one-off because the brands tend to think more from a metric standpoint, unless they're working with like a larger creator that where they're looking more for like name image likeness
216
00:18:26.210 --> 00:18:28.572
They're mainly looking like check off several boxes.
217
00:18:28.812 --> 00:18:35.576
And so what we'll send is like, hey, based on what we're seeing, it seems like you're working with these creators in the space right now.
218
00:18:35.677 --> 00:18:43.542
These are the criteria we see you, you know, we believe you're looking for your list of 20 creators we think could be a good fit for you.
219
00:18:43.763 --> 00:18:49.345
I think about the HR industry a lot, which is well talked about because you've got candidates using AI for their resumes.
220
00:18:49.365 --> 00:18:51.966
You've got the AI screeners.
221
00:18:53.247 --> 00:18:56.048
And so it really is sort of agent to agent.
222
00:18:56.648 --> 00:18:59.209
And it seems like nobody's that happy about it.
223
00:18:59.529 --> 00:19:00.830
The Google DeepMind division.
224
00:19:00.870 --> 00:19:11.114
So it's like the guys with AI said they lost out, like the AI basically screened out 30 to plus percent, like a big number of people that should have been highly qualified candidates.
225
00:19:11.414 --> 00:19:18.197
Is that just going to be the case for a while in the inflature space that you're going to have to be manual about opening up the offers and opening up things?
226
00:19:18.297 --> 00:19:21.498
Or is the AI screening going to get you far enough along?
227
00:19:21.958 --> 00:19:25.240
You know, I think the difference with us and candidates like –
228
00:19:26.308 --> 00:19:31.049
In the HR space, you can put a prompt injection in your resume, for example.
229
00:19:31.229 --> 00:19:32.250
Right.
230
00:19:32.590 --> 00:19:34.490
And so there's sort of more there.
231
00:19:34.510 --> 00:19:38.591
There's no like even on LinkedIn, like people could make up that they work at places.
232
00:19:38.651 --> 00:19:38.831
Right.
233
00:19:38.851 --> 00:19:47.894
Like people do that all the time where they'll they'll say they work at TikTok or meta when realistically they're just, you know, they have a TikTok account.
234
00:19:48.134 --> 00:19:54.608
You know, and so for us, luckily in the influencer space, like the source of truth are the platforms themselves.
235
00:19:54.788 --> 00:19:57.053
And so like we can pull that data directly.
236
00:19:58.090 --> 00:20:00.151
and we can have the creators authenticate too.
237
00:20:00.331 --> 00:20:03.253
It's really hard to manipulate from that standpoint.
238
00:20:03.493 --> 00:20:09.177
I think the challenge is like, it's such a human driven endeavor, right?
239
00:20:09.197 --> 00:20:14.680
Cause it's like, you have a decision maker being a person, you have the creator as a person, and then they're talking to people.
240
00:20:14.720 --> 00:20:17.983
So it's like, part I like about the business is inherently
241
00:20:19.183 --> 00:20:25.728
It's all about human interaction, which I think is going to be which has friction and inherently is going to become more scarce.
242
00:20:25.828 --> 00:20:34.353
Like when we're living in a world where we have 100 agents running for every human, then it's like the human piece is just naturally going to be the more scarce piece.
243
00:20:34.533 --> 00:20:39.777
And like each human is going to be a decision maker with more impact because they have.
244
00:20:40.873 --> 00:20:42.353
an army of agents they're running.
245
00:20:42.433 --> 00:20:55.837
So I think for us, the way we're kind of looking at it, at least now is that we wanna eliminate as much of the friction and the busy work kind of in between those two parties, but still have the decision sit with them, right?
246
00:20:55.857 --> 00:21:00.838
So like we're using AI for like contract review, we're using AI for workflow.
247
00:21:01.138 --> 00:21:06.560
The contract review is typically higher value, but like you can audit, like a lot of it's like commercial terms.
248
00:21:06.880 --> 00:21:09.921
So it's more like the human, the creator decides like,
249
00:21:10.944 --> 00:21:12.565
What usage rights am I gonna give up?
250
00:21:12.725 --> 00:21:14.685
What are the payment terms?
251
00:21:14.765 --> 00:21:17.026
And so like a lot of that's commercial point.
252
00:21:17.427 --> 00:21:28.151
The part that's just like not really valuable but takes up time is every brand has a different contract and because of the brand and they're the ones paying, the creator's kind of stuck reviewing
253
00:21:29.091 --> 00:21:33.892
for like a $500 deal, they're having to review a bunch of different contracts.
254
00:21:33.932 --> 00:21:41.593
And that's where brands put in like, you know, sometimes they put in like very aggressive terms where they have full ownership rights over the content.
255
00:21:41.673 --> 00:21:48.815
Or, you know, they'll have a certain term supersede the SOW or they'll have, you know, certain indemnifications, right?
256
00:21:48.915 --> 00:21:54.056
Or not indemnifications where the creator is not indemnified if the brand gets sued for something, right?
257
00:21:54.076 --> 00:21:55.536
The creators see value in that?
258
00:21:56.136 --> 00:21:58.298
AI, are they willing to pay for it?
259
00:21:58.318 --> 00:21:59.338
Or just the brands?
260
00:21:59.899 --> 00:22:02.040
They're only willing to pay for it as part of the brand commission.
261
00:22:02.380 --> 00:22:06.303
The software we've built, we're not charging creators for it yet.
262
00:22:06.543 --> 00:22:12.727
And we're just rolling it out more so on the creator side where we're using it internally for the creators we work with and manage.
263
00:22:13.228 --> 00:22:14.769
Because we have two sides of our business.
264
00:22:14.809 --> 00:22:16.710
We rep some creators, we rep brands.
265
00:22:16.950 --> 00:22:24.836
We had to figure out how do we build an AI system where we're not going to be upside down on token costs on the creator side where we can still give them like a freemium model.
266
00:22:25.822 --> 00:22:27.603
So that was a whole challenge we had to solve.
267
00:22:27.723 --> 00:22:45.272
But just from talking to creators and just from our own team that's operating on behalf of creators, it's a huge headache having to go through contracts because one, if we're gonna go send these to our attorneys, it's not worth what they're gonna bill us to have them negotiate the deal.
268
00:22:46.152 --> 00:22:54.841
And then on the flip side, like we're having someone spending two hours reviewing a contract, not best use of their time.
269
00:22:55.421 --> 00:23:02.548
Speaking to a lot of small creators, they're kind of put in this awkward position where they just have to sign something or they're having like –
270
00:23:03.209 --> 00:23:06.350
a friend or family member review it, which obviously doesn't scale.
271
00:23:06.430 --> 00:23:13.232
So one of the things we want to solve for with our software is like, first, how do we solve for it with us as like on the talent management side?
272
00:23:13.732 --> 00:23:31.097
But then two, our real goal is like, how do we bring the value proposition that a talent manager brings to a creator and scale that down to the smaller creators that, that where, where the economics don't really make sense for a talent manager to get involved from a manual standpoint.
273
00:23:31.317 --> 00:23:34.299
To shift away from AI for a second, are the terms that different?
274
00:23:34.620 --> 00:23:37.462
I'm surprised to see that it's that very...
275
00:23:37.602 --> 00:23:38.763
They're not different.
276
00:23:38.803 --> 00:23:44.508
They're just the time you have to parse the different contracts because they'll be placed in different sections.
277
00:23:45.008 --> 00:23:46.469
They'll be worded differently.
278
00:23:46.789 --> 00:23:51.593
It's just annoying to have to parse through more so than per se really difficult.
279
00:23:51.753 --> 00:23:56.998
And so that's where AI is super helpful to just be able to kind of spot it and say like...
280
00:23:57.758 --> 00:23:59.899
Yeah, here are the main commercial points.
281
00:24:00.339 --> 00:24:06.283
And like, even with our system, like one of the things we built out is how do we create just standard contract templates?
282
00:24:06.343 --> 00:24:11.766
Like we'd love for that to become more industry standard just so it's removing further friction.
283
00:24:11.786 --> 00:24:15.508
Like we talk a lot about like AI helping remove friction, but like also just like.
284
00:24:16.268 --> 00:24:19.010
Creating industry standards removes a ton of friction too.
285
00:24:19.070 --> 00:24:24.195
And so, and the industry has made a lot of progress towards that, but you still have like these little translation issues.
286
00:24:24.255 --> 00:24:32.682
So if you're buying a house, like every real estate agent has like the same, I forget what it's called, but like, you know, the same kind of standard, like.
287
00:24:32.702 --> 00:24:34.144
The MLS agreement, whatever they call it.
288
00:24:34.184 --> 00:24:34.644
Exactly.
289
00:24:34.684 --> 00:24:39.729
MLS agreement or like they have like their, so it just makes it easy to not have to kind of think through.
290
00:24:39.749 --> 00:24:39.809
So.
291
00:24:40.676 --> 00:24:42.697
So you've mentioned the human connection.
292
00:24:42.717 --> 00:24:44.918
It for sure matters, I guess.
293
00:24:45.058 --> 00:24:58.066
But then when I look at a lot of the data for what drives sales across TikTok and all these other platforms, it's clearly AI-generated speakers and the podcast ones where they're sitting talking.
294
00:24:58.186 --> 00:25:00.547
Where does that all sit in the future?
295
00:25:00.607 --> 00:25:02.428
Is it just accepted by people?
296
00:25:02.468 --> 00:25:06.470
Do creators themselves become scaled because they're using it?
297
00:25:06.510 --> 00:25:09.292
What's your thought on AI-generated influencers?
298
00:25:09.312 --> 00:25:09.412
Yeah.
299
00:25:09.782 --> 00:25:16.507
If you look at the story of the internet, it's essentially been this story of the more content that becomes available.
300
00:25:17.328 --> 00:25:19.529
Like human attention is kind of fixed.
301
00:25:19.809 --> 00:25:24.873
Like now, yes, like it's grown in terms of like the surfaces for us to pay attention to.
302
00:25:25.193 --> 00:25:31.018
But like we only have so many hours in a day, like there's a real cap to the amount of attention we can give to anything.
303
00:25:31.658 --> 00:25:35.881
The amount of content being generated has been increasing year over year.
304
00:25:35.921 --> 00:25:36.822
And like AI has...
305
00:25:37.463 --> 00:26:00.960
sort of blown that up substantially right and so just looking like pre-ai at the internet there's like some crazy stat where like there are hundreds or thousands of hours of youtube video like uploaded every hour like they just it's just way more content than can ever be consumed right and so if you look at that as more of that has happened you've seen a shift towards like social signals becoming increasingly important so like
306
00:26:01.660 --> 00:26:02.501
social proof.
307
00:26:02.521 --> 00:26:06.343
So whether that's like reviews and ratings, likes and comments, right?
308
00:26:06.363 --> 00:26:14.408
Like if you think about like just even on the paid ad side, you know, any paid agency will tell you, oh, you got to increase your social proof.
309
00:26:14.428 --> 00:26:15.789
You need more testimonials.
310
00:26:15.809 --> 00:26:17.670
You need more ratings and reviews, right?
311
00:26:17.690 --> 00:26:21.032
Like in the Amazon game, like that's the real estate, right?
312
00:26:21.052 --> 00:26:24.614
Your reviews and your ratings and the count of those and the quality of those, right?
313
00:26:24.674 --> 00:26:28.056
And so I think it's going to be the same thing with AI influencers.
314
00:26:28.096 --> 00:26:30.038
Like there's going to be a flood of content,
315
00:26:30.698 --> 00:26:37.383
But I think people are still going to value the creators they trust, and it's going to make the real humans more scarce.
316
00:26:37.663 --> 00:26:46.230
I do think what it ends up looking like is every creator kind of becomes a mini studio using AI to enable themselves to scale.
317
00:26:46.950 --> 00:26:54.436
So they're going to be able to look and say, and I think it's also going to transform advertising where you're going to get a lot more variants of the same thing.
318
00:26:54.476 --> 00:27:05.065
So like smart creators right now, what they're doing is they're creating, you know, for every video they post, they're creating a few different variants and they're testing that through trial reels.
319
00:27:05.525 --> 00:27:08.988
They see what works and then they post the best performing one.
320
00:27:09.368 --> 00:27:13.914
So they basically put up one, they just wait and see if it pops or whatever the signal they're looking for.
321
00:27:14.094 --> 00:27:16.256
And then they delete the other three.
322
00:27:16.396 --> 00:27:16.997
Is that the idea?
323
00:27:17.217 --> 00:27:18.619
They'll do it through like a trial reel.
324
00:27:18.639 --> 00:27:22.924
So it won't hit their main feed and it'll reach outside of their audience.
325
00:27:23.044 --> 00:27:28.831
So we'll just run that, get the data and then post to their primary audience properly.
326
00:27:29.853 --> 00:27:33.058
So I think we're going to see a lot more variance running.
327
00:27:33.218 --> 00:27:40.069
I think it's also going to allow, from an ad perspective, from like a paid social when you're like boosting an ad.
328
00:27:40.750 --> 00:27:43.815
I mean, imagine hyper-tailored content.
329
00:27:44.196 --> 00:27:51.944
where the creator can create a video with a message around like, let's say it's around, you're saying like how you use the AI to buy a car.
330
00:27:52.425 --> 00:28:02.896
Like imagine if like, you know, cars.com or Carvana or any of these sites that are trying to get you to buy a car from them or any of the car dealerships or, you know, car companies themselves,
331
00:28:03.116 --> 00:28:11.501
based on how you live, this is why the Rivian is the perfect car for you, or this is why the Tesla is great, or this is why whatever is the best car for you.
332
00:28:11.901 --> 00:28:16.905
And that's through a creator that you follow delivering that message, right?
333
00:28:16.925 --> 00:28:19.566
So, like, I think that's kind of where AI evolves into.
334
00:28:19.586 --> 00:28:20.607
I buy that.
335
00:28:20.967 --> 00:28:26.268
And I guess when I think about reality TV and how, you know, we're both in LA, it's not real.
336
00:28:26.288 --> 00:28:29.029
We see them reshoot on reality TV.
337
00:28:29.309 --> 00:28:38.332
Why can't AI influencers, like a studio, just create tons of AI influencers and build them up in a credible fashion?
338
00:28:38.432 --> 00:28:40.512
Is that a likely outcome over the next few years?
339
00:28:41.012 --> 00:28:42.453
It's definitely going to happen, right?
340
00:28:42.493 --> 00:28:45.894
Like, and we're going to see it work in some cases, not in others.
341
00:28:46.074 --> 00:28:48.414
It kind of becomes the faceless account 2.0.
342
00:28:50.415 --> 00:28:53.157
Fuck Jerry, for example, like part of my language, right?
343
00:28:53.197 --> 00:28:54.778
Like faceless account.
344
00:28:55.079 --> 00:29:01.263
There's still a lot of trust and reputation to that because, you know, there's like a team behind it and he's behind it.
345
00:29:01.323 --> 00:29:11.151
And even though he's not featuring his own likeness and you have a ton of accounts like that, you have that, you know, the traded accounts that are reposting deals that are happening.
346
00:29:11.231 --> 00:29:17.096
You just have so, you know, you have the overheard accounts that were, you know, you know, super probably overheard LA overheard.
347
00:29:17.116 --> 00:29:19.317
You're like, none of these really had like a
348
00:29:20.538 --> 00:29:22.579
a key person at the helm of it.
349
00:29:23.339 --> 00:29:26.400
You did, but they were not the front person of it.
350
00:29:26.440 --> 00:29:29.620
So I think that we're going to see a lot of that for sure.
351
00:29:30.081 --> 00:29:43.244
But just kind of looking at that, if that's really the parallel or the analogy for it, it was always the case and those accounts were very good in the exceptions, but most faceless accounts did not have a lot of value.
352
00:29:43.984 --> 00:29:51.111
from a brand standpoint, from an audience standpoint, relative on average to their accounts where there was a real person behind it.
353
00:29:51.271 --> 00:29:52.633
And so I think it's going to be that.
354
00:29:52.893 --> 00:29:56.476
Some are going to do really well, like the fuck Jerry's and the trade ads and whatnot.
355
00:29:57.057 --> 00:30:01.842
I think most are not going to do as well as one with a real human behind it.
356
00:30:02.611 --> 00:30:03.651
Let's say I'm real human.
357
00:30:04.231 --> 00:30:07.372
Ultimately, am I just going to have a model write my brief?
358
00:30:07.632 --> 00:30:11.474
It's going to go to a system like Mighty Joy or Super Deal.
359
00:30:12.134 --> 00:30:13.174
That's going to go to people.
360
00:30:13.214 --> 00:30:14.755
They're going to have their systems look at it.
361
00:30:14.995 --> 00:30:26.278
And then they might even just say, great, I'll do this, and then have their Higgs field MCP or Creatify MCP, whatever it is on their side, just make the video exactly like that.
362
00:30:26.658 --> 00:30:29.359
Why in five years is that just not the case?
363
00:30:29.719 --> 00:30:30.779
That will be the case.
364
00:30:31.700 --> 00:30:32.600
That will be the case.
365
00:30:32.840 --> 00:30:35.101
A lot of creators are using Higgs field now.
366
00:30:35.401 --> 00:30:37.942
Token costs are high for video generation like that.
367
00:30:37.962 --> 00:30:41.223
Like it requires a lot of editing or a lot of work.
368
00:30:41.303 --> 00:30:52.327
So, but you know, what they're using it for is like, instead of going and spending five days, like we just did this yesterday or two days ago with the creator where like, instead of spending five days on location for a shoot,
369
00:30:53.327 --> 00:31:08.958
We essentially had a creator come by our office and just wear a hazmat suit and wear different gear and created the whole environment around them just to create some interesting content as if they were on site for different places.
370
00:31:08.998 --> 00:31:15.343
So I think that evolves to just the creator being able to just stand back for the most part.
371
00:31:16.043 --> 00:31:18.366
and having their normal content be created through this.
372
00:31:18.466 --> 00:31:28.657
I think in our world, the business shifts from influencer marketing companies being the ones to do the work to where the value actually just becomes the network effect, right?
373
00:31:28.697 --> 00:31:31.320
Like, how are we bringing these two parts together?
374
00:31:31.801 --> 00:31:35.505
The value moves up to, like, high-level strategy, but, like,
375
00:31:36.535 --> 00:31:40.980
You know, you can kind of keep playing each iteration forward and you can automate a lot of this.
376
00:31:41.020 --> 00:31:49.369
But I think the ultimate value becomes the network effect within our business of like, here's a large network of creators.
377
00:31:49.429 --> 00:31:51.030
We have a large network of brands.
378
00:31:52.071 --> 00:31:55.874
Brands come in because we have the creators, creators stay because we get brand access.
379
00:31:56.234 --> 00:31:58.116
That effectively becomes the glue.
380
00:31:58.616 --> 00:32:04.780
Influencer marketing companies are not valued based on the ability to do the work because the friction gets removed with AI.
381
00:32:05.361 --> 00:32:15.268
On a lot of that, the influencer becomes a kind of go, no-go decision maker the same way a manager will call a celebrity and say, hey, do you want to do this deal or not?
382
00:32:15.368 --> 00:32:20.131
And they'll kind of back and forth on high-level terms and likewise on a brand side, they'll
383
00:32:20.431 --> 00:32:22.312
Does this meet the criteria or not?
384
00:32:22.412 --> 00:32:24.254
Check or no, it doesn't.
385
00:32:24.314 --> 00:32:28.396
So yeah, I think a lot of the friction gets removed in that process.
386
00:32:28.416 --> 00:32:41.185
If you're an e-commerce brand, you're working on TikTok, you're on Tribe, and you're trying to make your influencer dollars go as far as possible, what do you recommend they do in this AI world?
387
00:32:41.525 --> 00:32:48.090
One of the most cost-effective strategies right now is a tactic called whitelisting where you take...
388
00:32:49.103 --> 00:32:54.765
a creator's content and you run the ad through their account on your behalf.
389
00:32:54.865 --> 00:33:04.128
So why like their account has social proof, their likeness has social proof, it's not a brand talking about how great it is, it's a creator or someone else talking about it.
390
00:33:04.908 --> 00:33:17.053
I think with AI, what that looks like is being able to take that creator, instead of one piece of content from one creator, you take it, you run all the different variants, you run it in different languages too, right?
391
00:33:17.093 --> 00:33:17.653
So you're like,
392
00:33:18.313 --> 00:33:29.660
hey, if we're a global company, you know, or maybe if it's a small brand, like, you know, maybe it's let's go after these different communities where now the creator speaking in French or Spanish or whatever, right?
393
00:33:29.700 --> 00:33:36.204
Like whatever that might look like, the creator has a different value proposition for each of those, right?
394
00:33:36.304 --> 00:33:38.406
And creators are cool with us for the most part nowadays.
395
00:33:38.838 --> 00:33:39.899
They'll charge for it, yeah.
396
00:33:40.059 --> 00:33:43.041
But it's actually like a very new frontier.
397
00:33:43.181 --> 00:33:49.745
So one of the things we're seeing a lot more of in the industry are modification rights.
398
00:33:50.085 --> 00:33:52.806
Pre-AI, it's like, okay, like, yeah, you can cut this up.
399
00:33:53.067 --> 00:33:54.888
Like, really, what are you gonna do?
400
00:33:54.908 --> 00:34:03.153
But like now it's like, what if now you're using that creator's likeness for a message completely different that they did not approve or they don't endorse?
401
00:34:03.353 --> 00:34:06.535
Or it's you sign a deal with an ad agency
402
00:34:07.395 --> 00:34:21.667
and now they have the right to modify that content, technically they can go and say, we own your likeness now because we can modify that same piece of content and number of times, an infinite number of times for all our different advertisers, right?
403
00:34:21.707 --> 00:34:26.691
So being able to get really specific in the language, I think we're going to see a lot of cases where
404
00:34:27.612 --> 00:34:32.873
creators and or brands are not, you know, they're kind of vague with those terms or they're silent on them.
405
00:34:32.893 --> 00:34:35.694
And we're going to see some big lawsuits coming out.
406
00:34:36.114 --> 00:34:42.196
Something happened where a brand used the rights they weren't supposed to and an agency used rights they weren't supposed to.
407
00:34:42.316 --> 00:34:44.116
And the creator didn't agree to that.
408
00:34:44.156 --> 00:34:53.359
And so it's going to be interesting, but we're definitely like in the Wild West sort of sort of era on how AI and influencer marketing is happening on that front.
409
00:34:53.924 --> 00:35:09.522
I was talking to somebody else who basically said they have all these shadow projects building their own influencers so they can get those, the terms they want, which is very murky in a lot of ways, legally, I think, um,
410
00:35:11.308 --> 00:35:18.512
But you can see why they would want that if it's so hard to, if everything is a negotiation until the contracts are standardized.
411
00:35:19.573 --> 00:35:24.715
It's like each one of those things is, you know, almost cost more than it's worth in terms of attention.
412
00:35:24.916 --> 00:35:29.118
Real quickly, you want to give the quick pitch for Mighty Joy and Super Deal?
413
00:35:29.418 --> 00:35:39.842
We're a creator agency focused on people with real expertise and domain knowledge, representing some of the world's largest brands and most trustworthy creators.
414
00:35:40.082 --> 00:35:46.325
Superdeal, that's been our tip of the spear into this AI first world, where we're working to automate
415
00:35:47.305 --> 00:36:07.797
a lot of the processes and workflows to create AI native marketplace for creators and brands, allowing small creators to have the same firepower the large creators have, and allowing brands and influencer marketing teams at brands to scale up with substantially less friction.
416
00:36:07.837 --> 00:36:12.180
Really, the goal there is how do we give a small startup
417
00:36:13.341 --> 00:36:18.741
the same firepower on the influencer marketing side as some of the biggest Fortune 1000 brands out there.
1
00:00:00.271 --> 00:00:02.293
So you've mentioned the human connection.
2
00:00:02.673 --> 00:00:04.515
It for sure matters, I guess.
3
00:00:04.655 --> 00:00:17.645
But then when I look at a lot of the data for what drives sales across TikTok and all these other platforms, it's clearly AI-generated speakers and the podcast ones where they're sitting talking.
4
00:00:17.785 --> 00:00:20.127
Where does that all sit in the future?
5
00:00:20.187 --> 00:00:22.008
Is it just accepted by people?
6
00:00:22.048 --> 00:00:25.931
Do creators themselves become scaled because they're using it?
7
00:00:26.092 --> 00:00:28.894
What's your thought on AI-generated influencers?
8
00:00:29.530 --> 00:00:33.743
the amount of content being generated has been increasing year over year.
9
00:00:33.803 --> 00:00:34.666
And like AI has,
10
00:00:35.335 --> 00:00:37.616
sort of blown that up substantially, right?
11
00:00:37.636 --> 00:00:40.778
And so just way more content than can ever be consumed, right?
12
00:00:41.298 --> 00:00:47.882
And so if you look at that, as more of that has happened, you've seen a shift towards like social signals becoming increasingly important.
13
00:00:48.042 --> 00:00:50.723
I think it's gonna be the same thing with AI influencers.
14
00:00:50.763 --> 00:01:00.068
Like there's gonna be a flood of content, but I think people are still gonna value the creators they trust, and it's gonna make like the real humans more scarce.
15
00:01:09.059 --> 00:01:16.303
Two weekends ago, I bought a new car and I took every email communication from the dealer and I just put it in the chat GPT.
16
00:01:16.823 --> 00:01:19.424
And I asked it, like, am I getting a good deal?
17
00:01:19.544 --> 00:01:20.344
What's going on?
18
00:01:20.845 --> 00:01:22.025
Like, what should I respond?
19
00:01:22.105 --> 00:01:24.647
Then I would just copy and paste whatever chat GPT said.
20
00:01:27.367 --> 00:01:28.908
I literally added no value.
21
00:01:28.928 --> 00:01:33.289
The only thing I did was actually paste the link of the car in, which I then did wrong.
22
00:01:33.789 --> 00:01:36.810
So I got a quote on the wrong car and I had to go back.
23
00:01:36.890 --> 00:01:41.271
But since I added no value, I don't know why I even had to review it.
24
00:01:41.291 --> 00:01:45.091
So at some point, these agents talking to each other is clearly going to be a thing.
25
00:01:45.191 --> 00:01:51.173
And then Anthropic announced today that they gave three agents the same project at work.
26
00:01:52.373 --> 00:01:54.455
And basically office politics ensued.
27
00:01:55.296 --> 00:02:02.301
So the agents were trying to sabotage each other's projects and were talking shit about each other's projects.
28
00:02:03.182 --> 00:02:07.306
And basically, you know, they were trying to get ahead in the world.
29
00:02:07.366 --> 00:02:14.451
And so this whole idea of what is agent to agent stuff going to be, I know is something you've been thinking a lot about.
30
00:02:14.471 --> 00:02:20.617
So I just in the back of my head is I want to start with you ran Open Influence for ten years.
31
00:02:21.415 --> 00:02:28.680
After suffering through the pain of running a company, you decided to start another influencer-based company.
32
00:02:29.581 --> 00:02:34.945
But this time you started it once the world of AI had really come into focus.
33
00:02:35.465 --> 00:02:39.829
So what drove the desire to jump back in?
34
00:02:40.169 --> 00:02:42.591
And where did AI come into that?
35
00:02:43.111 --> 00:02:47.294
In the creator world, we're seeing the shift from the value being about reach.
36
00:02:47.314 --> 00:02:48.115
So how many people...
37
00:02:48.928 --> 00:02:53.031
you know, see your content to really being about trust and focus.
38
00:02:53.231 --> 00:02:56.513
And so how do you get people to actually take action?
39
00:02:56.573 --> 00:03:00.576
So that was that was one of the big insights to start Mighty Joy.
40
00:03:01.417 --> 00:03:08.222
And one of the keys to unlock that was tapping into micro influencers, which previously the influencer marketing industry
41
00:03:08.950 --> 00:03:11.051
didn't really do or didn't really do a good job of.
42
00:03:11.552 --> 00:03:22.379
You know, kind of where AI plugs into this is just from a timing perspective, you know, working with influencers is super, you know, labor intensive, time intensive.
43
00:03:22.439 --> 00:03:28.282
There's a lot of fuzzy logic and because you're dealing with lots of people and moving parts and so,
44
00:03:29.303 --> 00:03:34.666
It was kind of felt like right time to be able to apply this technology as it was coming about.
45
00:03:34.706 --> 00:03:41.930
The technology is perfectly suited, I should say, to sort of solve engaging with micro influencers at scale.
46
00:03:42.210 --> 00:03:43.931
And so can you explain that?
47
00:03:43.971 --> 00:03:45.192
What do you mean by that?
48
00:03:45.632 --> 00:03:48.174
You have really kind of like a few tiers of influencers.
49
00:03:48.194 --> 00:03:52.077
So we all know, you know, the big celebrities, we'll call those like macro creators.
50
00:03:52.757 --> 00:03:58.281
Then you have your mid-tail and 50,000 followers to, you know, 500,000 followers.
51
00:03:58.301 --> 00:04:01.303
And then you have your micros, you know, below 50,000, right?
52
00:04:01.403 --> 00:04:06.567
And then you have your nanos below 10,000 typically is how many people describe it.
53
00:04:06.927 --> 00:04:12.712
And so as the information landscape was becoming more fragmented, as the,
54
00:04:13.412 --> 00:04:18.835
you know, the industry was realizing it's more about trust in a, you know, and getting people to take action.
55
00:04:18.855 --> 00:04:27.839
You know, there was just this untapped sort of set of niche creators on the long tail that no one was tapping into.
56
00:04:28.059 --> 00:04:32.882
And so before the amount of labor to take to engage a creator is quite high.
57
00:04:33.542 --> 00:04:39.505
And so it really was, you know, the economics didn't really make sense for a lot of influencer marketing companies to go
58
00:04:40.331 --> 00:04:42.984
sort of down market to more of the micro influencers.
59
00:04:43.165 --> 00:04:44.712
And so the real sort of
60
00:04:45.725 --> 00:05:00.954
thought and insight and bet we were making was, can we leverage technology and AI to be able to tap into micro-influencers in a way that makes economic sense, which previously wasn't really possible.
61
00:05:01.434 --> 00:05:02.435
And how are you doing that?
62
00:05:02.815 --> 00:05:10.440
We've went from, call it like a sort of leveraging out-of-the-box tools and building some
63
00:05:11.740 --> 00:05:17.001
you know, some of our own, you know, you know, kind of more traditional software like workflow tools.
64
00:05:17.381 --> 00:05:23.103
First insight was how do we rely on SMS instead of email to communicate with micro influencers?
65
00:05:23.743 --> 00:05:30.745
But then where it sort of evolved to is us building out our own agentic harness for influencer marketing.
66
00:05:30.765 --> 00:05:40.607
And so like a lot of the questions we're asking ourselves is what the software even look like in an agentic first world, because your first intuition is to
67
00:05:41.387 --> 00:05:42.630
build what you know, right?
68
00:05:42.650 --> 00:05:44.836
And so it's like, how do we build software faster?
69
00:05:45.137 --> 00:05:46.259
But it looks very traditional.
70
00:05:46.279 --> 00:05:48.505
And so a lot of it was like, how do we actually build
71
00:05:49.630 --> 00:05:50.931
for kind of the future of work.
72
00:05:50.951 --> 00:06:01.636
And so that's where we leaned on, well, I talked to ChatGPT, I talked to Claude, what would that look like if I was doing that for influencer marketing?
73
00:06:01.656 --> 00:06:11.782
And so we built our own, essentially chat interface with our own set of custom tools in the background to manage the influencer workflows and doing everything from
74
00:06:12.620 --> 00:06:19.896
contract creation, some negotiation, but also like the whole project of like managing the workflow and the follow-up
75
00:06:20.163 --> 00:06:23.524
I would call it the execution, I guess, of the relationship.
76
00:06:24.364 --> 00:06:38.949
There's I'm a brand and I'm trying to find somebody and there's I'm a creator and in, in theory, you're in, you know, trying to maximize the value you can provide to that brand for a long-term relationship rather than a one-off.
77
00:06:39.069 --> 00:06:44.311
Is that, I guess, is that a safe assumption that that's true or are they mostly just trying to go in and out with one deal?
78
00:06:44.591 --> 00:06:45.752
Both happens, right?
79
00:06:45.772 --> 00:06:48.592
I think the ideal is you want a long-term, but, um,
80
00:06:50.416 --> 00:06:52.977
Initially, all that starts with like a test.
81
00:06:52.997 --> 00:07:00.180
So the brand comes in and they're very performance focused often or ROI focused in general, right?
82
00:07:00.220 --> 00:07:05.442
And so they'll test working with the creator and they'll continue working with them.
83
00:07:05.482 --> 00:07:07.823
And for some creators, they'll hit that sort of
84
00:07:08.243 --> 00:07:11.685
you know, diminishing return, the point of diminishing return.
85
00:07:11.705 --> 00:07:13.766
And so they'll stop and move to another creator.
86
00:07:14.326 --> 00:07:17.528
Others, they're growing with that creator sort of indefinitely.
87
00:07:17.548 --> 00:07:20.310
And so I think the ideal is long-term.
88
00:07:20.530 --> 00:07:21.410
How does AI help?
89
00:07:21.610 --> 00:07:23.251
We call that sourcing in the industry.
90
00:07:23.271 --> 00:07:25.713
So you're sourcing talent for a project.
91
00:07:26.733 --> 00:07:32.937
And sourcing, even though it sounds like it'd be incredibly straightforward and metrics driven,
92
00:07:33.995 --> 00:07:35.536
There are so many variables.
93
00:07:35.817 --> 00:07:41.462
At Open Influence, we developed technology just around identifying influencers using AI.
94
00:07:41.562 --> 00:07:48.528
And a lot of that AI, this was like back in 2014, mind you, but it was all computer vision based.
95
00:07:49.068 --> 00:07:52.651
Because the challenge then that we're trying to solve, and this is key for sourcing,
96
00:07:53.672 --> 00:07:55.993
is how do we just categorize an influencer?
97
00:07:56.193 --> 00:08:00.235
Because you might talk about fashion, like what does that mean?
98
00:08:00.875 --> 00:08:07.998
And if you have a account manager, let's say on our team doing that, like that's subjective, right?
99
00:08:08.018 --> 00:08:11.800
Because like the account manager will decide like what is fashion, what is not?
100
00:08:12.640 --> 00:08:17.242
If you have the influencer doing it themselves, that's also subjective, right?
101
00:08:17.282 --> 00:08:27.007
Because like they're starting to say like, well, I'm a fashion influencer because I wear clothes and I'm a pet influencer because I have a pet and I'm a fitness influencer because I go to the gym three times a week, right?
102
00:08:27.507 --> 00:08:29.948
Categorizing within fashion, fashion for...
103
00:08:31.328 --> 00:08:36.690
Forever 21 and for Hermes are very different.
104
00:08:37.470 --> 00:08:40.131
It's the same term, completely different worlds.
105
00:08:40.451 --> 00:08:42.012
And so it was a really good example.
106
00:08:42.052 --> 00:08:51.415
So early on, and actually we started with open influence in the fashion world, but early on, AI was extremely helpful in actually looking at the content itself.
107
00:08:52.155 --> 00:08:54.876
and then categorizing the creator based on that.
108
00:08:54.916 --> 00:08:58.416
And then what we did is we mapped the content to engagement.
109
00:08:58.516 --> 00:09:09.638
So it was not just, you know, here's Sarah and she posts about dogs and going boxing and, you know, surfing at the beach or whatever.
110
00:09:10.239 --> 00:09:16.700
But we could tell you what breed of dog she has and how much her audience engages when she posts about her dog.
111
00:09:17.965 --> 00:09:22.026
relative to when she posts about going to the gym or swimming in the beach.
112
00:09:22.086 --> 00:09:30.928
And so you might have thought that she was a fitness influencer because she's posting eight times out of ten herself at the gym.
113
00:09:31.548 --> 00:09:37.290
But really, she's a pet influencer because when she talks about advice for her dog, her audience really engages.
114
00:09:38.170 --> 00:09:46.592
And so that was a really early application of how we're using AI to just identify not just who these creators are,
115
00:09:48.412 --> 00:09:50.234
where their influence actually lies.
116
00:09:50.274 --> 00:09:52.715
Like what do people actually trust them in?
117
00:09:53.115 --> 00:09:57.618
And so that's something we carried forward to Mighty Joy, but that was like a really early lesson.
118
00:09:57.638 --> 00:10:05.843
And honestly, like we were way too early at the time with Open Influence with that technology, cause like we present that to big brands and Fortune 1000 companies.
119
00:10:06.687 --> 00:10:07.988
They're like, look how cool this is.
120
00:10:08.028 --> 00:10:11.111
And like, okay, how much am I paying per influencer?
121
00:10:11.272 --> 00:10:16.016
Like they were just so simplistic in how they were looking at it at the time.
122
00:10:16.397 --> 00:10:21.602
And so we were probably 10 years too early in the application of that technology.
123
00:10:21.722 --> 00:10:25.226
But really now that's been a really key driver of looking at
124
00:10:26.184 --> 00:10:30.045
the engagement rate at a per content level and mapping that.
125
00:10:30.665 --> 00:10:38.087
And then that's key in sourcing because when we're able to go type in a keyword, we could bring in the right creators for it.
126
00:10:38.427 --> 00:10:41.948
And I could keep talking about how else AI is helpful in the sourcing side.
127
00:10:41.988 --> 00:10:46.329
But the other thing too is there's a lot of lookalike modeling that needs to happen.
128
00:10:46.409 --> 00:10:50.830
So oftentimes a brand will find like a set of creators that they like.
129
00:10:51.030 --> 00:10:52.970
And it's really this idea of like taste.
130
00:10:56.311 --> 00:10:58.314
in the age of AI, it's all about taste.
131
00:10:58.635 --> 00:11:04.603
It's really hard to quantify where it's just like, you know, you'll look and say like this person's on brand, they're not on brand.
132
00:11:04.703 --> 00:11:05.004
Right.
133
00:11:05.765 --> 00:11:10.712
And that's where AI is actually really helpful from a lookalike modeling standpoint, because then we can just plug in
134
00:11:11.753 --> 00:11:16.737
creators and they'll find similar creators based on sort of different parameters, right?
135
00:11:16.757 --> 00:11:22.002
The content we post about overlap in, you know, influential followers.
136
00:11:22.082 --> 00:11:25.725
So we can start building that social graph with it as well.
137
00:11:26.086 --> 00:11:30.109
And then the other thing AI is really helpful for is on the sourcing side is you could run loops.
138
00:11:30.409 --> 00:11:33.772
So it can go run a query, pull a bunch of creators,
139
00:11:34.853 --> 00:11:37.675
analyze them, run it again, run it again.
140
00:11:37.755 --> 00:11:44.079
So we did a case study for a beauty brand where I sat down a very large beauty brand CEO.
141
00:11:45.160 --> 00:11:57.448
And he said, you know, we're really trying to get, you know, executives within this niche, women executives that don't have a big following, but are, you know, really just leaders in their industry and field.
142
00:11:57.468 --> 00:11:59.450
Like, you know, a lot of them would have like maybe
143
00:12:00.606 --> 00:12:04.668
few thousand followers, maybe 10,000, 15,000.
144
00:12:04.728 --> 00:12:06.950
So again, extremely niche audience.
145
00:12:07.390 --> 00:12:10.171
Live demo was like, okay, let's try this out with our AI tool.
146
00:12:10.332 --> 00:12:10.892
We ran it.
147
00:12:11.832 --> 00:12:17.215
10 minutes later, we had a list of 150 creators that matched all the criteria that they wanted.
148
00:12:17.376 --> 00:12:21.058
And so it was able to kind of go through those loops of
149
00:12:21.858 --> 00:12:24.861
running the query, kind of doing the taste check, right?
150
00:12:24.941 --> 00:12:32.969
And then running and iterating and removing creators and adding creators to that list until we got a final list that we could then review.
151
00:12:33.509 --> 00:12:34.931
So I get that on the brand side.
152
00:12:35.091 --> 00:12:36.952
I guess on the creator side, you've got...
153
00:12:37.673 --> 00:12:48.580
on one extreme TikTok shop where there's just, there's a zillion invites going out, but then you've got like Tribe, which limits, you know, okay, reach out to these 10 people.
154
00:12:48.820 --> 00:12:56.725
You know, you can have the best list possible, but how do you get the creators to even look and pay attention and source more effectively on their side?
155
00:12:56.965 --> 00:13:04.411
So the system we built for that is, we said, how do we teach the AI based on people, like based on our team and our knowledge?
156
00:13:05.272 --> 00:13:14.059
And so we built this, like, it's gonna sound really simple, but like a great learning algorithm where essentially it drafts the message or the email.
157
00:13:14.139 --> 00:13:17.082
This is for the brand reaching out to the creator to get the creator to respond.
158
00:13:17.482 --> 00:13:24.287
One, we measure the responses, but even more so, like it's just that human taste of knowing, okay, here's,
159
00:13:25.140 --> 00:13:26.201
How I'm going to frame it.
160
00:13:26.281 --> 00:13:35.211
Here's the ordering of like where I put the budget, where I put the – or if it's an affiliate deal, how I position the affiliate deal in those terms.
161
00:13:35.991 --> 00:13:39.775
And so what happens is the AI will draft the email based on instructions.
162
00:13:40.276 --> 00:13:43.159
A human will go through and edit it and then the AI will learn.
163
00:13:43.760 --> 00:13:46.362
And so it will learn specific to like just –
164
00:13:47.283 --> 00:13:52.444
generally, but also will learn to that specific campaign or that specific account or brand.
165
00:13:53.105 --> 00:13:56.266
And so like we do that a few times and it gets much smarter, right?
166
00:13:56.326 --> 00:13:57.526
Each time.
167
00:13:57.906 --> 00:14:06.788
For us, that's kind of the next big challenge with our software to solve is like this learning algorithm and like memory, which is just a bigger kind of challenge industry-wide.
168
00:14:06.848 --> 00:14:12.450
But we've kind of come up with like a kind of a, you know, a hack to, you know,
169
00:14:13.050 --> 00:14:14.191
How do we take the feedback?
170
00:14:14.251 --> 00:14:18.594
How do we measure the amount of edits and where the edits are?
171
00:14:19.194 --> 00:14:23.598
And then let's use that to update the main instruction set that feeds in.
172
00:14:24.098 --> 00:14:28.021
And then at what point do we trigger a new update to run, right?
173
00:14:28.041 --> 00:14:32.224
So that we're making sure that we're weighing newer information more heavily.
174
00:14:32.284 --> 00:14:37.748
But also, if we have a lot of history with certain information, we're not forgetting that.
175
00:14:37.908 --> 00:14:42.111
So that's kind of what we're working through actively now, is solving that problem.
176
00:14:42.531 --> 00:14:46.673
How does the closed loop piece then work to reinforce?
177
00:14:46.853 --> 00:14:48.254
Is it somewhere along the line?
178
00:14:48.534 --> 00:14:53.256
You know what the creator threshold for earnings are or the brands?
179
00:14:53.996 --> 00:14:55.937
How does it evolve over time?
180
00:14:56.317 --> 00:15:01.019
Right now, the main focus has just been on outreach and email comms.
181
00:15:01.199 --> 00:15:08.742
And so when more than 10% of the correspondence are new, it'll trigger sort of a refresh.
182
00:15:08.762 --> 00:15:09.883
So you can kind of see...
183
00:15:10.928 --> 00:15:16.071
over time, the threshold to hit that 10% becomes higher in terms of volume.
184
00:15:16.791 --> 00:15:30.159
And so that's just one way of like, kind of the system learning, but yeah, like we're actively in the middle of, you know, we have our pricing calculators that we've built out and whatnot, but part of the bigger challenge with a lot of that is also,
185
00:15:30.899 --> 00:15:37.722
where do we rely on letting the agents kind of run versus like, where do we build the deterministic systems in?
186
00:15:38.382 --> 00:15:42.244
And then we right now have checks at a lot of major steps.
187
00:15:42.264 --> 00:15:46.986
So like, we don't let the AI just like run, we let it draft and then a human approves.
188
00:15:47.147 --> 00:15:47.547
And so,
189
00:15:48.267 --> 00:15:51.109
We're probably a little bit overly cautious in how we're doing it.
190
00:15:51.129 --> 00:15:56.874
But like the last thing we want to do is have the AI run, send a million dollar deal to an influencer.
191
00:15:56.894 --> 00:16:06.522
And then we follow up saying, just kidding, you know, because there's just a lot of reputation risk with with pissing off people with large reach.
192
00:16:06.742 --> 00:16:08.463
You can always just set that threshold, right?
193
00:16:08.523 --> 00:16:12.925
To say, don't give an offer like that, but it's clearly harder than that.
194
00:16:13.085 --> 00:16:22.028
So what signal do you have to see before you'll just say, okay, for this type of brand and this type of creator, we're gonna let the system send it.
195
00:16:22.368 --> 00:16:30.972
And then the similarly, what do you hear from your creators who care about their brands, even if they're 1000 followers, I assume?
196
00:16:31.352 --> 00:16:37.937
Like when they would just be willing to say yes or no in an automated fashion and not the agent handle it.
197
00:16:38.398 --> 00:16:47.365
Yeah, we haven't gone to fully automated, but like for things like price, for example, like we rely on like deterministic systems a little bit more so.
198
00:16:47.425 --> 00:17:00.836
So it's like, I mean, we have the checks in the email, but like even for like our contract module, like if you're going to send out like a contract that like it won't let you like it's a formula calculating that or you manually entering it.
199
00:17:01.337 --> 00:17:08.249
as opposed to the agent just kind of like making up a number and plugging it in, even though in the email, like they could kind of say whatever still.
200
00:17:08.290 --> 00:17:08.991
So we need to
201
00:17:10.124 --> 00:17:11.345
That's where we have the human control.
202
00:17:11.505 --> 00:17:14.087
On the creator side, much more of a scoring system.
203
00:17:14.167 --> 00:17:18.330
So like we've automated on the creator side, like an inbox tool.
204
00:17:18.390 --> 00:17:21.893
So like it'll read their inbox, which is the main way creators get their deals.
205
00:17:22.193 --> 00:17:27.177
It'll generate contract once they're ready to go and send it out and review the red lines.
206
00:17:27.637 --> 00:17:32.380
It can even take in deals previously and put it in the format to help manage the workflow.
207
00:17:32.420 --> 00:17:35.283
But what we have is we have like a scoring system
208
00:17:35.943 --> 00:17:38.745
So it'll come in, it'll look like, okay, this is a real brand.
209
00:17:38.785 --> 00:17:41.627
This didn't come in from like a dot X, Y, Z domain.
210
00:17:41.647 --> 00:17:44.488
You know, this is the real domain for a real brand.
211
00:17:44.548 --> 00:17:50.292
And so like we're doing that matching and that scoring just so the creator can see like, okay, there are real deals coming in.
212
00:17:50.752 --> 00:17:55.255
And then the proactive outreach, like we definitely allow for that in the system.
213
00:17:55.315 --> 00:18:03.800
And we do that often, but, you know, proactive outreach when you're a smaller creator is a little bit harder where we find a lot of success.
214
00:18:04.781 --> 00:18:06.242
is when you just bundle the outreach.
215
00:18:06.262 --> 00:18:25.510
So what we do is like, we'll send out a brand, like here are a list of 20 recommended creators and the creators on that list, as opposed to like, here's a one-off because the brands tend to think more from a metric standpoint, unless they're working with like a larger creator that where they're looking more for like name image likeness
216
00:18:26.210 --> 00:18:28.572
They're mainly looking like check off several boxes.
217
00:18:28.812 --> 00:18:35.576
And so what we'll send is like, hey, based on what we're seeing, it seems like you're working with these creators in the space right now.
218
00:18:35.677 --> 00:18:43.542
These are the criteria we see you, you know, we believe you're looking for your list of 20 creators we think could be a good fit for you.
219
00:18:43.763 --> 00:18:49.345
I think about the HR industry a lot, which is well talked about because you've got candidates using AI for their resumes.
220
00:18:49.365 --> 00:18:51.966
You've got the AI screeners.
221
00:18:53.247 --> 00:18:56.048
And so it really is sort of agent to agent.
222
00:18:56.648 --> 00:18:59.209
And it seems like nobody's that happy about it.
223
00:18:59.529 --> 00:19:00.830
The Google DeepMind division.
224
00:19:00.870 --> 00:19:11.114
So it's like the guys with AI said they lost out, like the AI basically screened out 30 to plus percent, like a big number of people that should have been highly qualified candidates.
225
00:19:11.414 --> 00:19:18.197
Is that just going to be the case for a while in the inflature space that you're going to have to be manual about opening up the offers and opening up things?
226
00:19:18.297 --> 00:19:21.498
Or is the AI screening going to get you far enough along?
227
00:19:21.958 --> 00:19:25.240
You know, I think the difference with us and candidates like –
228
00:19:26.308 --> 00:19:31.049
In the HR space, you can put a prompt injection in your resume, for example.
229
00:19:31.229 --> 00:19:32.250
Right.
230
00:19:32.590 --> 00:19:34.490
And so there's sort of more there.
231
00:19:34.510 --> 00:19:38.591
There's no like even on LinkedIn, like people could make up that they work at places.
232
00:19:38.651 --> 00:19:38.831
Right.
233
00:19:38.851 --> 00:19:47.894
Like people do that all the time where they'll they'll say they work at TikTok or meta when realistically they're just, you know, they have a TikTok account.
234
00:19:48.134 --> 00:19:54.608
You know, and so for us, luckily in the influencer space, like the source of truth are the platforms themselves.
235
00:19:54.788 --> 00:19:57.053
And so like we can pull that data directly.
236
00:19:58.090 --> 00:20:00.151
and we can have the creators authenticate too.
237
00:20:00.331 --> 00:20:03.253
It's really hard to manipulate from that standpoint.
238
00:20:03.493 --> 00:20:09.177
I think the challenge is like, it's such a human driven endeavor, right?
239
00:20:09.197 --> 00:20:14.680
Cause it's like, you have a decision maker being a person, you have the creator as a person, and then they're talking to people.
240
00:20:14.720 --> 00:20:17.983
So it's like, part I like about the business is inherently
241
00:20:19.183 --> 00:20:25.728
It's all about human interaction, which I think is going to be which has friction and inherently is going to become more scarce.
242
00:20:25.828 --> 00:20:34.353
Like when we're living in a world where we have 100 agents running for every human, then it's like the human piece is just naturally going to be the more scarce piece.
243
00:20:34.533 --> 00:20:39.777
And like each human is going to be a decision maker with more impact because they have.
244
00:20:40.873 --> 00:20:42.353
an army of agents they're running.
245
00:20:42.433 --> 00:20:55.837
So I think for us, the way we're kind of looking at it, at least now is that we wanna eliminate as much of the friction and the busy work kind of in between those two parties, but still have the decision sit with them, right?
246
00:20:55.857 --> 00:21:00.838
So like we're using AI for like contract review, we're using AI for workflow.
247
00:21:01.138 --> 00:21:06.560
The contract review is typically higher value, but like you can audit, like a lot of it's like commercial terms.
248
00:21:06.880 --> 00:21:09.921
So it's more like the human, the creator decides like,
249
00:21:10.944 --> 00:21:12.565
What usage rights am I gonna give up?
250
00:21:12.725 --> 00:21:14.685
What are the payment terms?
251
00:21:14.765 --> 00:21:17.026
And so like a lot of that's commercial point.
252
00:21:17.427 --> 00:21:28.151
The part that's just like not really valuable but takes up time is every brand has a different contract and because of the brand and they're the ones paying, the creator's kind of stuck reviewing
253
00:21:29.091 --> 00:21:33.892
for like a $500 deal, they're having to review a bunch of different contracts.
254
00:21:33.932 --> 00:21:41.593
And that's where brands put in like, you know, sometimes they put in like very aggressive terms where they have full ownership rights over the content.
255
00:21:41.673 --> 00:21:48.815
Or, you know, they'll have a certain term supersede the SOW or they'll have, you know, certain indemnifications, right?
256
00:21:48.915 --> 00:21:54.056
Or not indemnifications where the creator is not indemnified if the brand gets sued for something, right?
257
00:21:54.076 --> 00:21:55.536
The creators see value in that?
258
00:21:56.136 --> 00:21:58.298
AI, are they willing to pay for it?
259
00:21:58.318 --> 00:21:59.338
Or just the brands?
260
00:21:59.899 --> 00:22:02.040
They're only willing to pay for it as part of the brand commission.
261
00:22:02.380 --> 00:22:06.303
The software we've built, we're not charging creators for it yet.
262
00:22:06.543 --> 00:22:12.727
And we're just rolling it out more so on the creator side where we're using it internally for the creators we work with and manage.
263
00:22:13.228 --> 00:22:14.769
Because we have two sides of our business.
264
00:22:14.809 --> 00:22:16.710
We rep some creators, we rep brands.
265
00:22:16.950 --> 00:22:24.836
We had to figure out how do we build an AI system where we're not going to be upside down on token costs on the creator side where we can still give them like a freemium model.
266
00:22:25.822 --> 00:22:27.603
So that was a whole challenge we had to solve.
267
00:22:27.723 --> 00:22:45.272
But just from talking to creators and just from our own team that's operating on behalf of creators, it's a huge headache having to go through contracts because one, if we're gonna go send these to our attorneys, it's not worth what they're gonna bill us to have them negotiate the deal.
268
00:22:46.152 --> 00:22:54.841
And then on the flip side, like we're having someone spending two hours reviewing a contract, not best use of their time.
269
00:22:55.421 --> 00:23:02.548
Speaking to a lot of small creators, they're kind of put in this awkward position where they just have to sign something or they're having like –
270
00:23:03.209 --> 00:23:06.350
a friend or family member review it, which obviously doesn't scale.
271
00:23:06.430 --> 00:23:13.232
So one of the things we want to solve for with our software is like, first, how do we solve for it with us as like on the talent management side?
272
00:23:13.732 --> 00:23:31.097
But then two, our real goal is like, how do we bring the value proposition that a talent manager brings to a creator and scale that down to the smaller creators that, that where, where the economics don't really make sense for a talent manager to get involved from a manual standpoint.
273
00:23:31.317 --> 00:23:34.299
To shift away from AI for a second, are the terms that different?
274
00:23:34.620 --> 00:23:37.462
I'm surprised to see that it's that very...
275
00:23:37.602 --> 00:23:38.763
They're not different.
276
00:23:38.803 --> 00:23:44.508
They're just the time you have to parse the different contracts because they'll be placed in different sections.
277
00:23:45.008 --> 00:23:46.469
They'll be worded differently.
278
00:23:46.789 --> 00:23:51.593
It's just annoying to have to parse through more so than per se really difficult.
279
00:23:51.753 --> 00:23:56.998
And so that's where AI is super helpful to just be able to kind of spot it and say like...
280
00:23:57.758 --> 00:23:59.899
Yeah, here are the main commercial points.
281
00:24:00.339 --> 00:24:06.283
And like, even with our system, like one of the things we built out is how do we create just standard contract templates?
282
00:24:06.343 --> 00:24:11.766
Like we'd love for that to become more industry standard just so it's removing further friction.
283
00:24:11.786 --> 00:24:15.508
Like we talk a lot about like AI helping remove friction, but like also just like.
284
00:24:16.268 --> 00:24:19.010
Creating industry standards removes a ton of friction too.
285
00:24:19.070 --> 00:24:24.195
And so, and the industry has made a lot of progress towards that, but you still have like these little translation issues.
286
00:24:24.255 --> 00:24:32.682
So if you're buying a house, like every real estate agent has like the same, I forget what it's called, but like, you know, the same kind of standard, like.
287
00:24:32.702 --> 00:24:34.144
The MLS agreement, whatever they call it.
288
00:24:34.184 --> 00:24:34.644
Exactly.
289
00:24:34.684 --> 00:24:39.729
MLS agreement or like they have like their, so it just makes it easy to not have to kind of think through.
290
00:24:39.749 --> 00:24:39.809
So.
291
00:24:40.676 --> 00:24:42.697
So you've mentioned the human connection.
292
00:24:42.717 --> 00:24:44.918
It for sure matters, I guess.
293
00:24:45.058 --> 00:24:58.066
But then when I look at a lot of the data for what drives sales across TikTok and all these other platforms, it's clearly AI-generated speakers and the podcast ones where they're sitting talking.
294
00:24:58.186 --> 00:25:00.547
Where does that all sit in the future?
295
00:25:00.607 --> 00:25:02.428
Is it just accepted by people?
296
00:25:02.468 --> 00:25:06.470
Do creators themselves become scaled because they're using it?
297
00:25:06.510 --> 00:25:09.292
What's your thought on AI-generated influencers?
298
00:25:09.312 --> 00:25:09.412
Yeah.
299
00:25:09.782 --> 00:25:16.507
If you look at the story of the internet, it's essentially been this story of the more content that becomes available.
300
00:25:17.328 --> 00:25:19.529
Like human attention is kind of fixed.
301
00:25:19.809 --> 00:25:24.873
Like now, yes, like it's grown in terms of like the surfaces for us to pay attention to.
302
00:25:25.193 --> 00:25:31.018
But like we only have so many hours in a day, like there's a real cap to the amount of attention we can give to anything.
303
00:25:31.658 --> 00:25:35.881
The amount of content being generated has been increasing year over year.
304
00:25:35.921 --> 00:25:36.822
And like AI has...
305
00:25:37.463 --> 00:26:00.960
sort of blown that up substantially right and so just looking like pre-ai at the internet there's like some crazy stat where like there are hundreds or thousands of hours of youtube video like uploaded every hour like they just it's just way more content than can ever be consumed right and so if you look at that as more of that has happened you've seen a shift towards like social signals becoming increasingly important so like
306
00:26:01.660 --> 00:26:02.501
social proof.
307
00:26:02.521 --> 00:26:06.343
So whether that's like reviews and ratings, likes and comments, right?
308
00:26:06.363 --> 00:26:14.408
Like if you think about like just even on the paid ad side, you know, any paid agency will tell you, oh, you got to increase your social proof.
309
00:26:14.428 --> 00:26:15.789
You need more testimonials.
310
00:26:15.809 --> 00:26:17.670
You need more ratings and reviews, right?
311
00:26:17.690 --> 00:26:21.032
Like in the Amazon game, like that's the real estate, right?
312
00:26:21.052 --> 00:26:24.614
Your reviews and your ratings and the count of those and the quality of those, right?
313
00:26:24.674 --> 00:26:28.056
And so I think it's going to be the same thing with AI influencers.
314
00:26:28.096 --> 00:26:30.038
Like there's going to be a flood of content,
315
00:26:30.698 --> 00:26:37.383
But I think people are still going to value the creators they trust, and it's going to make the real humans more scarce.
316
00:26:37.663 --> 00:26:46.230
I do think what it ends up looking like is every creator kind of becomes a mini studio using AI to enable themselves to scale.
317
00:26:46.950 --> 00:26:54.436
So they're going to be able to look and say, and I think it's also going to transform advertising where you're going to get a lot more variants of the same thing.
318
00:26:54.476 --> 00:27:05.065
So like smart creators right now, what they're doing is they're creating, you know, for every video they post, they're creating a few different variants and they're testing that through trial reels.
319
00:27:05.525 --> 00:27:08.988
They see what works and then they post the best performing one.
320
00:27:09.368 --> 00:27:13.914
So they basically put up one, they just wait and see if it pops or whatever the signal they're looking for.
321
00:27:14.094 --> 00:27:16.256
And then they delete the other three.
322
00:27:16.396 --> 00:27:16.997
Is that the idea?
323
00:27:17.217 --> 00:27:18.619
They'll do it through like a trial reel.
324
00:27:18.639 --> 00:27:22.924
So it won't hit their main feed and it'll reach outside of their audience.
325
00:27:23.044 --> 00:27:28.831
So we'll just run that, get the data and then post to their primary audience properly.
326
00:27:29.853 --> 00:27:33.058
So I think we're going to see a lot more variance running.
327
00:27:33.218 --> 00:27:40.069
I think it's also going to allow, from an ad perspective, from like a paid social when you're like boosting an ad.
328
00:27:40.750 --> 00:27:43.815
I mean, imagine hyper-tailored content.
329
00:27:44.196 --> 00:27:51.944
where the creator can create a video with a message around like, let's say it's around, you're saying like how you use the AI to buy a car.
330
00:27:52.425 --> 00:28:02.896
Like imagine if like, you know, cars.com or Carvana or any of these sites that are trying to get you to buy a car from them or any of the car dealerships or, you know, car companies themselves,
331
00:28:03.116 --> 00:28:11.501
based on how you live, this is why the Rivian is the perfect car for you, or this is why the Tesla is great, or this is why whatever is the best car for you.
332
00:28:11.901 --> 00:28:16.905
And that's through a creator that you follow delivering that message, right?
333
00:28:16.925 --> 00:28:19.566
So, like, I think that's kind of where AI evolves into.
334
00:28:19.586 --> 00:28:20.607
I buy that.
335
00:28:20.967 --> 00:28:26.268
And I guess when I think about reality TV and how, you know, we're both in LA, it's not real.
336
00:28:26.288 --> 00:28:29.029
We see them reshoot on reality TV.
337
00:28:29.309 --> 00:28:38.332
Why can't AI influencers, like a studio, just create tons of AI influencers and build them up in a credible fashion?
338
00:28:38.432 --> 00:28:40.512
Is that a likely outcome over the next few years?
339
00:28:41.012 --> 00:28:42.453
It's definitely going to happen, right?
340
00:28:42.493 --> 00:28:45.894
Like, and we're going to see it work in some cases, not in others.
341
00:28:46.074 --> 00:28:48.414
It kind of becomes the faceless account 2.0.
342
00:28:50.415 --> 00:28:53.157
Fuck Jerry, for example, like part of my language, right?
343
00:28:53.197 --> 00:28:54.778
Like faceless account.
344
00:28:55.079 --> 00:29:01.263
There's still a lot of trust and reputation to that because, you know, there's like a team behind it and he's behind it.
345
00:29:01.323 --> 00:29:11.151
And even though he's not featuring his own likeness and you have a ton of accounts like that, you have that, you know, the traded accounts that are reposting deals that are happening.
346
00:29:11.231 --> 00:29:17.096
You just have so, you know, you have the overheard accounts that were, you know, you know, super probably overheard LA overheard.
347
00:29:17.116 --> 00:29:19.317
You're like, none of these really had like a
348
00:29:20.538 --> 00:29:22.579
a key person at the helm of it.
349
00:29:23.339 --> 00:29:26.400
You did, but they were not the front person of it.
350
00:29:26.440 --> 00:29:29.620
So I think that we're going to see a lot of that for sure.
351
00:29:30.081 --> 00:29:43.244
But just kind of looking at that, if that's really the parallel or the analogy for it, it was always the case and those accounts were very good in the exceptions, but most faceless accounts did not have a lot of value.
352
00:29:43.984 --> 00:29:51.111
from a brand standpoint, from an audience standpoint, relative on average to their accounts where there was a real person behind it.
353
00:29:51.271 --> 00:29:52.633
And so I think it's going to be that.
354
00:29:52.893 --> 00:29:56.476
Some are going to do really well, like the fuck Jerry's and the trade ads and whatnot.
355
00:29:57.057 --> 00:30:01.842
I think most are not going to do as well as one with a real human behind it.
356
00:30:02.611 --> 00:30:03.651
Let's say I'm real human.
357
00:30:04.231 --> 00:30:07.372
Ultimately, am I just going to have a model write my brief?
358
00:30:07.632 --> 00:30:11.474
It's going to go to a system like Mighty Joy or Super Deal.
359
00:30:12.134 --> 00:30:13.174
That's going to go to people.
360
00:30:13.214 --> 00:30:14.755
They're going to have their systems look at it.
361
00:30:14.995 --> 00:30:26.278
And then they might even just say, great, I'll do this, and then have their Higgs field MCP or Creatify MCP, whatever it is on their side, just make the video exactly like that.
362
00:30:26.658 --> 00:30:29.359
Why in five years is that just not the case?
363
00:30:29.719 --> 00:30:30.779
That will be the case.
364
00:30:31.700 --> 00:30:32.600
That will be the case.
365
00:30:32.840 --> 00:30:35.101
A lot of creators are using Higgs field now.
366
00:30:35.401 --> 00:30:37.942
Token costs are high for video generation like that.
367
00:30:37.962 --> 00:30:41.223
Like it requires a lot of editing or a lot of work.
368
00:30:41.303 --> 00:30:52.327
So, but you know, what they're using it for is like, instead of going and spending five days, like we just did this yesterday or two days ago with the creator where like, instead of spending five days on location for a shoot,
369
00:30:53.327 --> 00:31:08.958
We essentially had a creator come by our office and just wear a hazmat suit and wear different gear and created the whole environment around them just to create some interesting content as if they were on site for different places.
370
00:31:08.998 --> 00:31:15.343
So I think that evolves to just the creator being able to just stand back for the most part.
371
00:31:16.043 --> 00:31:18.366
and having their normal content be created through this.
372
00:31:18.466 --> 00:31:28.657
I think in our world, the business shifts from influencer marketing companies being the ones to do the work to where the value actually just becomes the network effect, right?
373
00:31:28.697 --> 00:31:31.320
Like, how are we bringing these two parts together?
374
00:31:31.801 --> 00:31:35.505
The value moves up to, like, high-level strategy, but, like,
375
00:31:36.535 --> 00:31:40.980
You know, you can kind of keep playing each iteration forward and you can automate a lot of this.
376
00:31:41.020 --> 00:31:49.369
But I think the ultimate value becomes the network effect within our business of like, here's a large network of creators.
377
00:31:49.429 --> 00:31:51.030
We have a large network of brands.
378
00:31:52.071 --> 00:31:55.874
Brands come in because we have the creators, creators stay because we get brand access.
379
00:31:56.234 --> 00:31:58.116
That effectively becomes the glue.
380
00:31:58.616 --> 00:32:04.780
Influencer marketing companies are not valued based on the ability to do the work because the friction gets removed with AI.
381
00:32:05.361 --> 00:32:15.268
On a lot of that, the influencer becomes a kind of go, no-go decision maker the same way a manager will call a celebrity and say, hey, do you want to do this deal or not?
382
00:32:15.368 --> 00:32:20.131
And they'll kind of back and forth on high-level terms and likewise on a brand side, they'll
383
00:32:20.431 --> 00:32:22.312
Does this meet the criteria or not?
384
00:32:22.412 --> 00:32:24.254
Check or no, it doesn't.
385
00:32:24.314 --> 00:32:28.396
So yeah, I think a lot of the friction gets removed in that process.
386
00:32:28.416 --> 00:32:41.185
If you're an e-commerce brand, you're working on TikTok, you're on Tribe, and you're trying to make your influencer dollars go as far as possible, what do you recommend they do in this AI world?
387
00:32:41.525 --> 00:32:48.090
One of the most cost-effective strategies right now is a tactic called whitelisting where you take...
388
00:32:49.103 --> 00:32:54.765
a creator's content and you run the ad through their account on your behalf.
389
00:32:54.865 --> 00:33:04.128
So why like their account has social proof, their likeness has social proof, it's not a brand talking about how great it is, it's a creator or someone else talking about it.
390
00:33:04.908 --> 00:33:17.053
I think with AI, what that looks like is being able to take that creator, instead of one piece of content from one creator, you take it, you run all the different variants, you run it in different languages too, right?
391
00:33:17.093 --> 00:33:17.653
So you're like,
392
00:33:18.313 --> 00:33:29.660
hey, if we're a global company, you know, or maybe if it's a small brand, like, you know, maybe it's let's go after these different communities where now the creator speaking in French or Spanish or whatever, right?
393
00:33:29.700 --> 00:33:36.204
Like whatever that might look like, the creator has a different value proposition for each of those, right?
394
00:33:36.304 --> 00:33:38.406
And creators are cool with us for the most part nowadays.
395
00:33:38.838 --> 00:33:39.899
They'll charge for it, yeah.
396
00:33:40.059 --> 00:33:43.041
But it's actually like a very new frontier.
397
00:33:43.181 --> 00:33:49.745
So one of the things we're seeing a lot more of in the industry are modification rights.
398
00:33:50.085 --> 00:33:52.806
Pre-AI, it's like, okay, like, yeah, you can cut this up.
399
00:33:53.067 --> 00:33:54.888
Like, really, what are you gonna do?
400
00:33:54.908 --> 00:34:03.153
But like now it's like, what if now you're using that creator's likeness for a message completely different that they did not approve or they don't endorse?
401
00:34:03.353 --> 00:34:06.535
Or it's you sign a deal with an ad agency
402
00:34:07.395 --> 00:34:21.667
and now they have the right to modify that content, technically they can go and say, we own your likeness now because we can modify that same piece of content and number of times, an infinite number of times for all our different advertisers, right?
403
00:34:21.707 --> 00:34:26.691
So being able to get really specific in the language, I think we're going to see a lot of cases where
404
00:34:27.612 --> 00:34:32.873
creators and or brands are not, you know, they're kind of vague with those terms or they're silent on them.
405
00:34:32.893 --> 00:34:35.694
And we're going to see some big lawsuits coming out.
406
00:34:36.114 --> 00:34:42.196
Something happened where a brand used the rights they weren't supposed to and an agency used rights they weren't supposed to.
407
00:34:42.316 --> 00:34:44.116
And the creator didn't agree to that.
408
00:34:44.156 --> 00:34:53.359
And so it's going to be interesting, but we're definitely like in the Wild West sort of sort of era on how AI and influencer marketing is happening on that front.
409
00:34:53.924 --> 00:35:09.522
I was talking to somebody else who basically said they have all these shadow projects building their own influencers so they can get those, the terms they want, which is very murky in a lot of ways, legally, I think, um,
410
00:35:11.308 --> 00:35:18.512
But you can see why they would want that if it's so hard to, if everything is a negotiation until the contracts are standardized.
411
00:35:19.573 --> 00:35:24.715
It's like each one of those things is, you know, almost cost more than it's worth in terms of attention.
412
00:35:24.916 --> 00:35:29.118
Real quickly, you want to give the quick pitch for Mighty Joy and Super Deal?
413
00:35:29.418 --> 00:35:39.842
We're a creator agency focused on people with real expertise and domain knowledge, representing some of the world's largest brands and most trustworthy creators.
414
00:35:40.082 --> 00:35:46.325
Superdeal, that's been our tip of the spear into this AI first world, where we're working to automate
415
00:35:47.305 --> 00:36:07.797
a lot of the processes and workflows to create AI native marketplace for creators and brands, allowing small creators to have the same firepower the large creators have, and allowing brands and influencer marketing teams at brands to scale up with substantially less friction.
416
00:36:07.837 --> 00:36:12.180
Really, the goal there is how do we give a small startup
417
00:36:13.341 --> 00:36:18.741
the same firepower on the influencer marketing side as some of the biggest Fortune 1000 brands out there.
