Episode 4 - Proprietary Tech with Cadence Financial Planning

WEBVTT
00:00:02.000 --> 00:00:10.560
Welcome to another episode of Bionic Advisor, where we take a deep dive into all of the amazing pieces of tech being used by financial planning firms up and down the country.
00:00:10.800 --> 00:00:15.359
If you're a financial planner struggling to keep up on the tech treadmill, this is your home.
00:00:15.599 --> 00:00:25.600
I've got a bit of a different episode for you guys today because I'm joined by not one, not two, but three of the guys from Cadence Financial Planning.
00:00:25.760 --> 00:00:27.839
Alex, Craig, Keith, welcome.
00:00:28.000 --> 00:00:30.079
Thank you very much for joining us today.
00:00:30.480 --> 00:00:30.960
Hey Tom.
00:00:31.839 --> 00:00:32.240
Thanks, Tom.
00:00:33.359 --> 00:00:38.159
And uh this episode is not just a different episode because I'm outnumbered.
00:00:38.320 --> 00:00:45.920
It's also a different episode because it's the first one of what I hope is many uh what I'm going to call big picture episodes.
00:00:46.159 --> 00:00:58.240
And the idea behind a big picture episode is that rather than drilling down into one piece of tech specifically, I get to look under the bonnet of what some amazing forward-thinking firms are doing within their bus financial planning businesses.
00:00:58.560 --> 00:01:06.799
The hope from my side is that we can all share notes and uh about what maybe good looks like and everyone can improve their businesses as a result.
00:01:07.359 --> 00:01:11.920
I don't want to go into too much detail in fear of spoiling these guys' story.
00:01:12.079 --> 00:01:21.040
Uh, but to give you a bit of background, I reached out to Keith earlier this year, uh, and fair to say, Keith, that we shared a mutual love of being self-confessed geeks.
00:01:21.599 --> 00:01:41.040
Uh and in particular, I think it's fair to say that we shared an opinion which I keep referring to as my soapbox opinion, which is that financial planning firms of the next decade will be firms that own all of their own propriety tech, made and maintained in-house, built exactly how they want it.
00:01:41.519 --> 00:01:48.799
And I think it's fair to say uh at Cadence, you are, if you're not already doing that entirely, you are well on the way.
00:01:49.040 --> 00:01:57.920
Uh and I think if you're a financial planner and you're planning to still be in this industry in 10 years' time, then you seriously need to be considering doing the same, in my opinion.
00:01:58.239 --> 00:02:00.640
Uh right, let's get into it, shall we guys?
00:02:01.680 --> 00:02:01.840
Good.
00:02:03.439 --> 00:02:03.680
Good.
00:02:03.840 --> 00:02:04.159
Right.
00:02:04.319 --> 00:02:04.640
Okay.
00:02:04.959 --> 00:02:09.039
So for anyone listening who doesn't know Cadence, what's the firm?
00:02:09.199 --> 00:02:10.080
Who's involved?
00:02:10.159 --> 00:02:13.280
And what's the quick version of how it came together?
00:02:13.439 --> 00:02:15.360
Should we have that one with you, Keith?
00:02:15.680 --> 00:02:16.240
Yeah, yeah.
00:02:16.400 --> 00:02:21.039
Um, so Cadence started trading last November.
00:02:21.520 --> 00:02:28.080
Um we were formed, the three of us worked at uh Benchmark Capital together for a number of years.
00:02:28.240 --> 00:02:36.319
Um, we all left at different times, and I did a couple of couple of years in the wilderness uh just thinking about next steps, possibly looking at retiring.
00:02:36.400 --> 00:02:42.159
And then uh the three of us went on a cycle trip uh back in almost exactly two years ago.
00:02:42.319 --> 00:02:46.000
Um and they unretired me within about a minute.
00:02:46.159 --> 00:02:49.919
Um, and we came up with this idea that we might start our own firm.
00:02:50.080 --> 00:02:51.759
So we started putting a bit of thought to it.
00:02:51.919 --> 00:02:58.400
And uh a year after that we uh we basically went live and uh it's been a pretty journey so far.
00:02:58.479 --> 00:03:07.840
And one of one of the things we decided to do early on was look at because we all come from an organization that has built its own kit, uh, you know, back office platform client portal.
00:03:08.080 --> 00:03:14.319
Um we wanted to explore options of doing that again, but obviously on a much smaller scale with a much smaller firm.
00:03:14.479 --> 00:03:20.800
Um, because the tools that are available for everybody are uh, as we know, much more accessible than they would have been five, 10 years ago.
00:03:21.199 --> 00:03:21.840
Absolutely.
00:03:22.080 --> 00:03:29.439
I saw you shared recently, Keith, your uh original brainstorm mind map of what what you planned uh what you planned cadence to look like.
00:03:29.520 --> 00:03:32.639
And you said it's actually remarkably close to the original plan.
00:03:33.120 --> 00:03:42.879
Yeah, I mean, we'd spent a long time helping other firms grow and thrive, you know, it was part of the best practice network and um the directly authorized firms that used our kit back in the day.
00:03:42.960 --> 00:03:47.520
And we had second-hand experience of of how those firms ran.
00:03:47.919 --> 00:04:02.719
So, in a way, we've you know, our initial plan was to kind of cherry-pick the best bits of proper, as the trap guys would say, full fact financial planning firms, and um incorporate them into a business which was starting from a clean slate, and you know, that's pretty much how it's transpired.
00:04:04.000 --> 00:04:06.319
So I love the word unretired, by the way.
00:04:06.400 --> 00:04:07.520
That that's a great one.
00:04:07.680 --> 00:04:20.959
So, Craig, once you'd managed to persuade uh Keith to unretire, what what was it that made you decide to go and kind of build your own thing rather than going inside a bigger group or a network?
00:04:21.360 --> 00:04:25.680
Yeah, uh so for me, uh I think we're all kind of the same boat.
00:04:25.839 --> 00:04:28.879
We we all did a fairly long stint at benchmark.
00:04:28.959 --> 00:04:30.560
Um I did 10 years.
00:04:30.800 --> 00:04:35.680
And in that time, we just saw phenomenal growth, a lot of success stories.
00:04:35.839 --> 00:04:39.600
Um, my role in particular um was very broad when I joined.
00:04:39.680 --> 00:04:44.000
Uh, I was rolling up my sleeves, getting getting involved in quite a number of things.
00:04:44.240 --> 00:04:50.240
And as the company grows, you find that you can't keep doing all those things uh on your own.
00:04:50.399 --> 00:04:53.360
You need to kind of narrow down your role a little bit.
00:04:53.600 --> 00:04:58.319
Um and I just felt after 10 years I I'd achieved a lot, I'd seen a lot.
00:04:58.399 --> 00:05:01.519
I think I speak for the other guys, they they were in the same boat.
00:05:01.759 --> 00:05:10.240
Um, like you says, we had access to over 200 firms, and through that journey we could see what was going well, what was not going so well.
00:05:10.720 --> 00:05:15.360
And uh there was a bit of me that just wanted to try something new, try something fresh.
00:05:15.519 --> 00:05:26.480
Um, and when I got speaking to Alex in in Brighton, um, and then Lately and Keith in uh uh the cycling trip, it it just all kind of we all had these great ideas, and uh yeah, that's where it all kind of came from.
00:05:26.560 --> 00:05:33.120
And we thought we could only do that if we were doing it for ourselves and not uh and not doing it for someone else.
00:05:34.160 --> 00:05:35.040
Is it fair to say?
00:05:35.120 --> 00:05:45.759
I mean, you mentioned Keith about about the the background with benchmark uh and Craig, obviously you've seconded that, that that was a big part of understanding what like good looked like and what you wanted to work towards.
00:05:46.399 --> 00:05:52.720
That background obviously has has influenced your decision to build your own tech stack, fair to say?
00:05:53.120 --> 00:05:54.720
It has from both sides, actually.
00:05:54.879 --> 00:06:04.399
One from having built a version of that, and obviously Alex was a massive part of that build of the platform and the integration of the APIs between our various different bits of software.
00:06:04.560 --> 00:06:11.759
But from my perspective, I was a non-exec on a number of firms when Craig and I were looking after the acquisition side of the benchmark business.
00:06:11.839 --> 00:06:18.240
Uh, I would get sort of parachuted in as a non-exec director and a number of different firms that we either took a stake in or bought outright.
00:06:18.480 --> 00:06:29.600
And you just saw some amazing examples of um how the existing technology was being best used and how firms were building around the edges where our kit maybe didn't quite hit the mark.
00:06:29.680 --> 00:06:36.800
They would be much earlier times, but uh they'd be thinking about their own solutions or or buying in third-party solutions to fill in any gaps.
00:06:37.040 --> 00:06:43.279
And that sort of metaphor wind forward three, four years, you know, we very much took that approach.
00:06:43.360 --> 00:06:48.800
It was saying, let's look at the market, let's um assess what's out there from third parties.
00:06:49.040 --> 00:06:54.560
If it's non-core to us, um, then let's build around the edges of the core stuff.
00:06:54.959 --> 00:06:55.519
Absolutely.
00:06:55.680 --> 00:07:00.319
I mean, on that on that note, Alex, do you want to just talk us through what is the tech stack?
00:07:00.480 --> 00:07:01.199
What does it look like?
00:07:01.279 --> 00:07:02.800
What have you guys built yourself?
00:07:02.879 --> 00:07:03.759
What's bespoke?
00:07:04.000 --> 00:07:08.319
What are you what you're bringing in somewhere and what and where how do you link those?
00:07:08.959 --> 00:07:15.040
Yeah, so I mean the the core stack we have is uh Wealth Portal, which is the kind of core CRM.
00:07:15.199 --> 00:07:22.399
And this is built by um Ian Cook, who was the benchmark CEO and um the CTO Aaron.
00:07:22.800 --> 00:07:37.680
Um but but um because back in Benchmark, Ian was very tech focused, and so the whole business was kind of tech was really at the centre, um, which is quite rare, I think, um going back what 20 years, you know.
00:07:37.839 --> 00:07:44.319
Um so I think Ian was very kind of visionary on that level to see where tech could really be kind of core to the business.
00:07:44.480 --> 00:07:48.240
Um and he he left Benchmark and basically started again.
00:07:48.560 --> 00:07:54.399
Um, but um kind of the old products were enabling fusion, the T-Rom and the platform.
00:07:54.480 --> 00:08:01.279
Um, but he's built this 12th portal uh with Aaron over the last um, six, seven years.
00:08:01.680 --> 00:08:03.920
Um so built from the ground up.
00:08:04.000 --> 00:08:06.879
Um and it does pretty much the bulk of what we need.
00:08:07.279 --> 00:08:08.480
Um I mean it's incredible.
00:08:08.560 --> 00:08:15.680
We'll go into the detail uh later on, but um yeah, and so as Keith said, you know, that that does the bulk of it.
00:08:15.759 --> 00:08:21.279
But um we found that we wanted a stronger brand presence as a company.
00:08:21.680 --> 00:08:28.000
And um so the the bits that are kind of missing from Walf Portal, we've started to build ourselves.
00:08:28.240 --> 00:08:37.039
Um so in particular, this kind of onboarding tool uh we've we've bought to kind of bring on new prospects, um but also suitability.
00:08:37.120 --> 00:08:45.440
Um we kind of looked around at the market and uh thought that actually, you know, the the SMART these days is the the AI LLM behind the scenes.
00:08:45.600 --> 00:08:54.639
So actually we kind of when you go direct to the LLM, you can kind of build your own reporting engine kind of relatively easy uh these days if you know what you're doing.
00:08:54.799 --> 00:09:03.519
So yeah, so though that that's the kind of core stack and obviously M365 and the kind of core recording app called Rink Central.
00:09:03.759 --> 00:09:10.720
Um but yeah, it's I've seen the onboarding tool and it is it is really cool, really, really cool.
00:09:10.960 --> 00:09:23.519
But uh I obviously you've just you've mentioned there about AI AI tools, and uh I think it's fair to say you can put me in my place, Alex, because uh I I use Claude quite a bit now.
00:09:23.679 --> 00:09:26.720
Uh but I am I am the epitome, and I do say this to everyone.
00:09:26.879 --> 00:09:30.879
The problem with AI in general is you don't know what you don't know, right?
00:09:30.960 --> 00:09:39.600
You know, it's this it's the same principle why uh I do capital gains tax reports for my clients and I use Claude to produce that.
00:09:39.759 --> 00:09:54.559
But if a client said to me, Oh, I've done my capital gains tax report in Claude, I would say, you are mad, like you are mental, because obviously I can just take one glance at that capital gains tax report, check the cells and go, yeah, that's right, they won't have a clue.
00:09:54.639 --> 00:09:55.919
They would just blindly assume.
00:09:56.080 --> 00:10:01.679
And obviously, I am well aware that when it comes to building software, that's the exact same problem, right?
00:10:01.759 --> 00:10:11.919
Is that to someone like yourself who is using Claude, and you're fair to say you are a huge fan just as much as I am, but you can spot when it's gone wrong in seconds.
00:10:12.080 --> 00:10:16.639
I can just walk blindly into an absolute minefield and just get blown to smithereens.
00:10:17.200 --> 00:10:17.759
Absolutely.
00:10:17.919 --> 00:10:28.879
And you know, it it's incredible because this onboarding tool, the the flow was, you know, I started with this um kind of wireframe design and claw design, which was great.
00:10:28.960 --> 00:10:33.759
You know, you kind of get these, you can iterate on these different uh UI styles and so on.
00:10:34.159 --> 00:10:40.559
But like you say, you didn't, I I took that and gave it to Claude and it it ran off and built the whole app in one kind of shot.
00:10:40.879 --> 00:10:47.519
And you know, at the time I was like, wow, this is amazing, you know, look at this, it's a fully functional, you can log in, it's got a database and all this.
00:10:47.600 --> 00:10:53.919
And but then I dug into the code and you know, looking at what it had produced, you know, there's so many problems with it.
00:10:54.159 --> 00:11:02.799
And you know, and like you say, unless you don't know these things, it's um you you would never know that there's anything wrong because it kind of on the surface it was functional.
00:11:33.360 --> 00:11:33.919
Yeah.
00:11:34.240 --> 00:11:43.519
Yeah, a software developer friend of mine, he always says you've try you have to treat it like the absolute most junior of junior of software developers.
00:11:43.679 --> 00:11:55.759
It's like you you'll you'll tell it, press the red button and then press the blue button, and it comes back and goes, I've pressed the blue button, and you're like, oh geez, no, no, that's not what I told you to do.
00:11:55.919 --> 00:11:57.600
You really have to spell everything out.
00:11:57.679 --> 00:11:59.519
And I and and that's the thing that I'm worried about.
00:11:59.600 --> 00:12:04.320
Is like I'm I'm building bits of software that are more uh hobby-ish, I go, I suppose.
00:12:04.559 --> 00:12:09.840
But for me personally, I can't go anywhere near anything with client stuff because I'm like, I don't know enough about it.
00:12:10.000 --> 00:12:14.240
You know, I need I need an Alex, that's what I need that in in my business.
00:12:14.559 --> 00:12:25.919
Um speak speaking of which, though, uh obviously I mentioned earlier on, I am a massive, massive believer that if you are looking to scale a business, this is the route you've got to go down, right?
00:12:26.080 --> 00:12:34.559
Like we we we have a huge problem in financial services, more so than any industry I know where we have the most fragmented tech stack.
00:12:34.639 --> 00:12:42.559
I think you know the average advisor is probably using somewhere between 10 and 20 bits of software on a regular basis just to service their clients.
00:12:42.879 --> 00:12:47.600
So I do believe that this is the direction that firms need to go in.
00:12:47.919 --> 00:12:54.559
But um for you, Keith, what would you say is a realistic need before a firm could attempt this?
00:12:55.519 --> 00:13:01.759
In terms of skill set or in terms of um the the solutions they need.
00:13:02.159 --> 00:13:07.360
Well, I think in terms of skill set, um, you need to be curious first and foremost.
00:13:07.519 --> 00:13:18.799
I don't think there's any issue in a sole advisor with an administrator researching what's out there either from third parties or things that they could build themselves.
00:13:19.039 --> 00:13:22.240
Um if it's just for their own use.
00:13:22.879 --> 00:13:30.240
And I think that's maybe where we differed a little bit is because we're in the process of building a firm that may eventually turn into 15 to 20 advisors.
00:13:30.639 --> 00:13:30.960
Exactly.
00:13:31.200 --> 00:13:37.039
To get consistency across not just the code base but also the usage of those tools.
00:13:37.600 --> 00:13:46.159
You just need an additional layer of um structure, which I think from a development perspective, you probably need somebody who's had development experience.
00:13:46.399 --> 00:13:49.200
Um that's why you know you need an Alex, exactly.
00:13:49.759 --> 00:13:52.960
Um But it's not to say you can't try things yourself.
00:13:53.039 --> 00:14:03.519
I I as you said, I think we'll come on to talk about it later, is the concern about hallucinations and incorrect answers can be partially solved by inserting a human at the right place.
00:14:03.919 --> 00:14:14.799
And that's that's what we've done in all our applications so far is that yeah, there's a lot of automation, but there's always human intervention before something goes back to the client or could be sort of taken out of context.
00:14:15.679 --> 00:14:18.879
Yeah, the human in the loop that we keep talking about is massive, isn't it?
00:14:18.960 --> 00:14:21.679
You've gotta you've gotta have someone checking it at this stage.
00:14:21.840 --> 00:14:25.519
It's not at the point of uh of being flawless.
00:14:27.919 --> 00:14:43.039
In the um in the spirit of uh of this podcast, which I do always say we are honest but not cynical, Craig, what would you say has been the biggest headache of doing it this way that people don't talk about enough?
00:14:43.360 --> 00:14:50.240
Building everything from scratch, nothing off the shelf, just trying to every day iterate on your own stuff.
00:14:50.559 --> 00:14:54.639
Uh I mean you're asking an accountant not to be cynical, um, but I could be very honest.
00:14:55.200 --> 00:14:57.759
You could be you could be a bit cynical, I'm not very cynical.
00:14:58.559 --> 00:15:02.480
Uh no, well, I mean, ultimately it's our responsibility here.
00:15:02.559 --> 00:15:06.879
Um, we've got to take responsibility for what we build, we haven't outsourced it to anyone.
00:15:07.200 --> 00:15:10.000
Um, there's constant technology changes.
00:15:10.159 --> 00:15:22.480
Uh there is a period earlier on this year, I think Q1, where Alex, this is his background, he was we're constantly having conversations on Monday morning about the latest thing that we were reading over the weekend, and that was just in the tech world.
00:15:22.639 --> 00:15:27.440
You then have a regulatory uh landscape that that is offering change as well.
00:15:27.519 --> 00:15:29.200
So you combine those two.
00:15:29.360 --> 00:15:34.720
There's this constant movement, and uh, it's not perfect, the outputs change.
00:15:34.960 --> 00:15:39.600
So uh we're talking about the suitability uh producer uh report producer.
00:15:40.159 --> 00:15:46.159
Um you can run that twice uh and come up with slightly different uh reports.
00:15:46.320 --> 00:15:49.519
The the context is the same, but they'll look different sometimes.
00:15:49.759 --> 00:15:52.559
So you always need that human oversight into it.
00:15:52.639 --> 00:15:59.600
Um what that means is uh the biggest headache is we often have to roll our sleeves up and do things that are slightly outside of our kind of comfort zone.
00:15:59.840 --> 00:16:04.240
Um, but you know, that that's what we knew we were getting into, and that's what we're happy to do.
00:16:04.720 --> 00:16:06.480
Yeah, it's an investment as well, right?
00:16:06.559 --> 00:16:13.360
You know, it's it's it's very easy to just get something off the shelf that already does it and save yourself the hundred hours of building it.
00:16:13.519 --> 00:16:17.919
But obviously, once you've built it, you've got it and you can reap the benefits for the rest of time.
00:16:18.080 --> 00:16:31.759
So I I mean, I even I in my tiny little firm, I experienced that regularly is it worth me spending 20 hours on building something that's gonna save me, you know, if it's gonna save me 20 minutes on each client, it's a no-brainer.
00:16:31.919 --> 00:16:38.720
But if it's gonna save me a minute on each client, that's gonna take a lot longer for it to it to pay pay dividends for me, right?
00:16:40.159 --> 00:16:55.279
Yeah, there's that, but there's also, I mean, we went through a full market suite um last year, and we saw Saturn Advisory AI, Intelliflow, Planner, lots of different um tools that had overlapping functionality.
00:16:55.440 --> 00:17:09.359
Um but the nicest thing for me in building our own kit was having absolute control and the end result and having it very much tailored towards our advice process, our client proposition, our investment proposition, and the outcomes.
00:17:09.680 --> 00:17:14.400
So much that way outweighed the effort of actually building it.
00:17:14.559 --> 00:17:18.720
You know, it was it was having that level because we're all control freaks at the end of the day, you know.
00:17:18.799 --> 00:17:26.160
Um as Craig mentioned earlier, it was or sorry, I'll I'll mention earlier, it was control of brand, but it was also control of experience.
00:17:26.319 --> 00:17:30.880
Um, particularly with things like the onboarding tool where it's a client's first experience of you.
00:17:31.440 --> 00:17:38.640
Um, I wouldn't have wanted to trust that to any third party, not because they're not competent, but because it would we couldn't define it exactly as the way we wanted it.
00:17:39.039 --> 00:17:40.799
Exactly, totally agree, Keith.
00:17:40.880 --> 00:17:48.799
This is the thing I'm keep keep preaching on about is every firm in the UK has their processes slightly different.
00:17:48.960 --> 00:18:01.839
And it's there's no right or wrong answer, but everyone has a way that they've developed and owned over time, and there is no firm that can tell me that when they get one of these off-the-shelf solutions, oh yeah, guess what?
00:18:02.000 --> 00:18:03.839
It works exactly how we work.
00:18:04.240 --> 00:18:11.519
To some degree, they have had to shape their process to fit the mold of the tech that they are using.
00:18:11.680 --> 00:18:16.559
Whereas if you build it yourself, you can do the opposite, and I just think that that's so powerful.
00:18:16.880 --> 00:18:26.559
Um, and also because uh you know the issue with this fragmented tech stack is well the original solution was having open AI.
00:18:27.039 --> 00:18:28.559
Sorry, open APIs.
00:18:29.039 --> 00:18:38.880
But the uh the issue with open APIs is one, you're reliant on both parties to maintain that API and decide what's actually gonna be available through the API.
00:18:39.119 --> 00:18:44.559
Um but also it's it's not perfect, like stuff goes wrong.
00:18:44.720 --> 00:19:01.680
Um I I'm not gonna call anyone out on this podcast, but there are a lot of companies that are claiming to have links between each other, and when you go and use it as an actual user, you go, This is barely functional, like barely, like so many errors.
00:19:01.920 --> 00:19:04.960
And obviously, having the tech in-house, you'd have to worry about any of that.
00:19:05.039 --> 00:19:08.240
Like you can fix it yourself, you can prioritize things yourself.
00:19:08.480 --> 00:19:22.480
If you've got this one part of your process that every other firm in the country considers really, really minor, and therefore, when you pitch it to the software that you're using and say we would really like this as an improvement, it falls to the bottom of the pile.
00:19:22.559 --> 00:19:26.319
I think, Keith, you you used to me the term death by democracy.
00:19:26.480 --> 00:19:31.279
Uh I think it's a perfect example of the problem.
00:19:31.440 --> 00:19:44.000
Whereas if you are in charge yourself, it might seem really, really niche to any other firm, but if to you it's a central part of your proposition, you can literally accelerate it and get it out as quickly as possible, which I think that's the most powerful thing.
00:19:44.319 --> 00:19:44.880
Yeah, absolutely.
00:19:44.960 --> 00:19:47.359
And it could be a competitive differentiator in some respects.
00:19:47.519 --> 00:19:56.640
If you're able to do things that 95% of other firms, or even the big national firms, can't do or have never thought of doing, get it to market quickly.
00:19:56.799 --> 00:20:00.079
I mean, the the example I'll give you is the the onboarding tool itself.
00:20:00.160 --> 00:20:06.799
We we sat around, the five of us sat around, I think it was on a Thursday out and started sketching things out.
00:20:06.880 --> 00:20:09.039
And by Monday, we had a working prototype.
00:20:09.200 --> 00:20:14.720
And it was it was released into our our version of the wild within about a week of that.
00:20:14.880 --> 00:20:21.680
And, you know, I mean, not pointing any fingers, but we've we've all known places where that's taken a year, 18 months to get to that stage, you know.
00:20:21.759 --> 00:20:32.799
So I I think there's a real a real kind of spirit within cadence of try quickly, finesse, and then either continue or just forget about it and and go again.
00:20:32.960 --> 00:20:41.119
And and you know, we've got we've done that with it with the I think we're on version 12 or 13 of the of the suitability tool, and it's four months old, you know.
00:20:41.200 --> 00:20:46.000
So it's uh um iteration is is really good for the soul, I think.
00:20:47.200 --> 00:20:49.680
Constantly, constantly, that's never finished, is it?
00:20:49.759 --> 00:20:50.400
That's the problem.
00:20:50.559 --> 00:20:51.039
Yeah, exactly.
00:20:51.279 --> 00:20:59.200
Um Al, we we were talking before about uh, you know, I joked that everyone needs needs an Alex on their on their team.
00:20:59.519 --> 00:21:11.599
So if there are any planners, say like myself listening, that they have the appetite, but maybe not the technical skill in-house, do you genuinely think at this stage it's even possible for them to do something like this?
00:21:11.920 --> 00:21:21.119
Or option B, is it something that you think will be possible in the future, but it's not the AI isn't there right now, or is it something they always need an Alex?
00:21:21.359 --> 00:21:22.880
Is that gonna be your your answer?
00:21:23.279 --> 00:21:25.359
Well, that's probably been my default, yeah.
00:21:25.599 --> 00:21:31.119
But uh, you know, I I follow it a lot, and um I think you know it's nearly there.
00:21:31.279 --> 00:21:38.400
I mean, this this term vibe coding, you know, where you can basically build apps without ever looking at the code.
00:21:38.559 --> 00:21:41.519
Um and they're they're getting better and better.
00:21:41.759 --> 00:21:50.960
Um I think the things that you know we're seeing in the in the kind of industry is it's the classic thing of software where I think building that first version is actually quite easy.
00:21:51.119 --> 00:21:55.119
You know, you kind of no users, you're just you know knocking out the features, it's great.
00:21:55.359 --> 00:22:03.759
Um but once you start, you know, getting into production, you've got people using it, relying on Everyday maintaining software is quite challenging.
00:22:04.240 --> 00:22:08.160
In particular, you don't want to break old features when you release new ones.
00:22:08.240 --> 00:22:11.200
And you're going to hear stories of these kind of vibe-coded apps.
00:22:11.920 --> 00:22:14.799
The quality of the code is probably not the best.
00:22:14.960 --> 00:22:17.759
And so yeah, maintaining it becomes difficult.
00:22:18.240 --> 00:22:23.119
Security as well, you know, it's obviously a massive, you know, part of our industry.
00:22:23.200 --> 00:22:29.599
And again, you hear horror stories of vibe-coded apps that connect to a database with no security at all.
00:22:29.759 --> 00:22:31.599
So all the data can be leaked.
00:22:31.759 --> 00:22:36.000
And I think like you mentioned earlier, like if you don't know, you you don't know what you don't know.
00:22:36.079 --> 00:22:37.839
So you wouldn't know these things are there.
00:22:38.880 --> 00:22:46.720
But having said that, you know, I just I'm just optimistic, you know, like um, you know, internally, both Keith and Craig have been building apps.
00:22:46.880 --> 00:22:59.759
Um so I see my role as more just trying to, you know, guide guide the kind of process and you know, making sure data's secure, but um and you know, providing the data you know to them to allow them to build what they want to do.
00:22:59.839 --> 00:23:07.039
Because I think what what's happening is it's you know, business users know what they want more than the developers, obviously.
00:23:07.200 --> 00:23:12.160
So building your own stuff just how you want it is um you know, it's it's incredible.
00:23:12.240 --> 00:23:15.440
And it's um yeah, like I say, it's so nearly there.
00:23:15.599 --> 00:23:24.960
So um, you know, yeah, the technology is moving on so quickly that um I wouldn't be surprised if we're kind of not far off pretty perfect.
00:23:25.680 --> 00:23:41.680
Keith, uh Keith has joked of me before, and he said that you know you've had experience with uh running your own teams of software developers, and actually your your team that you're running now at Cadence is your favourite, and that's because it's all just sub agents of Claude.
00:23:42.079 --> 00:23:45.759
Do you want to talk me through like what what what kind of subagents are you using?
00:23:46.000 --> 00:23:52.880
Because uh I think it's a it's a an area that certainly people who are not software developers, I think it's underutilized at the minute.
00:23:52.960 --> 00:24:06.559
Um I I did a post on LinkedIn a while ago talking about my use of subagents, which is actually in in my suitability reports, which is you know, getting each subagent to check the various aspects of the report in isolation.
00:24:06.640 --> 00:24:11.519
So, you know, one being like a compliance check, one being just a general copyright proofread check.
00:24:11.759 --> 00:24:20.559
Um, but what what are you using, someone who obviously knows a lot more than me on a daily basis, what sort of subagents are are running and how are they communicating with each other?
00:24:21.279 --> 00:24:23.359
Yeah, so I mean it it's fascinating, isn't it?
00:24:23.519 --> 00:24:27.920
Because I think this um this technique and things move so quickly.
00:24:28.079 --> 00:24:32.960
So I started off um I created a subagent team within Claude Code.
00:24:33.200 --> 00:24:37.920
So um uh yeah, this is you know obviously the the kind of coding part of Claude.
00:24:38.319 --> 00:24:45.279
Um and so what I did was created different personas that kind of matched how we used to write code in in a big organization.
00:24:45.440 --> 00:24:48.480
So you'd have a product owner that would define the spec.
00:24:48.720 --> 00:24:56.799
Um so I'd work with them to build this spec, and then they'd output that, that would be handed over to an architect who could design a kind of high-level architecture.
00:24:56.960 --> 00:25:04.559
Then you have developers that then take both of those and implement features and um testers that do the testing and then a code reviewer.
00:25:04.720 --> 00:25:13.440
And um yeah, it's just it was incredible, you know, because it's kind of like you know, a whole virtual team that kind of just does exactly what I tell them to do.
00:25:13.599 --> 00:25:16.799
You know, I can control them, it's bullying, no one talks back.
00:25:17.680 --> 00:25:20.799
They don't need holiday, they all agree through the night.
00:25:21.119 --> 00:25:21.440
Yeah.
00:25:22.400 --> 00:25:27.680
I've already joked in Arboretum that uh there's there's only myself, my wife, Emma, and Claude.
00:25:27.759 --> 00:25:32.240
And I have already joked to them that Claude's gonna win employee of the year every single time.
00:25:32.480 --> 00:25:34.240
So she's got no chance.
00:25:34.480 --> 00:25:40.079
He works through the night, he's never stops, never complains, he's he's the best employee.
00:25:40.559 --> 00:25:42.079
Yeah, it's brilliant, isn't it?
00:25:42.160 --> 00:25:42.720
You know, yeah.
00:25:42.880 --> 00:25:48.319
The um yeah, and so but that um that kind of that worked really well.
00:25:48.400 --> 00:26:02.799
But then since then, you know, things just moved so quickly that I've actually moved on from that to um there's a really good plugin called Superpowers, which has a kind of an opinionated process um for developing, and that's using spec-driven development.
00:26:02.960 --> 00:26:12.160
Um so again, you kind of brainstorm the requirements, um, but you do that really quite uh a lot more intensely, and it will output this specification.
00:26:12.400 --> 00:26:29.680
And then at that point you review it, iterate it, and once that's you know done, you then hand that over to its own kind of sub agent um uh driven development, and that's using uh has developer uh testers and code review.
00:26:29.839 --> 00:26:53.920
But um it's using this kind of loop engineering concept now, which is this idea that you know you have you know the output that you you're kind of your goal, what you're kind of aiming for, um, and then you have a way of evaluating whether you've reached that goal so that you know you do the work or it does the work, you then have something else review it and says, well, this is good, but this needs you know tweaking or so on.
00:26:54.000 --> 00:27:00.000
And you kind of loop on that, it's you know, it uh iterating on that um until it's done.
00:27:00.240 --> 00:27:10.960
And so the idea behind that is you know, if you have if you get it right, you can literally kick stuff off at night and come back in the morning to kind of this whole kind of working system.
00:27:11.440 --> 00:27:14.240
And yeah, that's kind of the frontier labs are doing now.
00:27:14.319 --> 00:27:15.599
It's um yeah, incredible.
00:27:15.920 --> 00:27:20.400
How much would you say then are you uh of what you're getting um the agents to run?
00:27:20.559 --> 00:27:26.079
How much of that is genuinely unattended compared to you know how much work we talked about human in the loop before?
00:27:26.160 --> 00:27:30.240
What what sort of percentage are we talking of terms of stuff that you need to intervene in?
00:27:30.960 --> 00:27:36.000
Yeah, I mean, well, I mean, I'd say uh nearly all of the code has been written by Claude.
00:27:36.160 --> 00:27:40.640
Um so my role, the human in the loop, is now just reviewing the code.
00:27:40.880 --> 00:27:53.200
So I use um GitHub and uh so they've got a pull request kind of processed, kind of a well-established software engineering where you know anyone who does any work, you kind of review it before it gets um approved.
00:27:53.359 --> 00:28:04.559
Um so that's my role really now, is just you know reviewing the code to make sure it's um you know, um adhering to the the kind of standards that the coding standards that that I've set.
00:28:04.799 --> 00:28:10.480
But even that, you know, I've got um on GitHub you can have another AI agent to do the review with you.
00:28:10.640 --> 00:28:14.880
So it does its own review and then I review its review and add my own.
00:28:15.039 --> 00:28:19.759
So it's um yeah, very nearly kind of no need for me at all.
00:28:22.400 --> 00:28:28.960
Have you got any stories, Craig, of where like it might have uh gone wrong, or maybe just where you don't trust the agents yet?
00:28:29.039 --> 00:28:34.160
Like what any any no-goes as to what you're not gonna let uh AI build for you?
00:28:34.799 --> 00:28:39.599
Uh yeah, so uh no-goes.
00:28:39.839 --> 00:28:45.200
We haven't identified any hard stops on that, but there's a few bits at the moment we wouldn't let it near.
00:28:45.680 --> 00:28:51.039
Um, particularly around uh just letting it run wild and send out reports without checking them.
00:28:51.279 --> 00:28:53.680
I I've I think you mentioned about CGT.
00:28:53.759 --> 00:29:00.240
I I've noticed just some fundamental errors that that it's coming back with, and that that's right across the piece, not just Claude.
00:29:00.880 --> 00:29:03.759
Um it hallucinates quite often.
00:29:03.920 --> 00:29:05.359
Um let's be real.
00:29:05.599 --> 00:29:15.359
Um I've I've had an instance in the suitability report where it um it raised a we've got this kind of secondary check.
00:29:15.440 --> 00:29:21.839
So a bit like you, we have a we we have a section that builds report, and then we have a section within Claude that reviews reports.
00:29:22.079 --> 00:29:41.279
And the reviewer came back with a uh regulatory issue that our fee that we were charged for this particular client was over uh 0.75% for the advice fee, which was uh higher than the FCA threshold, and therefore it required a higher level of consumer duty.
00:29:41.359 --> 00:29:46.079
And I thought that that's 0.75, that's not a rule I know.
00:29:46.400 --> 00:30:02.720
And uh when I queried it, what it had can what it had done, um it had found an article somewhere online in one of the kind of money marketing or uh Citywire, whereby there was a there was an article about fee charges, and there is a concentration, as we know, around 1%.
00:30:03.039 --> 00:30:05.440
There's a concentration of fees around 0.75.
00:30:05.920 --> 00:30:14.960
And it could it it took from that that that was a threshold that the FCA would deem as as appropriate for enhanced um consumer duty reporting.
00:30:15.680 --> 00:30:20.079
So I queried it, and it it it it kind of apologized, didn't apologize back to me.
00:30:20.240 --> 00:30:36.799
Um so I think it just goes back to my an earlier point that it it it's it's a great tool for a lot of things, uh but I I I'm still very conscious of what it can do, what it shouldn't be doing, and it definitely needs that second pair of eyes, and uh I yeah, don't let it loose yet.
00:30:36.960 --> 00:30:38.799
But uh it's a powerful tool.
00:30:38.960 --> 00:30:41.519
Um yeah, it keeps evolving, like I say.
00:30:41.680 --> 00:30:57.440
Um anything, anything that we sorry Tom, I was gonna say anything that we um that we think would be uh from a reputational perspective or from a regulatory perspective, or even from just a presentational perspective, client impactful.
00:30:57.599 --> 00:31:01.440
We always have somebody who will check something that's been automatically generated.
00:31:01.519 --> 00:31:07.519
You know, whether it's suitability report, whether it's um flyers that go out to uh you know newsletters, all that kind of stuff.
00:31:07.680 --> 00:31:14.319
So it's I and it quite often, as Craig suggested there, that your own expertise can spot things very quickly.
00:31:14.559 --> 00:31:28.720
So I think that's where we're we're getting into is that it's not just checking the work, it's actually um using your experience and your understanding of the subject area to just spot instantly when something just feels out of context or incorrect.
00:31:29.519 --> 00:31:29.920
That's it.
00:31:30.000 --> 00:31:31.119
It's like I was saying before, isn't it?
00:31:31.200 --> 00:31:32.319
It's you don't know what you don't know.
00:31:32.480 --> 00:31:41.440
And if you're using it in a subject area that you are very, very confident in, it's absolutely amazing because it can save 90% of time, do all of the grunt work.
00:31:41.599 --> 00:31:44.400
And if it does get something wrong, you'll spot it in a heartbeat.
00:31:44.480 --> 00:31:46.079
It's just it's not there yet.
00:31:46.240 --> 00:31:52.799
Like with like Alex is saying, it you know, it might get there because it's certainly improving, but it's not there yet for stuff you actually don't know.
00:31:52.880 --> 00:31:54.000
Uh but there we go.
00:31:54.480 --> 00:32:03.119
But we but we had um just to bring that to life in an example, the first case that we ever had as cadence was uh quite a complex pension consolidation case.
00:32:03.200 --> 00:32:06.240
There was other bits and pieces around, so not a simple piece of advice.
00:32:06.480 --> 00:32:18.480
And because we didn't have any full-time staff uh who were administrators of power planners at the time, we outsourced the whole amount of of that work to uh a third-party uh powerplanning admin.
00:32:18.640 --> 00:32:24.000
And it was about 35 hours worth of work from end to end to get the uh the work done.
00:32:24.319 --> 00:32:30.160
Uh, as part of the testing for the, I think it was a version three or four of the suitability report generator.
00:32:30.319 --> 00:32:35.279
Uh, Craig ran exactly the same case through the engine and it took us two and a half hours.
00:32:35.519 --> 00:32:49.440
Uh and I think you mentioned at the time, Craig, most of that was clawed time, most of that was just waiting on responses rather than so the actual effort of checking is actually a relatively small part of the overall uh workflow, but it's it's so vital.
00:32:49.839 --> 00:32:50.960
Yeah, absolutely.
00:32:51.039 --> 00:32:51.920
I'm a big fan of this.
00:32:52.000 --> 00:32:54.799
I I this is this is what I've been uh referring to.
00:32:54.880 --> 00:33:03.680
I don't think I've gone into full detail on it on LinkedIn, but uh I obviously started like everyone else of just seeing if we can get Claude to write a suitability report.
00:33:03.839 --> 00:33:10.880
And I think the first one took something like 25 minutes, um, and it was just you know full of errors.
00:33:10.960 --> 00:33:20.400
It I think into that first one took me about an hour and a half to uh fix, I mean, uh how much uh fix it and how much just generally rewriting you're doing at that point.
00:33:20.559 --> 00:33:46.000
Um, but when I started doing the subagent approach, like Alex was referring to, um, and I think it's it's running at four or five subagents now that I have, and essentially we have the the original agent that's writing the report, and then it goes to each of the each of the subagents, and they have their own criteria to uh pass or fail, kind of very similar to what Craig was describing with the uh the 0.75%.
00:33:46.319 --> 00:33:48.400
They all have their own certain criteria.
00:33:48.480 --> 00:34:02.720
Uh, and they can either they can pass the report, say, yeah, it's good to go, uh, they can reject it uh and give grounds of what they need to change in before they'll accept it, or they can uh accept it but say that there are notes that they want the advisor to read.
00:34:02.880 --> 00:34:12.719
And then at the end, when the report has finally been generated, perhaps forms a separate note that uh I can read through where each agent has like a, yeah, you just might want to be aware of this.
00:34:12.880 --> 00:34:22.800
Um like one example was the um one of the subagents approved it, but they just noted that the client's retirement age didn't match up to the illustration.
00:34:22.880 --> 00:34:29.760
Um so obviously rejecting the suitability report would be stupid in that case because it'll never get past it because it was just something that we had missed.
00:34:29.920 --> 00:34:32.400
Um but yeah, that was just a note that we we tweet.
00:34:32.559 --> 00:34:34.719
Um so the these these run circular.
00:34:34.800 --> 00:34:39.679
So all of the subagents, they'll just keep sending it back until we get to five loops, because I don't want it to run forever.
00:34:39.840 --> 00:34:44.079
Uh, if we get to five loops and they've still not signed it off, then uh it gets sent back to me.
00:34:44.239 --> 00:34:45.679
I've never got there yet, that's good.
00:34:45.760 --> 00:34:47.599
Uh we've never got to the five loops.
00:34:47.760 --> 00:34:55.840
Um but obviously the downside of this, and that's what you were referring to there, is the report writing time has now gone up to something like three hours.
00:34:56.159 --> 00:34:57.440
But do I care about that?
00:34:57.679 --> 00:34:58.000
No.
00:34:58.239 --> 00:35:00.400
I don't, I don't have to sit there for three hours.
00:35:00.559 --> 00:35:06.079
Most commonly it will run when I'm asleep and I'll wake up in the morning and there'll be a suitability report.
00:35:06.159 --> 00:35:12.000
And I'm now down to, in terms of my check-in time of suitability reports, like 10 minutes, 15 minutes.
00:35:12.239 --> 00:35:18.000
So although the time taken has gone up, just like you said, a huge chunk of that is Claude time.
00:35:18.079 --> 00:35:20.480
I I don't, sorry, Claude, but I don't care about your time.
00:35:20.559 --> 00:35:21.920
You can spend as long as you like.
00:35:22.159 --> 00:35:27.360
What I want is my check-in time to go from an hour and a half, hour and a quarter down to 10 minutes.
00:35:27.519 --> 00:35:33.360
So even though the overall time has gone up, the use of my time has come down.
00:35:33.599 --> 00:35:34.159
Yeah, absolutely.
00:35:34.400 --> 00:35:37.360
So I think we need more, more, more needs to be doing stuff like this.
00:35:37.440 --> 00:35:40.000
Uh trying to convert as many people as possible.
00:35:40.480 --> 00:35:50.719
On that note then, Keith, if you have got a financial planner listening who likes the sound of what we're talking about, but they they are loyal to the core systems, you know.
00:35:50.800 --> 00:36:06.880
Uh I I I I'm really reluctant to name them, but uh you you know, the likes of Advisory AI, Marlo, Saturn, any of these off-the-shelf tech solutions, they're happy with them, they get it, they don't have to go under the bonnet, they don't have to understand it.
00:36:07.199 --> 00:36:13.039
If they were just tempted by what we're talking about, what would your direct message be to them?
00:36:13.760 --> 00:36:19.119
Well, I think come and chat to you or I about it and just uh show them what we've built.
00:36:19.360 --> 00:36:27.519
You know, and I I think we are a firm of six people, you know, and uh you can punch above your weight from a technology perspective.
00:36:27.679 --> 00:36:32.159
If you're happy with the solutions, I think it's still worth looking around, you know.
00:36:32.239 --> 00:36:36.400
Um, because you may be happy in a context where the the world has moved on around you.
00:36:36.480 --> 00:36:44.000
You might you might just be thinking that things are okay when uh you know there are other solutions out there that could significantly improve your lot.
00:36:44.079 --> 00:36:54.719
And I think it's it's always my advice would be always be curious, you know, um spend a bit of time on on web conferences and and the sort of latest thinking type podcasts.
00:36:54.880 --> 00:37:07.519
Um try things, you know, even if it's uh like I when I started out with Claude, the first thing I did was build a financial personality calculator and uh d determine your child's financial superpower app.
00:37:08.239 --> 00:37:13.360
It was just something that no other provider or um core solution would build.
00:37:13.440 --> 00:37:16.400
So I just thought, well, let's just do something that's completely out there.
00:37:16.800 --> 00:37:27.119
And um, you know, it was it was what it was, but it it really got me interested in the way that Claude works and and the thinking that needs to go into building things with Claude.
00:37:27.519 --> 00:37:39.039
Um But I just yeah, stay curious, um ask, you know, you know, I think Alistair Walker's got a peer group that runs on a regular basis that uh I think he tries to get together every few months.
00:37:39.119 --> 00:37:42.960
Um I I think you mentioned you were planning to attend it if you're not part of it already.
00:37:43.039 --> 00:37:56.239
But um I think finding groups of like-minded people who who just can throw some ideas around and excite you, you know, that's that's uh you know, as we're both geeps, that's what really is interesting to me is to learn off others and share as well.
00:37:56.400 --> 00:38:00.000
So if anyone is interested in anything that we do, we're not licensing it.
00:38:00.079 --> 00:38:06.559
You know, I mean it's it's it's not very transferable to anyone else, but the uh essence of what we've built, we're more than happy to share.
00:38:07.840 --> 00:38:10.159
And Keith, you and I have talked about this before.
00:38:10.320 --> 00:38:14.559
The we are looking for northern tech geeks in financial services.
00:38:14.719 --> 00:38:20.239
If there's any in the north, reach out to us because me and Keith are desperate to find more members.
00:38:20.480 --> 00:38:22.960
It's one of the one of the things I'm hoping as part of this.
00:38:23.119 --> 00:38:23.920
We want to get more.
00:38:24.079 --> 00:38:26.239
We want to build a little community of us geeks.
00:38:26.400 --> 00:38:27.920
Like we we have some good ideas.
00:38:28.239 --> 00:38:29.440
We're the future.
00:38:29.760 --> 00:38:31.679
Absolutely, absolutely.
00:38:32.000 --> 00:38:38.239
Um, Craig, if you were if you were starting this whole journey again from scratch, what would you have done differently, do you think?
00:38:39.760 --> 00:38:43.599
Um I don't I don't think there's an awful lot I would have I would have changed.
00:38:43.760 --> 00:38:51.119
Um knowing what we know now, there was a there was a period where we picked a slightly different LMM at the beginning.
00:38:51.280 --> 00:38:56.960
Um I'm struggling for anything really, but uh you know, it'd be if it'd have pivoted towards Claude a bit quicker.
00:38:57.119 --> 00:38:59.199
Um What did you go over at the start?
00:39:00.880 --> 00:39:04.239
Uh it was we were using a uh OpenAI.
00:39:04.559 --> 00:39:06.960
Um and and it was fine.
00:39:07.039 --> 00:39:11.119
It and it's a bit like the d Donning Kruger uh curve.
00:39:11.360 --> 00:39:14.480
You know, what we were using, we didn't know any better, it was like great, this is fantastic.
00:39:14.719 --> 00:39:27.440
Then I started to hear a bit of noise about Claude and and Al had already gone, yeah, I I I've been using it personally for for other things, and uh I think we spent a few days just looking through it and we um we pivoted the following week.
00:39:27.679 --> 00:39:28.800
It was as quick as that.
00:39:29.039 --> 00:39:34.079
Uh very much realistically, we might really our strategy.
00:39:34.159 --> 00:39:38.880
Keith uh Keith shared his mind mark uh on LinkedIn, anyone wants to go and look at it, it's still there.
00:39:39.119 --> 00:39:41.280
Uh that's what we set out at the beginning.
00:39:41.440 --> 00:39:45.840
I uh we we had a bit of a chuckle because we looked at it and we thought that's pretty much where we are now.
00:39:46.079 --> 00:39:53.519
Tactically, we might have done little things slightly different in a different order, but strategically we we are where we are, and that's where we plan to be.
00:39:53.679 --> 00:39:55.840
And I wouldn't change any of it really.
00:39:55.920 --> 00:39:58.079
It's it's been an absolute fantastic journey.
00:39:58.159 --> 00:40:01.519
I've learned a lot uh over the last uh 12 months.
00:40:01.679 --> 00:40:02.880
It's been it's been brilliant.
00:40:02.960 --> 00:40:09.119
Um so uh not to say we knew what we were doing from the beginning, not far from it, but yeah, I wouldn't I wouldn't have changed anything.
00:40:09.440 --> 00:40:12.320
Just wish you'd started the uh hop to Claude sooner.
00:40:12.480 --> 00:40:13.599
That's the same as me.
00:40:13.679 --> 00:40:15.519
I I was definitely I'd open AI.
00:40:15.599 --> 00:40:23.039
Well, I I remember when when we first, you know, if you if you think back to when ChatGBT first came out and everyone was like, chat what?
00:40:23.280 --> 00:40:38.400
You know, uh and then we were immediately hearing stories of like where ChatGBT had written essays for university students and it had got these kids in high school A plus grades, and you see the essays that it wrote, and everyone just kind of saw it as magical this first time.
00:40:38.800 --> 00:40:49.280
And then I think that lasted about two months, and then I just seemed to spend every single waking hour with it after that, just being disappointed of like, oh, it could do oh no, it's failed.
00:40:49.440 --> 00:40:50.400
No, it's got it all wrong.
00:40:50.480 --> 00:40:51.840
No, it's hallucinated.
00:40:52.159 --> 00:40:59.679
And I just I just found myself just constantly arguing with it, and then I was like, why am I arguing with this nonsentient being?
00:40:59.840 --> 00:41:01.119
Like it's just ridiculous.
00:41:01.280 --> 00:41:08.400
And it was actually um, it was Alan Smith that could that that that kind of nudged me in the direction of uh of Claude.
00:41:08.880 --> 00:41:13.280
And I remember I got I got the one month free trial, I think.
00:41:13.519 --> 00:41:23.199
And um I I I I think the first thing I did was I was just interested, could it um could it write something in my tone of voice?
00:41:23.360 --> 00:41:28.480
Because every time I would read I I'd see stuff on LinkedIn, I'd be like, that's ChatGPT.
00:41:28.559 --> 00:41:30.320
Like I can just tell from the wording.
00:41:30.719 --> 00:41:35.280
Uh and I remember I just I gave it some of my articles and I said, Can you write something in my voice?
00:41:35.440 --> 00:41:40.559
And it it wrote, and I remember looking at it and going, Did I write that?
00:41:40.719 --> 00:41:50.400
Like and at that point I was like, I was like, ah, okay, yeah, it's time to cancel my open AI subscription and move over to Claude.
00:41:50.639 --> 00:41:53.039
And I never looked back, never looked back after that.
00:41:53.119 --> 00:41:56.000
And it's I've I've been Claude's been my best friend ever since.
00:41:56.239 --> 00:41:56.639
Yeah.
00:41:58.719 --> 00:41:59.760
I was just gonna add to that.
00:41:59.840 --> 00:42:05.039
Sorry, uh, there is one other thing that we have we haven't touched on really, but Claude is quite handy.
00:42:05.199 --> 00:42:10.000
Um on uh to do the co co-working part, which is a feature they have.
00:42:10.239 --> 00:42:11.119
It's it's very handy.
00:42:11.199 --> 00:42:16.800
We haven't really touched on that, and it's probably a separate section, but um people should go and look that up if they if they're interested.
00:42:16.960 --> 00:42:25.679
So, what I've used that for is we had a load of policies, internal policies, compliance, HR, tech, all the policies to expect the company to to have.
00:42:25.760 --> 00:42:32.400
And and anyone who's looked after policies knows it's it's hard work, it's it's constantly checking regulation updates and so on.
00:42:32.639 --> 00:42:39.440
I just set Claude to look at that folder on our system, and I said, here's all our policies.
00:42:39.519 --> 00:42:45.119
We're a UK-regulated firm, we're subject to FCA, HMRC, Companies Act, everything.
00:42:45.360 --> 00:42:46.960
Give it whole scope.
00:42:47.199 --> 00:42:59.119
And I said, organize it, rebrand it, version control it, give it the proper um authority it needs, and give me a diary um for review.
00:42:59.280 --> 00:43:02.639
So some of them needed a six-month review, some of an annual review, and everything.
00:43:02.880 --> 00:43:08.159
And it started the process and it asked me four or five questions before it really got going.
00:43:08.320 --> 00:43:12.000
I walked away for an hour, came back, and I could not believe what I saw.
00:43:12.159 --> 00:43:14.559
And then it said at the end, do you want me to put this in your diary?
00:43:14.719 --> 00:43:19.599
Now we have a rule about um Outlook and Claude, we don't we don't connect the two.
00:43:20.000 --> 00:43:21.519
But it was phenomenal.
00:43:21.599 --> 00:43:29.519
It literally just did everything, and it's it's already it it found a couple of conflicts where one policy said one thing and it conflicted with another in another document.
00:43:29.599 --> 00:43:31.519
It said which one's the correct one.
00:43:31.760 --> 00:43:37.440
It it basically did probably, I reckon, one month's worth of work in an hour.
00:43:37.840 --> 00:43:41.519
Uh so check check that out if you if you're interested.
00:43:41.840 --> 00:43:48.960
Yeah, you've mentioned another point there, actually, Grey, which I'd forgotten uh because it's so long ago going back to my my open AI days.
00:43:49.119 --> 00:43:54.559
But that was another thing that surprised me in my early days of using Claude is that it actually asks questions.
00:43:54.719 --> 00:44:00.880
Like, you know, you say to it, Hey, do this job, and OpenAI goes, No problem.
00:44:01.039 --> 00:44:02.079
I'm off to work.
00:44:02.239 --> 00:44:05.039
And you're like, I've not even given you the data yet.
00:44:05.199 --> 00:44:07.199
Like, how are you expecting to do this job?
00:44:07.360 --> 00:44:09.760
Like, I was going to upload the files in a minute.
00:44:10.000 --> 00:44:16.800
Whereas, like, as soon as I was doing the same things with Claude, it'd be like coming back straight away, like, hmm, seems like this information is missing.
00:44:16.960 --> 00:44:17.760
What do you want me to do?
00:44:17.840 --> 00:44:20.480
And it's got the recommended or the other options or type your own.
00:44:20.559 --> 00:44:26.000
And then you'd go through like almost like a mini wizard, and you'd be like, I'd be like thinking, oh gosh, yeah, I didn't actually give it that.
00:44:26.079 --> 00:44:30.719
Like, how was I expecting it to write a suitability report when I didn't even tell it the client's attitude to risk?
00:44:30.880 --> 00:44:32.800
Like it's just little things like that.
00:44:32.960 --> 00:44:38.079
Uh, whereas open AI just would pure blag it, just be like, well, he's not told me, but let's just go for it.
00:44:38.239 --> 00:44:39.280
Uh, there are nine.
00:44:39.360 --> 00:44:42.320
Uh, you're like, where did you get that information from?
00:44:42.559 --> 00:44:44.880
Hey, I'd I'd forgotten that until you just said.
00:44:45.119 --> 00:44:47.039
Yeah, it's it's great.
00:44:47.599 --> 00:44:50.079
Alex, where's where's the tech stack going next?
00:44:50.239 --> 00:44:54.400
What are you uh what are you building or automating now that you weren't able to 12 months ago?
00:44:54.559 --> 00:45:01.119
Obviously, recently Claude has launched uh Fable 5.1, is it, if I'm remembering my numbers right?
00:45:01.280 --> 00:45:03.199
Uh, which I've had a bit of a play with.
00:45:03.280 --> 00:45:10.400
And um, well, I'm I'm simply just happy to have a Fable version that's uh included in your membership and is not is not billable by the token.
00:45:10.559 --> 00:45:14.400
But uh has that opened anything new up or more generally?
00:45:15.360 --> 00:45:21.199
Yeah, I mean I think well, we're still on the kind of team plan, so we don't get that as part of our license.
00:45:21.360 --> 00:45:27.119
We have to pay the usage, but we you know we've got um a bunch of free credits at the moment, we're so we're trialing it out.
00:45:27.280 --> 00:45:30.960
Um but yeah, I mean that it's another step change, isn't it, Fable?
00:45:31.599 --> 00:45:34.159
Um I mean it's just changing all the time.
00:45:34.320 --> 00:45:39.039
So obviously, as you know, like I think GPT 6 Astra was released last night.
00:45:39.199 --> 00:45:44.400
So that's uh OpenAI, you know, they did definitely lag behind Claude.
00:45:44.559 --> 00:45:57.840
I think they kind of I think they spread themselves a bit too thin, um, you know, with all the kind of video stuff they're trying, but they uh they've gone back and focused on the kind of enterprise and coding and they've kind of got got stronger again.
00:45:57.920 --> 00:46:04.559
So it's uh you know, it just changes so often, like one nodes ahead and the other one, you know, and so on and so forth.
00:46:04.719 --> 00:46:12.320
So um, you know, probably be looking at OpenAI just to see what their you know their the tool's up to from a coding point of view anyway.
00:46:12.639 --> 00:46:15.039
Um but definitely Claude, we'll stick with Claude.
00:46:15.679 --> 00:46:25.280
Um but the thing that's enabling us to do is to do uh you know, the the two kind of big projects that we've kind of got in mind are being more proactive for clients.
00:46:25.519 --> 00:46:35.360
So um, you know, with all the data now that we can access it through Wealth Portals API, um we want to do this kind of quarterly um PDF review.
00:46:35.440 --> 00:46:45.119
Um so rather than just kind of waiting for the annual review, we can kind of be a bit more proactive and say, well, here's a kind of bunch of things we've checked on your kind of financial plan.
00:46:45.440 --> 00:46:46.880
Are you on target for your goal?
00:46:47.039 --> 00:46:49.920
Is there anything that's kind of going a bit off piste?
00:46:50.559 --> 00:47:01.679
Um so you know, it's a great example of how the tech can be, you know, bring us more proactive, kind of bringing more value to the client um on a regular basis.
00:47:01.840 --> 00:47:11.199
And and the other one is um cash flow modelling that Craig, you know, he's he's building, you know, his own because obviously you know it's um he's kept that quiet.
00:47:13.599 --> 00:47:23.519
Uh you know, with his background and you know, Excel, you know, wizardry, um again, it's just you know, one of those where people can build their own.
00:47:23.599 --> 00:47:28.800
And so my role is to kind of look at how how can we get the data in there and sort of make it skewer and and so on.
00:47:28.960 --> 00:47:32.079
So uh but yeah, it's exciting, you know.
00:47:32.400 --> 00:47:37.360
I've I've been talking about this a lot because I am a I'm a massive Voyant user, so I'm a big, big fan of Voyant.
00:47:37.519 --> 00:47:38.960
I do think it's very, very good.
00:47:39.519 --> 00:47:46.719
But coming back to the whole thing, I'm I'm reliant on Voyant to push new updates, I'm reliant on what they decide is a priority to work on.
00:47:46.960 --> 00:48:00.159
And I mean, they've been telling me for over 12 months now that they've got a module for uh businesses, you can't really simulate limited owner, you know, a director of a limited business, you can't put the company in, you can't, it's really clunky.
00:48:00.639 --> 00:48:03.119
And they've been telling me for over a year that it's nearly coming.
00:48:03.440 --> 00:48:16.159
And with the stuff that I have used Claude to build that does similar stuff to a cashier modeling, I keep saying, and voyant users, even though I am a fellow voyant user, they they shoot me down straight away.
00:48:16.239 --> 00:48:17.599
They're like, Tom, you're wrong.
00:48:17.760 --> 00:48:18.239
You're wrong.
00:48:18.480 --> 00:48:40.480
So uh, but I'm I'm a big believer that if you if you have shares in these companies, I would be like, I'd be terrified because I think, you know, Claude, it's one of the first things that I think a firm could build their own version of is a cash flow modeling tool, exactly how they want it, using all of the assumptions that they want, being able to do all of the insights that they need to do.
00:48:40.719 --> 00:48:42.960
Because it's just it's just a gigantic spreadsheet.
00:48:43.039 --> 00:48:44.719
That's that's all a cash flow modeling is.
00:48:44.800 --> 00:48:47.599
It's just a visual version of a gi massive spreadsheet.
00:48:47.840 --> 00:48:58.719
And certainly in the period where I was allowed to use Fable free of charge, the stuff that it was outputting in calculations, it was finding errors that I wasn't finding.
00:48:58.800 --> 00:49:06.559
Like I was checking it, going through it, looking at every formula, and then I would give it to Fable, and Fable would go, Yeah, but there's also these errors.
00:49:06.639 --> 00:49:09.280
And I'd look at them like, oh yeah, there is.
00:49:09.599 --> 00:49:20.800
So I don't think we're that far away from firms like Cadence, like Arboretum, being able to say, actually, I don't want an off-the-shelf cash for modeling solution.
00:49:20.880 --> 00:49:23.199
I want to build my own exactly how I want it.
00:49:23.360 --> 00:49:27.599
And then you've got the luxury of it being able to come into your client portal and everything.
00:49:27.840 --> 00:49:30.000
Is that the direction you're going for, Craig?
00:49:30.719 --> 00:49:38.400
Yeah, there were there were a couple of limitations that we identified using we realities cash calc and then we we move to Voyant very quickly.
00:49:38.480 --> 00:49:46.480
Um and there are limitations because you you can't build uh a system for thousands and thousands of users that's going to do everything for everyone.
00:49:46.800 --> 00:49:47.119
Exactly.
00:49:47.360 --> 00:49:52.559
So I I do think the mid-card SaaS firms might struggle um in in the long run.
00:49:52.800 --> 00:49:55.199
Um so let's wait to see where it's going.
00:49:55.280 --> 00:49:59.920
But we we've built we we've got a particular type of target market client.
00:50:00.000 --> 00:50:07.679
Um we built Claude to to cater, uh we've built a cash flow model through Claude to cater for the vast majority of those individuals.
00:50:08.000 --> 00:50:14.719
Um and I do know a few things that Voyant can't do that I've now added into mine.
00:50:14.880 --> 00:50:20.960
Um but I'm sure there's plenty of things in Voyant that we we've not yet um added into the features on ours.
00:50:21.199 --> 00:50:32.239
But the beauty, the the biggest beauty that we've got is I can produce a cash flow model with a client that's already on our system uh in Wealth Portal.
00:50:32.400 --> 00:50:38.320
I can take that fact find, I can put it into our cash flow model and have a cash flow within five minutes.
00:50:38.480 --> 00:50:41.920
Whereas we found with Voyant, you had to replicate a lot of the data.
00:50:42.079 --> 00:50:46.719
Uh the Monte Carlo simulation took uh quite a lot of time to run through.
00:50:46.880 --> 00:50:58.480
We had to look we use um some model portfolios that uh not widely available, and therefore we had to rebuild all those in Voyant, whereas I can just dump that in through FE into into Claude.
00:50:58.719 --> 00:51:03.679
So all those things just cut the corners for us and made it a lot easier to to navigate.
00:51:03.840 --> 00:51:04.639
So that's why we did it.
00:51:05.039 --> 00:51:08.639
It was out of necessity more than trying to cut costs or anything like that.
00:51:08.800 --> 00:51:10.880
And uh it's it's in a good place.
00:51:11.360 --> 00:51:12.880
Comes back to that story though, doesn't it?
00:51:12.960 --> 00:51:19.199
Of you know, owning your own tech, and this is the power is you can decide where to prioritize it, you can decide how it fits into your overall tech stack.
00:51:19.360 --> 00:51:25.039
Whereas if you've if you've got a circular piece that has been provided by Voyant, you can't put it in that square hole.
00:51:25.119 --> 00:51:26.559
You've just got to fit it where you can.
00:51:26.719 --> 00:51:44.000
Whereas if you build it yourself, it can exactly mold it, like you say, you can build it with your own clients in mind and what they're what they consider a priority to see in their cash flow model, that's what you can prioritize rather than loads of parts of Voyant that are very, very useful to other firms, but maybe not so much to your firm, right?
00:51:44.239 --> 00:51:45.360
Yeah, exactly that.
00:51:45.519 --> 00:51:46.000
Exactly that.
00:51:46.239 --> 00:51:54.639
And to your earlier point on while Voyant does have an API, it wasn't rich enough for our needs, didn't have all the kind of points covered.
00:51:54.800 --> 00:52:08.800
And um, you know, we're lucky enough that with Ian and Wealth Portal, um, pretty much any item of data that's that we've already keyed in through the onboarding tool in there, then into Wealth Portal, is available.
00:52:08.880 --> 00:52:13.440
In fact, it could go straight from onboarding tool into cash flow model, it doesn't even need wealth portal.
00:52:13.519 --> 00:52:16.960
It gives you options that we just wouldn't have if there was a third-party tool there.
00:52:17.280 --> 00:52:27.039
And it also allows us to um tailor the branding on the output and the experience of getting to that point entirely to Cadence's brand guidelines.
00:52:28.639 --> 00:52:31.119
The uh you mentioned the API link.
00:52:31.199 --> 00:52:42.559
The the one of the ones that I was alluding to earlier is the one between Voyant and Planner, which I try to use on a regular basis and always just decide that manual entry is actually faster.
00:52:42.800 --> 00:52:54.559
Because one of the things it likes to do is when the client has filled out the fact-find in planner and they've done all of their expenditure, bear in mind, you know, if you imagine each one is effectively at one field.
00:52:54.800 --> 00:53:05.119
So you're talking, you know, for a household you might have like something like between 30 and 40 different lines, and every single one voyant will ask me when I import, what is this?
00:53:05.360 --> 00:53:10.480
And you'll have to tell it it's an expense that occurs every year for this period of time.
00:53:10.719 --> 00:53:13.519
And you'll go through, and the thing is, I'm not even going to pull this data through.
00:53:13.679 --> 00:53:20.320
I don't want this data because I'm just gonna lump it all together as one like goal of like, you know, household spending, discretionary spending.
00:53:20.400 --> 00:53:30.880
But I still have to go through the wizard of answering 30 times this is a multi-year expense, this is a multi-year expense, this is a multi-year expense to get all the way through to then go, I don't actually want any of that.
00:53:31.119 --> 00:53:34.480
It's just it's absolutely awful to deal with.
00:53:34.639 --> 00:53:37.360
Um let's uh let's do a quick fire close.
00:53:37.519 --> 00:53:39.039
Thank you very much, guys, uh, for me.
00:53:39.119 --> 00:53:48.960
But this last question for each of you, you can answer in any order you like one bit of tech that you could not run your business without, and one bit of tech that you think is overrated.
00:53:49.119 --> 00:53:52.480
You can decline to answer either part of them if you don't want to be held to account.
00:53:55.039 --> 00:53:56.079
Who wants to go first?
00:53:56.239 --> 00:53:56.880
I I'll go first.
00:53:57.039 --> 00:53:58.320
I I I don't mind going first.
00:53:58.480 --> 00:54:01.519
Um, so tech we couldn't do without.
00:54:01.840 --> 00:54:09.360
Uh for me, we talked a lot about Claude, but you know, I I'm as faithful as to Claude as the next best AI model.
00:54:09.760 --> 00:54:13.280
So if something comes along that's better, I I'm I can jump ship.
00:54:13.440 --> 00:54:16.320
Uh but for me, wealth portal is key.
00:54:16.400 --> 00:54:18.239
Uh that that's such a great piece of kit.
00:54:18.320 --> 00:54:22.159
So I want to give a shout out to Ian and Aaron who and the on the team.
00:54:22.320 --> 00:54:23.599
It's it's an amazing bit of kit.
00:54:23.679 --> 00:54:29.360
We haven't talked about it much today today, but that that's the one thing we couldn't at the moment live without.
00:54:30.559 --> 00:54:35.199
Um and bit of should I do the next bit of tech that we're gonna do?
00:54:35.360 --> 00:54:40.079
Uh probably I I'll just say blockchain and leave out that.
00:54:40.559 --> 00:54:42.159
Okay, all right, yeah.
00:54:43.199 --> 00:54:45.199
Keith, Al, who wants to go next?
00:54:46.480 --> 00:54:48.480
Okay, yeah, yeah, you go Al, yeah.
00:54:49.039 --> 00:54:50.960
Um well for me, yeah.
00:54:51.039 --> 00:54:57.039
I mean, so the thing we couldn't live without, um, it's still um development tech.
00:54:57.360 --> 00:55:04.400
So you know, ID and and Azure to deploy all the code to, um kind of traditional, I guess.
00:55:04.559 --> 00:55:11.119
Um but yeah, I mean that's uh still fundamental to um to powering kind of web-based solutions.
00:55:11.199 --> 00:55:13.920
So that's that's yeah, that's mine.
00:55:14.079 --> 00:55:16.559
And then the the tech, what was it, overhyped?
00:55:16.639 --> 00:55:17.920
Um overhyped.
00:55:18.159 --> 00:55:18.800
Yeah, I don't know.
00:55:20.400 --> 00:55:22.800
Well, I guess at the moment, maybe the robotics.
00:55:23.039 --> 00:55:33.039
Um because hearing how they're gonna kind of take over the world, but I don't know if you've seen the uh the kind of robotics Olympics recently and just uh it's not quite there, to be saying.
00:55:33.280 --> 00:55:33.599
Yeah.
00:55:33.760 --> 00:55:34.559
Um button.
00:55:34.880 --> 00:55:36.079
Not quite as entertaining.
00:55:36.239 --> 00:55:36.480
Yeah.
00:55:36.719 --> 00:55:38.960
Well maybe it is entertaining, but just for very wrong reasons.
00:55:39.280 --> 00:55:39.920
Yeah, exactly.
00:55:40.000 --> 00:55:40.320
Yeah.
00:55:40.480 --> 00:55:43.440
Lots of explosions and yeah, malfunction.
00:55:43.679 --> 00:55:47.840
So yeah, and also like 80% of the entrants are Chinese.
00:55:47.920 --> 00:55:50.800
Like there's it's not really quite representative of the whole world yet.
00:55:51.039 --> 00:55:52.159
Yeah, exactly.
00:55:52.400 --> 00:55:53.599
What about you, Keith?
00:55:54.239 --> 00:55:57.199
So Al mentioned about building a virtual development team.
00:55:57.360 --> 00:56:00.400
I built myself a virtual marketing department using Canva.
00:56:00.800 --> 00:56:10.000
Um, and we use the Claude connector into Canva so that you can do plain text uh generation of any marketing document to set a brand guidelines.
00:56:10.480 --> 00:56:14.079
Um, it is absolutely phenomenal, and it's like 20 quid a month per user.
00:56:14.159 --> 00:56:15.679
You know, it's uh they've got their pricing.
00:56:15.760 --> 00:56:27.119
I don't know how they ever make any money, I know they do, but it is it is phenomenal when you pair it with Claude and the skills that you can inject into Canva and get back again.
00:56:27.360 --> 00:56:49.360
Um, I mean, I I just a good example and do uh coming out of retirement from a presentation perspective and doing a chamber of commerce presentation in a couple of weeks' time, and I got Claude to do a draft script for the presentation, and I said, could you just build the presentation in Canva with graphics and uh pull it back and I'll I'll review it and then we can get to version two.
00:56:49.519 --> 00:56:57.280
And it did that in about three minutes with some real live examples pulled from the internet about things I wanted to talk about.
00:56:57.360 --> 00:57:03.440
And you know, it was 95% there in a presentation that would have taken me three or four days to do otherwise.
00:57:03.760 --> 00:57:05.679
Yeah, so I'm a big, big fan of Canada.
00:57:05.760 --> 00:57:14.079
I'm not I'm not using the integrations anywhere near enough, but I I do I do love Canva and I am always surprised at how little I have to pay for it compared to anything in the industry.
00:57:14.239 --> 00:57:16.000
What about your overrated tech, Keith?
00:57:16.079 --> 00:57:16.880
What would you say?
00:57:17.199 --> 00:57:19.440
The metaverse for me is the most overrated.
00:57:20.800 --> 00:57:27.920
I uh I got my kids uh a VR headset, uh MetaQuest 3 uh last Christmas, Christmas 24.
00:57:28.320 --> 00:57:40.079
And um, if you want the future of business to be nine-year-old kids playing virtual paintball with each other and swearing uh to the extent that their parents should be ashamed, that's what the metaverse is.
00:57:40.559 --> 00:57:43.360
Yeah, I think that's probably exactly what Mark Zuckerberg had in mind, doesn't it?
00:57:44.719 --> 00:57:46.960
I worriedly think that may be the case, actually.
00:57:48.159 --> 00:57:50.480
Okay, uh that's us dumber for another episode.
00:57:50.639 --> 00:57:53.760
Alex, Craig, Keith, thank you all very much for joining me.
00:57:53.920 --> 00:57:59.519
Uh if you found this episode useful, please do leave us a review, like, share, subscribe, all the good stuff.
00:57:59.599 --> 00:58:07.519
Uh, and more importantly, don't forget if you've got an idea for a piece of tech that you want us to cover on the podcast, head over head over to bionicadvisor.co.uk.
00:58:08.079 --> 00:58:10.559
Submit that contact form, and I will be in touch.
00:58:10.719 --> 00:58:13.199
Until next time, look after yourselves, everyone.
00:58:13.360 --> 00:58:14.480
Thank you, guys.
00:58:14.800 --> 00:58:15.679
Cheers time.
00:58:15.920 --> 00:58:16.239
Bye-bye.