文字稿 🔗
WEBVTT
00:00:00.160 --> 00:00:05.120
Chief sustainability officers are not a really durable role in themselves as a C-suite role.
00:00:05.280 --> 00:00:06.000
We're seeing that.
00:00:06.080 --> 00:00:11.759
I've seen within my network already multiple chief sustainability officers be removed from their role.
00:00:11.919 --> 00:00:16.640
That role is likely being downgraded to a different level of hierarchy within the organization.
00:00:17.120 --> 00:00:21.280
And most sustainability teams as sustainability functions are also taking a hit.
00:00:21.440 --> 00:00:26.480
Now, it is not the 100% value for 100% of the cost.
00:00:26.559 --> 00:00:29.440
It is the 80% of value for 60% of the cost.
00:00:29.600 --> 00:00:34.960
And I think that's what sustainability teams, at least for the next few years, need to get used to as a concept.
00:00:40.799 --> 00:00:43.920
Welcome back to another episode of the State of Sustainability.
00:00:44.000 --> 00:00:47.679
I'm your host, Sef Hamid, founder and CEO of Altruistic.
00:00:47.920 --> 00:00:52.799
And I'm uh excited to spend this episode chatting about AI.
00:00:52.960 --> 00:01:00.159
We've done a few episodes on AI on this podcast, but we've done different flavors, and this is a rapidly evolving space.
00:01:00.320 --> 00:01:09.920
So longtime listeners may remember our episode on Centaurs and Cyborgs, which I think is already almost a year old, at least eight, nine months old.
00:01:10.079 --> 00:01:23.519
And in that episode, we talked about how there are these two different ways of using AI, where the cyborg approach was where you kind of have a human taking a task and interchangeably with AI performing the task.
00:01:23.680 --> 00:01:30.319
So for instance, let's say you're working on an email and you're refining the email and you're asking AI to improve little bits of the email.
00:01:30.480 --> 00:01:31.760
That's a cyborg model.
00:01:31.840 --> 00:01:37.120
You're kind of iterating away, but fundamentally you're both working towards the same little outcome or little piece of work.
00:01:37.280 --> 00:01:43.280
And the Centaur approach was where you basically say, okay, fine, this is your task, this is my task.
00:01:43.439 --> 00:01:53.519
So for instance, I'm going to get an AI to uh fully draft the email, and I'm going to work on a little uh table of numbers that I'm going to embed in the email later.
00:01:53.599 --> 00:01:58.319
But the AI is fully writing the text, and I'm fully working on the quantitative bit, for instance.
00:01:58.560 --> 00:02:00.400
Those aren't my original terms.
00:02:00.560 --> 00:02:10.800
I stole them from a report that was done between Harvard Business School and BCG, where they had a range of consulting teams working on using AI.
00:02:10.960 --> 00:02:18.240
And they basically were able to see that all the teams that used AI did better than the teams that didn't on the metrics that they were tracking.
00:02:18.400 --> 00:02:21.439
And the teams that did use AI used it in these two different approaches.
00:02:21.680 --> 00:02:35.039
And the reason that I'm kind of mentioning that is to recap a little the episode that we ran many months ago, but also to say that I feel like that is already almost entirely out of date because the space has evolved so much.
00:02:35.199 --> 00:02:42.960
At Altruistic, we are trying to really be at the frontier of this topic in terms of how we use AI within our business.
00:02:43.120 --> 00:02:53.919
And we're sort of, I would say, at least a year and a half into our journey, where the first thing was how do we get AI to run most of our product engineering work, for instance?
00:02:54.000 --> 00:03:02.400
So AI writing 80% or more of our code, assisting in all of our code, and really leverage it to the max in terms of our own product development.
00:03:02.639 --> 00:03:08.800
Next phase was how do we start bringing it to customers and embedding the use of AI in our product, in our platform?
00:03:09.039 --> 00:03:14.879
And then the third is kind of how do we actually have agents running amock across all of our business operations.
00:03:15.199 --> 00:03:27.360
And so based on the experiences we're going through and have been through, I wanted to spend time on this episode talking about AI, agentic AI, and AI native services.
00:03:27.520 --> 00:03:29.680
And I see those as three different things.
00:03:29.919 --> 00:03:33.360
In some ways, it's like a bit of an evolution, but in some ways not.
00:03:33.520 --> 00:03:42.319
And I want to talk about what I see as the implications for sustainability teams and sustainability professionals of AI, agentic AI, and AI native services.
00:03:42.479 --> 00:03:55.759
And I think this is an important topic to talk about because, in my experience, most large corporations are actually far behind on these topics versus where they could be based on where the technology is.
00:03:56.080 --> 00:04:05.120
And most sustainability teams, I would say, seem to me to be at least where their organizations are and maybe even in some ways a little further behind.
00:04:05.199 --> 00:04:12.159
And so I want to spend this time just talking about it in that sustainability context to help bring everyone onto the same page.
00:04:12.400 --> 00:04:20.000
So maybe let's start with a quick recap of AI because we tend to use AI and LLMs now interchangeably.
00:04:20.240 --> 00:04:22.480
AI and LLMs are obviously not the same thing.
00:04:22.720 --> 00:04:28.720
AI is a broad space which includes many different approaches layered on top of each other.
00:04:28.879 --> 00:04:41.759
So you might have, let's say, uh machine learning, you'll have deep learning, you'll have rule-based AI, and then you'll have, you know, these are maybe different foundational approaches of AI, and then you'll have different applications built on top.
00:04:41.920 --> 00:04:55.680
And so large language models are where you where you have a model that has been trained on lots
Load full transcript
of text, uh, lots of language, and is predicting the next best word, basically, the next word in the sequence that is that is likely based on what you've given it.
00:04:55.759 --> 00:05:00.879
Uh, but you might also have a version of this that is based on pixels where it's trying to predict the next pixel.
00:05:01.040 --> 00:05:04.399
Think about those captures that we've all uh we've all done.
00:05:04.639 --> 00:05:09.519
And then you you might have other models as well that are based on predicting the next data object, for instance.
00:05:09.680 --> 00:05:13.040
So you'll you could have different different engines or different applications running this.
00:05:13.199 --> 00:05:23.759
We have over the last few years all plunged into LLMs, and LLMs are sort of right now being seen as the dominant model or the dominant approach or application.
00:05:23.920 --> 00:05:27.120
There are lots of flaws or weaknesses with LLMs.
00:05:27.199 --> 00:05:29.120
So, for instance, they're prone to hallucinations.
00:05:29.279 --> 00:05:33.680
Some people talk about how the hallucinations can be erased over time.
00:05:33.839 --> 00:05:40.480
Others say actually you're never gonna get rid of the hallucinations because the hallucinations are intrinsic to how LLMs operate.
00:05:40.639 --> 00:05:45.360
Uh, sycophancy is another thing that we're all aware of where the LLM is always trying to please you.
00:05:45.439 --> 00:05:52.000
Uh, you know, some people say you can get rid of the sycophancy and it only crept in in certain releases and then became prevalent.
00:05:52.079 --> 00:05:55.360
Uh others say actually the sycophancy is always gonna be a factor.
00:05:55.600 --> 00:06:03.040
Another thing is they're really expensive to run because they tend to be these large models running with huge amounts of compute and server capacity and energy.
00:06:03.279 --> 00:06:06.720
Obviously, there are experiments to try and make those lighter, touch, and easier.
00:06:06.800 --> 00:06:14.319
Uh so point being, there are lots of drawbacks of LLMs, but at the same time, LLMs have right now emerged as a dominant model.
00:06:14.480 --> 00:06:17.920
I'm saying this just to make sure that our listeners are aware.
00:06:18.160 --> 00:06:27.279
I'm just saying this to make sure that you are aware that it may well be that in three years we're talking about a different application and we've all decided actually LLMs are toast.
00:06:27.519 --> 00:06:29.040
Uh but for now, let's run with it.
00:06:29.120 --> 00:06:30.160
Let's talk about LLMs.
00:06:30.240 --> 00:06:31.199
And so this is AI.
00:06:31.279 --> 00:06:37.120
And so if you think about, let's say, how AI has entered the workplace, uh, we're using AI.
00:06:37.199 --> 00:06:47.360
You know, most people in most companies are going to be using ChatGPT either formally or on the side, or Copilot, or Claude, or whatever they might might be using to help with their everyday tasks.
00:06:47.600 --> 00:06:52.240
And that's not
agentic AI, that's just using AI-based chat, really.
00:06:52.399 --> 00:06:58.000
That's using a chat-based product that is leveraging an LLM and you're using it yourself.
00:06:58.240 --> 00:07:07.040
The step from AI to agentic AI comes in when you start to connect up that AI to tools and to context.
00:07:07.199 --> 00:07:14.000
And you start to say, well, I'm going to give this AI access to certain products or applications that I'm using.
00:07:14.160 --> 00:07:19.519
So let's say I give an LLM access to my email, for instance.
00:07:19.759 --> 00:07:23.680
And so now actually this AI can access my email directly.
00:07:23.759 --> 00:07:27.920
It can read my emails and it can summarize all my emails for me.
00:07:28.079 --> 00:07:33.920
And maybe it can also write emails and it can push those emails to my email client or application.
00:07:34.079 --> 00:07:37.199
Let's say it's Gmail, and it can maybe even uh send that email.
00:07:37.360 --> 00:07:44.399
Currently, you can't send it from Gmail directly using an AI, but uh, let's say on Superhuman or many other email clients, you can.
00:07:44.560 --> 00:07:49.120
And so that's me connecting up an AI to tools, and you can connect it up to many tools.
00:07:49.279 --> 00:07:59.920
Now, let's say I also give that AI context, and so that context might be a combination of, let's say, instructions and standing information that that AI needs to perform its task.
00:08:00.000 --> 00:08:09.120
So let's say it always needs to know that Seth only writes emails in formal British English, always has capitalized eyes, always uses impeccable grammar.
00:08:09.279 --> 00:08:16.319
Uh, but because Seth grew up in an emerging market, the standard for his written English was generally kept quite high all through his life.
00:08:16.560 --> 00:08:20.079
And so that's how you should you should write his emails for him.
00:08:20.319 --> 00:08:23.120
And you know, maybe there's certain instructions given alongside that.
00:08:23.199 --> 00:08:28.959
So I've basically given this AI access to some fixed context, almost like a personality.
00:08:29.040 --> 00:08:32.799
Uh, and I've also given it access to tools so it can actually go and do stuff.
00:08:32.960 --> 00:08:39.279
And I've therefore created an agent, an agent that can act on my behalf to perform functions and deliver outcomes.
00:08:39.440 --> 00:08:42.080
This is important because we're gonna come back to it later.
00:08:42.320 --> 00:08:53.360
Now, most companies that I observe, and I think about our customers and companies that we work with at Altruistic, are really early in their use of agents in an advanced way.
00:08:53.519 --> 00:08:57.679
That doesn't mean that there are not pockets within every business that are making use of agents.
00:08:57.759 --> 00:08:58.639
I'm certain there are.
00:08:58.720 --> 00:09:09.360
I'm certain that IT teams and digital transformation teams and the like are experimenting with agents in every large business today,
but it hasn't yet entered the mainstream of the business.
00:09:09.600 --> 00:09:32.320
And what I mean by that is that, you know, the average function, the average team within a large company, a Fortune 500 business, is not getting agents to perform tasks either autonomously, which means fully independently, or even with a human in the loop, an H-I-TL process where a human is sort of embedded within a process that is otherwise being run by agents.
00:09:32.559 --> 00:09:37.759
Uh, therefore, needless to say, sustainability teams are even more early on this.
00:09:38.000 --> 00:09:40.000
So that is agentic AI.
00:09:40.080 --> 00:09:42.960
Now let's come to AI native services, which is an even newer space.
00:09:43.120 --> 00:09:51.200
Just to contextualize this, I talked at the start about how AI had this sort of cyborg central approach and AI was, you know, how AI was developing.
00:09:51.279 --> 00:09:53.120
That was only eight, nine, ten months ago.
00:09:53.440 --> 00:09:55.039
Agentic AI is even newer.
00:09:55.279 --> 00:10:08.159
I would say that a company like ours, which is inherently forward-leaning, being a software company, we've been experimenting with agents at some scale starting eight, nine months ago, at much more significant scale starting three months ago.
00:10:08.240 --> 00:10:17.360
And I would say it's six months from now, it will be ubiquitous within our business or everywhere within our business, embedded in most processes.
00:10:17.519 --> 00:10:19.679
Uh, so agentic AI is still pretty new.
00:10:19.840 --> 00:10:22.080
AI native services, I would say, is even newer.
00:10:22.320 --> 00:10:31.039
And I don't expect most of our customers or companies that we work with to have really come across AI native services in a big way, but you will within a few months.
00:10:31.120 --> 00:10:38.320
As in, you will start getting pictures, you'll start seeing it being offered to you by consulting firms and uh solution providers.
00:10:38.559 --> 00:10:50.960
AI native services is where you basically say, I now have a hybrid where I have humans in the loop on processes, and I have agents built all around those humans.
00:10:51.440 --> 00:10:58.960
And uh this combination of human plus agents is responsible for delivering outcomes autonomously.
00:10:59.120 --> 00:11:03.519
And so it is substituting for a person on the side of the business.
00:11:03.679 --> 00:11:12.159
So you could say, for instance, that let's say there's a supply chain manager within your company, and the supply chain manager is responsible for two or three different outcomes.
00:11:12.320 --> 00:11:24.480
Uh, one is let's say they need to understand the risk of environmental or human rights or other catastrophe across their suppliers.
00:11:24.559 --> 00:11:33.440
Uh, they need to gather data from their suppliers to complete, let's say, their scope three footprint, you know, whether it's PCFs or corporate carbon footprints or whatever, they need to get that together.
00:11:33.600 --> 00:11:39.600
And they need to perform, let's say, some sort of environmental audit on a supplier when a new supplier is being onboarded.
00:11:39.679 --> 00:11:44.799
I've taken three relatively simple outcomes that that supply chain manager is responsible for.
00:11:45.039 --> 00:11:55.039
One is this kind of broad running scan of risks, second is a data acquisition, and third is an onboarding related diligence of a supplier, those three outcomes.
00:11:55.279 --> 00:12:16.000
Now, in an AI native services world, the way that that might be delivered is that a solution provider might come to you and say, look, I can deliver these three outcomes for you as a business, and I can take responsibility for those outcomes, and I'm going to deliver those outcomes to you with a combination of agents and humans in the loop or humans running processes.
00:12:16.240 --> 00:12:29.360
One of the key tells of AI native services versus just consultancy is that AI native services tends to anchor on outcomes, whereas consultancy tends to anchor on capacity.
00:12:29.679 --> 00:12:43.039
And so usually when a consultant sells you a project, they will say, I'm gonna give you a manager and two people for four weeks full time, or I'm gonna give you a person at 50% capacity for six weeks or something.
00:12:43.120 --> 00:12:45.279
It is usually a capacity-linked model.
00:12:45.519 --> 00:12:51.759
And the way that they charge for it or bill for it, or the economics internally, are also on this sort of day rate or per diem.
00:12:52.159 --> 00:12:54.159
I say this as a recovering consultant.
00:12:54.320 --> 00:13:09.759
Whereas with AI Native Services, they don't want to pitch to you on that basis because they don't want to be beholden for giving you a certain amount of capacity because they are going to automate a lot of the work that they're gonna do for you, and therefore locking in capacity doesn't make sense economically.
00:13:09.919 --> 00:13:14.000
So they will commit on the basis of outcomes and they will say, I will own these three outcomes.
00:13:14.080 --> 00:13:14.960
Let's recap.
00:13:15.200 --> 00:13:19.360
One is a broad-based risk assessment that is ongoing of your suppliers.
00:13:19.519 --> 00:13:23.679
The second is a data acquisition, let's call it PCFs or corporate carbon footprints.
00:13:23.759 --> 00:13:26.720
And a third is uh onboarding related diligence.
00:13:26.879 --> 00:13:29.600
And so you say, yes, great, let's let's come on board, let's do this.
00:13:29.679 --> 00:13:38.159
Now, the way that that AI native service provider might uh support you in this is they might have, let's say, the ongoing diligence is a scan that just runs.
00:13:38.320 --> 00:14:04.480
And every time that something triggers an exception, so every time there's a risk uh that gets triggered, you will receive, let's say, an email in your inbox, or the business will receive an email in its inbox, or it might even receive a message in
Slack or Teams or whatever the business uses saying, hey, there's an exception, there's an alert, the human rights like violation or risk has been spotted based on some press release, or an environmental risk has been spotted based on something.
00:14:04.720 --> 00:14:07.039
Want you to be aware, and so it is exception-based.
00:14:07.279 --> 00:14:15.440
That entire stream could be managed by an agent, which is again, let's say, AI connected up to tool access, connected up to context.
00:14:15.600 --> 00:14:30.159
Tools might include uh something where it can pull satellite-based uh data feeds, it can pull news feeds, it can pull other things uh together because it has tool access and it can deposit something in your inbox or your Slack or Teams because it has access to that as well.
00:14:30.320 --> 00:14:41.600
And it has enough context to manage for minimizing hallucinations, minimizing all the other AI sort of uh flaws, and really just giving you very focused outcomes and output.
00:14:41.759 --> 00:14:45.039
And that's like one outcome that is then fully uh agentic.
00:14:45.120 --> 00:14:51.919
Now, there might be a human in the loop there somewhere just to monitor for quality control and someone, but it is otherwise a fully uh agentic process.
00:14:52.080 --> 00:14:56.159
Now, on the second one where you have data acquisition, this gets a bit messier.
00:14:56.240 --> 00:15:13.279
Uh, maybe you still want to have on the solution provider side, there might still be a human engaging in a first call with a supplier, but all of the data verification, validation, et cetera, could be done by by agents as well, uh, which are then depositing the acquired data into another solution on your side.
00:15:13.360 --> 00:15:19.759
It could be wherever you're doing your corporate footprint, it could be an internal BI tool or Power BI or tableau or something that you're using.
00:15:19.919 --> 00:15:23.279
So it's a combination of humans doing some bits and AI doing others.
00:15:23.440 --> 00:15:38.879
And then on that third thing, which is the diligence, where you have a supplier being onboarded and you want to have an environmental diligence, let's say, done on that, or an environmental audit done on that supplier, you could again similarly see how certain bits of that could be done entirely by agents.
00:15:38.960 --> 00:15:41.039
Like the agents could be scrutinizing the data.
00:15:41.120 --> 00:15:46.879
You could have agents scrutinizing the output of other agents to make sure you have multiple lines of review.
00:15:47.039 --> 00:15:49.840
And then you might have a human, let's say, having the final look.
00:15:50.000 --> 00:15:58.720
And maybe if the supplier wants to have a conversation because they feel they'd be unfairly graded as part of that audit, then they want to speak to a human, and then you have a human coming in to have a conversation.
00:15:58.879 --> 00:16:05.120
And so in this way, you can see how you've got like a human at the center of an ecosystem of agents doing stuff.
00:16:05.360 --> 00:16:15.200
Now, I've talked a little bit at length about AI native services, and the reason I'm doing that is because I want to get down to implications for sustainability teams.
00:16:15.919 --> 00:16:25.440
I think that's like it's not even really an opinion, but what I'm seeing this year is I'm seeing lots and lots of hemorrhage within sustainability teams.
00:16:25.519 --> 00:16:28.799
And we've been talking about that for a while on this podcast, so it's nothing new.
00:16:28.960 --> 00:16:38.480
But I'm you know, we we've been sort of predicting, I would say for a couple of years now, that chief sustainability officers are not a really durable role in themselves as a C-suite role.
00:16:38.639 --> 00:16:39.279
We're seeing that.
00:16:39.440 --> 00:16:45.039
I've seen within my network already multiple chief sustainability officers be removed from their role.
00:16:45.200 --> 00:16:50.000
That role is likely being downgraded to a different level of hierarchy within the organization.
00:16:50.399 --> 00:16:54.559
And most sustainability teams and sustainability functions are also taking a hit.
00:16:54.799 --> 00:17:00.320
Now, there are a few reasons why that is likely going to accelerate and continue.
00:17:00.639 --> 00:17:03.840
And I'm gonna bring this back to AI and AI native services.
00:17:04.160 --> 00:17:11.440
Most businesses right now are under some form of pressure to think about more utilization of AI.
00:17:11.920 --> 00:17:21.440
Now, early in this year, there was this idea of like token maxing, particularly within software companies, which was how do you maximize the use of AI?
00:17:21.680 --> 00:17:32.319
You measure token usage, the number of tokens that are being used, and you want to encourage the business to use as many tokens as possible because that means the business is experimenting and really pushing the envelope.
00:17:32.559 --> 00:17:37.759
I think that there's a bit of a step back from that because it was really expensive and not clearly correlated with results.
00:17:38.000 --> 00:17:41.119
I think there's a lot of sophistication now emerging.
00:17:41.279 --> 00:17:59.359
I think it was a Harvard Business Review uh analysis from a few months ago that said that when a manager believes that an AI tool that they're using has personality and is anthropomorphosized, they are less likely to review their outcomes for quality than if they think of the AI as a tool.
00:17:59.440 --> 00:18:04.640
And so there's all these nuances now around how organizations are gonna be using AI and getting more value from it.
00:18:04.799 --> 00:18:10.240
But the bottom line is most organizations are being pushed to use AI more and to push the envelope.
00:18:10.480 --> 00:18:13.519
And the envelope basically means labor savings.
00:18:13.839 --> 00:18:25.200
This is really the little spoken, you know, open secret uh of the AI world, which is any value that AI is going
to create is almost certainly going to come from loss of labor.
00:18:25.440 --> 00:18:34.799
It is not that AI is, I mean, AI may create a lot of net new value, but we're basically saying we have invented as a society artificial intelligence that can do stuff.
00:18:35.039 --> 00:18:42.079
And the only thing that you could see that substituting for is human intelligence because that's the only other form of intelligence in the workplace.
00:18:42.400 --> 00:18:44.880
So, what is that labor saving going to look like?
00:18:44.960 --> 00:18:51.039
It's going to start with things that are predictable, repetitive, not cool.
00:18:51.279 --> 00:18:54.880
And unfortunately, sustainability right now is not considered very cool.
00:18:55.039 --> 00:19:00.480
Like you have this sort of years, it's against the Zeitgeist in the same way that it was with the Zeitgeist a few years ago.
00:19:00.880 --> 00:19:02.559
Much of the work is repetitive.
00:19:02.720 --> 00:19:04.640
It happens every year, like clockwork.
00:19:04.799 --> 00:19:06.480
It should happen in a certain cycle.
00:19:06.640 --> 00:19:14.559
And actually, it is kind of proper white-collar work in that it requires a certain amount of subject matter knowledge, it requires some intelligence.
00:19:14.880 --> 00:19:20.400
But those are exactly the types of jobs that AI is well geared to tackle.
00:19:20.559 --> 00:19:28.880
And so I would say that sustainability is uh going to very soon be in the headlights for automation with agentic AI.
00:19:29.039 --> 00:19:33.599
And I would say AI Native Services is the competition for internal agentic AI.
00:19:33.759 --> 00:19:37.359
A business can either set up its own agents to do these things themselves.
00:19:37.440 --> 00:19:46.960
So let's say to uh you know, clean and scrub data before it gets used in a corporate footprint, to run the corporate footprint calculation, maybe in certain cases.
00:19:47.119 --> 00:19:50.240
Uh, you could see more complex uh uses as well.
00:19:50.400 --> 00:19:53.759
And so if the business is ready to build out their own agents, they might do that.
00:19:53.839 --> 00:20:00.960
If they're not ready to build out their own agents, they might work with a solution provider that is able to provide this hybrid blend of AI native services.
00:20:01.119 --> 00:20:05.599
And because that is using a lot of automation on the solution provider side, it will still be cheaper.
00:20:05.839 --> 00:20:17.279
I would say the way that sustainability teams could think about this is that I suspect most businesses would be, would be happy with 80% of the value for 60% of the cost.
00:20:17.599 --> 00:20:23.039
And that could be achieved through internal agentification or through AI native services.
00:20:23.279 --> 00:20:41.119
So now if you say, okay, that is what is coming for me, I hate to be in this position where I'm trying to defend sustainability teams against their organizations or sort of advise them on how to immunize themselves against the onslaught of AI, because I want to be also on the side of the corporation.
00:20:41.279 --> 00:20:50.240
But I think that actually there's a way that both those streams can come together a little, where if you are a chief sustainability officer, you need to be thinking about leverage.
00:20:50.400 --> 00:20:53.519
If you're a sustainability manager, you need to be thinking about leverage.
00:20:53.599 --> 00:20:56.640
If you're a sustainability analyst, you need to be thinking about leverage.
00:20:56.880 --> 00:20:59.920
What I mean by that is it plays out a little differently at each level.
00:21:00.000 --> 00:21:02.720
So let's start at the highest level, the chief sustainability officer.
00:21:02.880 --> 00:21:10.400
If you're a chief sustainability officer, I don't think that the right answer is that you are a one-person show and everything else is done by agents.
00:21:10.559 --> 00:21:29.200
You may well find that you'd actually need certain types of people for creative thinking, for system design, for stakeholder interactions, for a bunch of stuff that actually AI doesn't really do well now and is unlikely to do really very well for at least the next few years.
00:21:29.440 --> 00:21:35.440
However, those types of people may actually not be the types of people who have been valuable to you two years ago.
00:21:35.759 --> 00:21:42.240
Two or three years ago, the most valuable people to a chief sustainability officer were often the deep subject matter experts.
00:21:42.400 --> 00:21:47.440
Because the chief sustainability officer was in some cases themselves a good subject matter expert.
00:21:47.680 --> 00:21:53.200
It is very rare that a chief sustainability officer in any large organization is a deep subject matter expert.
00:21:53.279 --> 00:21:54.720
You just have too many other things to do.
00:21:54.880 --> 00:22:08.079
Now, the deep subject matter expertise, in a like Yoda like way, where it's Someone you want to say, hey, what are the arcane rules of flag accounting per the latest hypothesis on where the greenhouse gas protocol will land?
00:22:08.319 --> 00:22:20.799
Like that is, you know, that was really useful two, three years ago to have that person who could just like give you an answer to a question like this and know that and and and help you make decisions on on the basis of that sort of expertise.
00:22:20.880 --> 00:22:28.400
Uh, but that's exactly the kind of thing that AI is going to make quite easy, and in many cases is already made quite easy and will progressively make quite easy.
00:22:28.480 --> 00:22:33.200
Now you could say, well, my subject matter expert is deeper, better, badder than anyone else's.
00:22:33.279 --> 00:22:38.480
And then, you know, AI has been trained on the average, and therefore my person is just so much better.
00:22:38.720 --> 00:22:41.839
That's a really diminishing little island, basically.
00:22:42.000 --> 00:22:43.200
That island is sinking.
00:22:43.359 -->
00:22:57.599
That difference between your SME, your subject matter expert, and what AI can give you is, you know, it's kind of like that survey they did like of Scandinavian drivers, where 95% or something of drivers thought they were above average, right?
00:22:57.680 --> 00:22:59.920
I forget the exact numbers, but it's a little like that.
00:23:00.079 --> 00:23:06.559
AI is in most cases going to be better at giving this sort of advice or opinion than whoever you have in your team.
00:23:06.799 --> 00:23:11.119
But actually, like these other areas where you think about system design, how do all these things play together?
00:23:11.279 --> 00:23:12.000
How do they play well?
00:23:12.160 --> 00:23:19.119
Where do I have a system of records storing stuff that is a database where my AI can actually pull from and use?
00:23:19.359 --> 00:23:20.960
What does the context need to look like?
00:23:21.119 --> 00:23:23.519
Who should kind of create and maintain that context?
00:23:23.680 --> 00:23:27.839
Uh, what are the different flows of how agents plug into my processes today?
00:23:28.000 --> 00:23:37.920
Like some of this is an IT thing, but a lot of this will require an understanding of what the outcomes should be, what good looks like, and how you get all these things to play together.
00:23:38.160 --> 00:23:46.000
And so there will be certain roles that you really need as a chief sustainability officer, but they will be likely different to the roles that you needed two or three years ago.
00:23:46.160 --> 00:23:48.640
Let's move a level down to a sustainability manager.
00:23:48.799 --> 00:23:55.839
Now, my opinion, my hypothesis is that we're going to see a lot of flattening of that manager and analyst layer.
00:24:54.429 --> 00:24:56.509
And I think that that's true for two reasons.
00:24:56.669 --> 00:25:02.429
Both of which have to do with how organizational hierarchies have developed over time in corporations.
00:25:02.750 --> 00:25:10.269
From my perspective, the value of an organizational hierarchy tends to be around managing three things.
00:25:10.509 --> 00:25:15.069
Um, my team will say there's always three things because it's an old consultant trope.
00:25:15.309 --> 00:25:19.149
But one of those things is how do you manage the flow of information?
00:25:19.309 --> 00:25:22.909
Uh, typically, let's say upwards, but often sometimes downwards.
00:25:23.069 --> 00:25:29.389
How do I convey information upwards into the organization so that the senior echelons of my organization know what's going on?
00:25:29.549 --> 00:25:38.029
I need to have gatekeepers of that information, otherwise, my poor CEO is going to get overwhelmed with too much data and won't be able to make sense of things.
00:25:38.269 --> 00:25:52.990
How do I manage the flow of decision making downwards so that actually, again, the CEO and the upper echelons are able to give certain decisions, and those decisions can be packaged in certain ways that different nodes within
the organization can now go ahead and implement them?
00:25:53.230 --> 00:26:07.709
And the third is performance management, where now, given I have to have all these nodes for information flow and decision making, how do I also performance manage and coach and mentor and apprentice the different nodes so that everyone can develop into a better version of themselves?
00:26:07.869 --> 00:26:18.669
Now, the reason that I say that that sustainability manager and sustainability analyst are likely to be collapsed is because agenc AI and AI in general, I think, will start to make a lot of these things a lot easier.
00:26:18.829 --> 00:26:28.509
It becomes easier to contextualize and structure information flowing upwards if intelligence becomes super cheap.
00:26:28.669 --> 00:26:37.789
Because actually you have all this artificial intelligence that can now look through all the data and process it and parse it and tell the CEO and tell the oper echelons this is what you actually need to know.
00:26:37.869 --> 00:26:38.990
This is the signal that's important.
00:26:39.069 --> 00:26:40.429
You don't need all those gatekeepers.
00:26:40.589 --> 00:26:53.469
Similarly, with decision making, it becomes much easier to take a high-level objective or goal and actually have it packaged down into what the implications are for all the other nodes in the organization using AI and agents.
00:26:53.709 --> 00:26:57.069
Now, on the performance management piece, that's actually derivative of the other two.
00:26:57.149 --> 00:27:01.309
If you don't need the other two, do you really need all these people to be coached and mentored and apprenticed?
00:27:01.629 --> 00:27:02.349
Probably not.
00:27:02.509 --> 00:27:06.909
So I think that the manager and analyst layer will collapse a little on themselves.
00:27:07.069 --> 00:27:13.069
And I think that there's an important and different learning for both, both the sustainability manager and the sustainability analyst.
00:27:13.230 --> 00:27:20.269
I think for the sustainability manager, you have to appreciate that your progression is no longer going to be linked to the size of team that you manage.
00:27:20.429 --> 00:27:25.549
Because the organizational momentum is going to be towards collapsing these teams and having fewer and fewer people.
00:27:25.709 --> 00:27:34.990
So as a manager, I would say you need to start taking the lead and thinking through, okay, actually, do I want to be part of my organization building out all its own agents?
00:27:35.149 --> 00:27:40.990
Or do I want to be part of my organization leveraging external solutions that can drive down cost in a meaningful way?
00:27:41.230 --> 00:27:49.629
And if my organization is going to be happy with 80% of the value, where do I shave off the bits of the value that actually I don't need perfect?
00:27:49.949 --> 00:27:57.309
I have firsthand exposure to multiple organizations that I would say have tremendously powerful sustainability teams.
00:27:57.469 -->
00:28:09.629
By sustainability teams, I mean sustainability teams that do their job really well, that have a super high bar for quality, and that are recognized within their industry and within other sustainability teams as being real trailblazers.
00:28:09.789 --> 00:28:16.990
And I know that many of the individuals within those teams are actually going to be looking for jobs over the next weeks.
00:28:17.230 --> 00:28:22.829
The simple reason is that that level of bar is no longer what's most important in these organizations.
00:28:23.149 --> 00:28:27.869
It is not the 100% value for 100% of the cost.
00:28:27.949 --> 00:28:30.829
It is the 80% of value for 60% of the cost.
00:28:30.990 --> 00:28:35.869
And I think that's what sustainability teams, at least for the next few years, need to get used to as a concept.
00:28:36.029 --> 00:28:39.869
And so as a manager, you need to start thinking about okay, how do I get leverage?
00:28:40.109 --> 00:28:41.709
What does that 80% mean?
00:28:41.789 --> 00:28:48.829
And how do I get cost down to 60% for delivering my bit, my set of outcomes that I'm responsible for as a manager?
00:28:49.149 --> 00:28:59.389
As a sustainability analyst, I think you need to start thinking about the same thing, but almost in the inverse, which is how do I make it so that actually a manager is not needed over me?
00:28:59.629 --> 00:29:05.469
If that layer is going to collapse, how do I make sure that I am the one that actually can win out?
00:29:05.629 --> 00:29:08.109
I don't mean to pit managers against analysts, right?
00:29:08.189 --> 00:29:24.909
But I'm just saying that if there is one seat to fill and three people, a manager and two juniors, let's say, who are right now in the organization, I want to give each of those three people whatever insights you know I can to get that seat.
00:29:24.990 --> 00:29:29.230
And I would say for the analyst, it needs to be how do you make sure that you don't need the manager anymore?
00:29:29.389 --> 00:29:36.829
How do you make sure that actually whatever the level above will need from your unit can be provided in as simple, as easy a way as possible?
00:29:36.990 --> 00:29:45.869
That that data is available in a really easy way for making decisions, uh, that actually you don't need many more people to do many more things, you don't need software tools.
00:29:46.029 --> 00:29:52.109
I realize I'm I'm I'm uh cannibalizing my own business here, but you, you know, I it is an area for you to look at.
00:29:52.269 --> 00:29:53.149
Do you need software?
00:29:53.309 --> 00:29:54.189
Do you need people?
00:29:54.349 --> 00:30:05.069
Do you need more resource where you can actually get stuff done using your own agents or again using solution providers who can club together the value that you might be getting from software and people today?
00:30:05.230 --> 00:30:08.669
And so I think that as an analyst, you want to think about how do you collapse that layer in your favor.
00:30:08.750
--> 00:30:11.469
And as a manager, you need to think about again how do you consolidate.
00:30:11.629 --> 00:30:20.429
But both manager and analyst need to figure out how they deliver the composite of outcomes that the organization needs with fewer people and less resource.
00:30:20.669 --> 00:30:31.069
Maybe just lastly, right, I think just a note on what it feels like to be in the Zeitgeist or with the Zeitgeist and then against the Zeitgeist.
00:30:31.309 --> 00:30:41.149
I've shared this example on an earlier podcast at some point, or the earlier episode at some point, but I used to be way back when a consultant to the energy industry with McKinsey.
00:30:41.309 --> 00:30:49.149
And I was part of a team running a very long-run cost optimization exercise at a big utility business.
00:30:49.309 --> 00:30:53.549
And the utility business, I think, started out with 12,000 employees.
00:30:53.709 --> 00:30:58.269
And by the time I got there, which was like two years into the transformation, they were down to 7,000.
00:30:58.429 --> 00:31:03.949
So uh McKinsey had already slashed, help helped the organization slash, you know, a big chunk.
00:31:04.029 --> 00:31:05.549
And we were down to now 7,000.
00:31:05.709 --> 00:31:12.349
Now, within that 7,000, one of my roles was to figure out, okay, where do the next, where do the next cuts come from?
00:31:12.509 --> 00:31:16.269
Uh, this is where I started racking up all those sins I need to work off now.
00:31:16.509 --> 00:31:18.909
And I was kind of looking at all of these different functions.
00:31:19.149 --> 00:31:22.269
You had like sales and marketing and finance, et cetera.
00:31:22.429 --> 00:31:28.669
And I found a pool of 1,200 people in an organization called the Change Org.
00:31:29.149 --> 00:31:33.309
And everyone I asked about, well, you know, what does the change org do?
00:31:33.549 --> 00:31:36.429
No one could tell me really what the change org does.
00:31:36.669 --> 00:31:39.629
And this this you know, this is like 10 years ago or something.
00:31:39.789 --> 00:31:47.549
Uh, and I've learned this lesson actually since then, which is when someone can't tell you what a team does, that is a problem.
00:31:47.789 --> 00:31:57.629
Like when someone when when others in the organization don't know what an entire function, that is like a fifth of the organization, does and is responsible for, really there is something wrong.
00:31:57.789 --> 00:31:59.230
And so in this case, no one knew.
00:31:59.469 --> 00:32:14.429
Eventually, after many conversations, what I realized was that the change org was a team of translators, a 1200-person team to translate what the IT function needed with respect to the rest of the business.
00:32:14.509 --> 00:32:18.429
So there was a thousand-person IT function, thousand, fifteen hundred people in IT.
00:32:18.589 --> 00:32:23.230
There was 1200 in this change function, and then there was all these other people in the other functions.
00:32:23.389 --> 00:32:32.909
And because no one knew how to speak to IT people, and because IT people didn't know how to speak to everyone else, you had to have this function in between, the change function to bridge the gap.
00:32:33.149 --> 00:32:38.750
And at some point, I'm sure that the idea of having an IT translator made a lot of sense.
00:32:38.909 --> 00:32:41.069
And I know that because I've seen the pitch decks.
00:32:41.309 --> 00:32:49.949
Uh, having translators to, you know, IT translators eventually later even became people who can work with data scientists, the translator for the data scientists.
00:32:50.029 --> 00:32:57.309
Like corporate speak has evolved this translator role, and every five years thinks about how translators are suddenly needed within the organization.
00:32:57.389 --> 00:33:03.230
Uh, but by the time I came to this place, translators were so out of fashion that they were just a target for cost-cutting.
00:33:03.389 --> 00:33:06.189
No one even remembered why they had 1200 translators.
00:33:06.349 --> 00:33:08.269
But all this was needed at some point.
00:33:08.349 --> 00:33:12.589
It was considered valuable and cool, basically, for CEOs.
00:33:12.750 --> 00:33:20.750
And I'm not saying it's that bad for sustainability, but I would just be wary of being a translator in a world that no longer knows what a translator does.
00:33:20.990 --> 00:33:25.869
Uh so with that, I hope you enjoyed this episode of the State of Sustainability Podcast.
00:33:26.509 --> 00:33:31.469
As always, if you did like the show, please hit follow so you never miss another episode.
00:33:31.629 --> 00:33:34.429
Rate us, review us, and share us with a friend.
00:33:34.589 --> 00:33:37.069
All these steps very much appreciated.