Ep 132: Untangling AI Readiness and Prioritizing Workforce Data with Olivier Vidal

[00:00:09] Hey everyone, it's Bob. Welcome back to Elevate Your AIQ, your go-to source for insightful conversations on human-centric AI readiness, talent transformation, responsible innovation, and the future of work. Today I'm joined by Olivier Vidal, who's the founder of a new AI readiness practice for workforce products called Sightline. That's at sightline-ai.co. He's a long-time technology product leader at organizations like WPP, ADECO, and General Assembly.

[00:00:36] Olivier and I dig into why so many enterprise ambitions around AI are outpacing the underlying data and organizational readiness needed to actually deliver on them. We talk about what it takes to close the gap between big promises and the messy reality of feeding these systems well, remember garbage in garbage out as far as your data. We also get into how human and AI collaboration is reshaping decision making, why trust in your data matters more than the technology itself.

[00:01:03] So stick around for a candid, wide-ranging conversation and thanks as always for listening to the show. Hey everyone, welcome back to another episode of Elevate Your AIQ. I am your host Bob Pulver and I am looking forward to this overdue conversation with Olivier Vidal. How are you today, Olivier? I'm very good. I am baking like a rotisserie chicken here in London.

[00:01:27] I think it's 100 degrees and my little office has got a flat roof and yeah, it's too much. But hopefully the weather's broken by the time this goes out. Yeah, I mean that's crazy. I thought New York was hot so that's pretty bad. So stay hydrated. Olivier, we've got a whole lot of things to talk about and as we've spoken over the last year, year and a half, I think some of the topics are consistent and some of them have evolved.

[00:01:56] Certainly your career and some of the projects that you're working on have evolved and I want to dig into all of that. But why don't we start with just you giving my listeners a little bit about your backstory and some of the work that you've been doing across WPP and GA and Chapter 2, lots of things. Sure. So if I was at a dinner party, I guess I'd say to someone that I'm a product manager. But equally, those of us who've been around a while, we understand there are many, many flavors of product manager.

[00:02:26] And I guess for the last 10, 15 years or so, I've been involved in building, breaking, rescuing, HR tech tooling. Often as that interpreter between the hopes and dreams of senior management and the nitty gritty dull reality of what you can do with a given pool of software engineers, data scientists, whatever it might be at your disposal. I, what's to say about my background?

[00:02:55] So I've, I've been in HR tech now for, for, I guess the best part of nearly 20 years. Well, 12, 15 in, in, in HR tech proper and, and eight years before that in recruiting and exec search. I ran a startup in my late twenties, which taught me an awful lot about trying to use technology to fix problems.

[00:03:18] And, and I slowly have then got involved in running technical teams, data science teams, decision engineers, all of whom have been focused on how do you build tools that basically match people to work at different velocities.

[00:03:37] So whether that is talent marketplaces, whether that is resourcing solutions, and maybe as we drift into the agentic era, will we need faster and faster versions of that same category of technology to associate non human resources with, with, with work. But that's kind of always been my shtuck.

[00:03:58] And as you alluded to, I've really been primarily been either in small startups or I've been in the technology teams of services companies, ADECO, WPP, where they were trying to build in-house tooling to associate people with work in, in different ways and forms. I mean, you and I have both spent some time, you know, at big companies and, and small companies.

[00:04:26] I mean, in terms of, in terms of, in terms of the current technology, I mean, how do you think, I guess, which group is better prepared when it comes to, I know we're going to get into this topic of readiness, AI readiness, data readiness. But I guess, you know, what's, is there any commonality between the big and the small or is everybody just?

[00:04:50] I think the commonality in my experience and it's frame everything by what was missing, but essentially the, in an awful lot of the projects I've been involved in, the hopes and dreams of senior management or the hopes and dreams of product builders and designers have been miles ahead of what the underlying, whether we talk about data maturity, knowledge maturity, rules, whatever it might be.

[00:05:16] But essentially miles ahead of the actual preparedness of a business to feed a given system with the information it needs to make decisions.

[00:05:27] So, you know, at Adadeco, you know, there were, there were kind of big, highfalutin kind of Fortune 500 style, like dreams about, you know, what, what, what we were going to do to the market and how we were going to use data science and AI to achieve that.

[00:05:49] But there was a big disconnect between those ambitions and the base layer preparedness of the business to, to, to make that happen. Not in the technical building, but in the technical feeding, if you like, or the, the, the ingestion. And that, that's kind of always been the theme. I'm, I'm sorry to say, and I think it's probably common across HR tech, right? I don't think this is like weird to me.

[00:06:16] This is an industry wide problem that we either need to kind of work out solutions to. Part of it is agreeing on language or we're just going to slightly leave our head in the, in the sand ostrich like, and just sort of say, well, we've built some amazing tools. Do they work? Oh, well, they work in abstract, but they don't, they don't really work in reality. I hear what you're saying. And it does seem like, you know, technology has evolved, products have evolved, got more capabilities.

[00:06:47] And yet some of the fundamentals about how we do that, how we find that sort of win-win scenario, right? Matching the right people to the right jobs and the right context and, you know, what have you. We're still having, it's a very complex problem that is still not yet solved. So I think some of this goes back to, I mean, what your conversations have revealed, which is, you know, you still have these fundamental data problem that we can't trust. Well, trust is paramount, right?

[00:07:17] So we've got to trust each other. We've got to trust, you know, the other members of our ecosystems. And we've got to trust the data because we already know AI is held to a higher standard in terms of accuracy and, you know, bias and things like that than human, you know, judgment and bias, right? So how do we get past that without actually digging in? And I know you've been, you know, working with others.

[00:07:44] How do we sort of mature that space and get people to sort of rethink this, whether you're talking about skills or you're talking about, you know, HRIS, you know, core systems of record, you know, data, performance data. I mean, there's so much out there that everyone's trying to mix and match and coalesce. A hundred percent. And yet they're not coming up with, you know, the silver bullet.

[00:08:05] No, and I think I won't use names because it's not appropriate, but, you know, I was talking to a senior product person at one of the big, one of the big industry leading suppliers.

[00:08:17] And they were kind of trying to both encourage me, but also to use a preferred argument of theirs that essentially the data set about work and people is in the next five years going to massively move up the pecking order in terms of its value as a data set inside a classic enterprise.

[00:08:43] So historically, financial data was given a massive premium inside an organization. Same around kind of sales and acquisition and the go-to-market pipeline that had huge kind of scrutiny and value and people working on those data sets were, had some level of status inside an organization.

[00:09:05] Whilst historically, things like people analytics or talent intelligence were specialist subunits of the CPO's world and they didn't. The data sets that they were working on were primarily there to inform relatively slow-paced decision making and might be trusted or might not.

[00:09:32] And it didn't, to a degree, it didn't terribly matter because as a data set, it wasn't terribly trustworthy. So the people analytics people were just doing their best to make sense of whatever they could access. But the data set itself was not a refined and valued asset of the business, right? I think we're just generally moving into a phase where maybe that will be the case.

[00:10:02] And I think the tipping point is probably one of two things. It's either that companies want to get some fluidity in how they engage with their internal staff alongside external providers.

[00:10:18] So whether that be contractors or people who are in third-party kind of consulting organizations, but effectively, if a management team wants to take that fluidity between our people and other people seriously, then that data set becomes jolly important. And then equally, if they want to start to use agents to take on tasks or be part of a pre-designed value chain, then again, that's a potential tipping point.

[00:10:47] But without either of those two impetuses, then probably this data set will remain kind of a poor cousin. The poor cousin of this, of others. Yeah, I mean, it just seems like you've got a common, in a way, a common denominator across some of these big, potentially multi-year initiatives, right? Job architectures and skills taxonomies and all of this stuff.

[00:11:16] And so I don't know that there's one team that can necessarily solve it all, but I hear what you're saying. Like, you bring in the expertise, you coalesce, you know, you bring in as much data as you can. You figure out, you know, how to sort of have the, I guess, the scope of the product and some of your deliverables, you know, address those. And yet, ultimately, I guess, harkening back to some of the work you've done on decision engines.

[00:11:46] Ultimately, when it comes to, you know, feeding a decision support system with both, you know, human and AI, you know, contributing to that, you still have a foundational sort of flaw. Yeah. Yeah. And the human and AI teams, I'm also thinking about at the, not within the HR world and not within the people analytics world, but it's in the front line of how a business generates value, right?

[00:12:12] So if you're a, I don't know, a pharmaceutical development business and you, you know, you previously had a people analytics team that probably did, made the best of the little bit of information they could harness on those, on people. And they used that for, but resource management would probably exist in the operations world.

[00:12:33] But basically, in the traditional version of a business, the people data was used to help describe what's going on. Whilst as soon as you've got agents, say, within that drug development process, you need to be able to orchestrate human and agent to some degree within that value chain. And it has to be a designed experience.

[00:12:56] And suddenly the fidelity of information you need on people and what they can do is exponentially higher than it was when all you were trying to do was put together some plausible feeling reports on what are our people doing and what can they do? And I think we'll kind of wrap up this topic because there's some other things I want to make sure we talk about. But I do think, you know, it's not just about, I want to make sure people realize it's not just about data, you know, quality.

[00:13:26] I mean, this is, you know, is the data, you know, trustworthy? Do I know that it was, you know, what do I know about the data, right? And how does that inform the level of trust that I have once it's now commingled with other data? So it's data, you know, consistency, it's quality, it's provenance, it's veracity. There's all these elements that feed in to this.

[00:13:53] And so, unfortunately, that can manifest, if that's not addressed, that manifests in the lack of trust in whatever the numbers say. And then you get back to leaders trusting their gut because they feel like they can't trust the data. All of that is true. And at the same time, in most enterprises, the number of tools that hold this kind of category of information is much higher than would be ideal if you would design it from the ground up.

[00:14:21] And the HRIS often only holds pretty rigid structured amounts of information that doesn't really tell you anything that you can use at a tactical level about, you know, essentially who's going to do what or could do what or is ready to do which given task or project or whatever unit of work you're interested in.

[00:14:44] And additionally, again, there is, you know, as we move into this world of interviews being recorded and so much more long form unstructured information being generated on people and what they do and how they use tools and so on and so forth.

[00:15:01] All of that is a sort of a whole secondary sort of source of potential information about people that the classical database driven HRISs and so on have to respond to work out how it's going to be, how it's going to be used.

[00:15:21] Yeah, I mean, I think about this in the context of, you know, the work you've been doing and I know the conversations that I've been privy to in the talent intelligence collective, which spans a lot of different experts across some of the areas that you mentioned, talent intelligence, people analytics, strategic workforce planning, all of these elements.

[00:15:44] And we think about, you know, we think about, you know, we think about, you know, the labor market intelligence and dipping further into talent intelligence in terms of the labor market, in terms of extended workforces. We've got a growing as expected gig economy.

[00:15:58] So as you look at a comprehensive view about the available talent plus available or to be built, you know, agents and authentic capabilities, it gets, the puzzle gets more and more complex. So if we don't fix that foundation, you know, things could easily get messier before they ever get cleaner, I suppose.

[00:16:25] Yeah, but I think the good news is there are some amazingly foresighted product building teams out there in the market that are absolutely trying to, you know, trying to create AI enabled software that will give companies a fighting chance of managing these much more complicated fluid workforces. Like, I don't think it's on enterprises, enterprise teams entirely to kind of work all this out for themselves.

[00:16:54] What I think is shifting is that the, I have no idea where the balance comes in, but the kind of old school, well, we'll buy a tool, we'll configure it, we'll plug it in, and then the job is done. Is anybody who's kind of got that attitude towards HR tools in the next 10 years, I think it's going to be sorely disappointed.

[00:17:17] It's because ultimately, even the very best in class of these new tools are a beautiful looking user interface. Great. God knows we need that. We, then there's systems of intelligence and computation and so on that are being used to make decisions and inferences and so on.

[00:17:39] But ultimately, they're still just got to be fed by a sound foundation of whether we call it data or knowledge or information or context. And I think that's where a lot of the challenge sits.

[00:17:54] And ultimately, the thing that I wouldn't say to kind of the SaaS vendors necessarily is if you follow the logic through to its sort of maturity, ultimately, the company's own data set is the product.

[00:18:13] That is its ability to have an operating system that lets them combine their people and resources to deliver value to their own customers.

[00:18:25] And the products that they purchase to provide a interpretation and a well-designed front end that they layer on top of that information is almost secondary.

[00:18:45] And they should be able to move between tools in the future that are all basically making sense of the database, the data set that the company itself cares for and curates and values. Because if they don't value it, they're just going to end up with loads and loads of different disaggregated systems. It's going to be really, really painful.

[00:19:11] Yeah, I mean, I certainly did not want to give the impression that, you know, the goal is perfection as that is a fool's errand. But I think we can all agree that, you know, some of the metrics that we care about and metrics that truly matter are not necessarily going in the right direction.

[00:19:31] So I think you need some target goals to improve, to reverse some of those trends or just incrementally improve some of the metrics that we're seeing, whether that's, you know, quality of hire or, you know, reverse the shrinking tenures trend, improve morale and engagement.

[00:19:53] I mean, there's lots of things that we can do to sort of nibble at some of these challenges that are just foreign and many, many leaders side.

[00:20:04] Yeah, and I think what I've been with the practice I'm putting together, I've been trying to work backwards from the use case or the problem at hand like you describe and try and see how you would remedy that within the source data, the source information that feeds those decisions. Because saying, well, we need to sort our data out, it is too broad, too general for anybody to really care about, right? It's just too frightening.

[00:20:33] But if you look at it on a product basis and a use case basis, then suddenly it kind of becomes manageable, even if it's still a little intimidating.

[00:20:43] Olivia, I wanted to get your product expert perspective on this concept of evaluations of AI solutions and how different it is from the traditional software testing cycles, right?

[00:21:06] I've spoken to others in the past and I've got some personal experience with this going through different testing phases, you know, user acceptance testing and, you know, technical verification testing. There's, you know, five or six different sort of phases, but everything is very sort of structured, right?

[00:21:25] It's like there's only so many, you know, drop downs and choices and radio buttons and things for a user to do as they interact with your product with a traditional software or even a SaaS solution oftentimes. So your testing can be, it can be vast, but it can still be sort of a finite sort of, you know, you know, frame around, you know, what we're testing and those are the things that could ever possibly break.

[00:21:53] And with AI solutions, that's not the case. If you have, you know, any user interacting in natural language, you have absolutely no idea what someone may say or work around as they may try to get the AI to do. And there's some, there's just a lot of nuance to and variability into what you can expect. So how do you think about that as someone who works in product on the AI side these days?

[00:22:23] I would, I'd be cautious here just about me and my level of expertise and what I'm good at and not so good at. So the, I think technical eval evaluation of different models is a discipline in and of its own right that is, that is maturing. And a product professional has responsibility to be asking the right questions and bringing in the right experts at the right time.

[00:22:48] There are folk like Warden AI who are doing exemplary work alongside the vendors to stress test and look at the potential biases within decision systems and AI tooling. So what I would, it's a discipline within its own right, which I am not claiming.

[00:23:14] I'm not, you know, I'm just, I'd be super cautious there about trying to claim any personal expertise there. I would say it is quite separate from the business cases and the strategic value of a product. It is a, it's a, yeah, as you say, it's part of testing and it's a discipline within its own right, not to be, not to be sort of undertaken by amateurs, basically.

[00:23:40] Because as you say, how do you, tactics for breaking or for forcing answers out of particularly chat-based front end user interfaces is a whole discipline within its own, within its own right. Yeah, I, uh, I've worked for myself since, you know, chat-GBT came out.

[00:24:01] So I've been only seeing this from, you know, the courses that I've taken and some of the personal, you know, experimentation, building agents and, and routines and workflows and things like that. And it seems like in addition to sort of fixing a lot of the slop that gets generated, I'm also at least a couple times a week also finding some gap in the way that the workflow is running.

[00:24:29] It could be simple, like, you know, it was supposed to wait for this other routine to finish before it moved on. And now it's carrying forth either stale data or inaccurate data or both. And so I'm constantly sort of fixing those things. So I, I have an appreciation for exactly what you're saying. Like it's a whole sort of discipline unto itself.

[00:24:50] How do you, I imagine that AI would help you flush all those things out and thoroughly do the, the evaluations. But sometimes it is, it takes human, you know, judgment and what people are calling taste to actually say, is this the output that I, that I want, that I expect and that users are going to expect if this was something that I'm building for others.

[00:25:14] Yeah. And I think the example you've given there, you are super experienced person who's like, like yourself when you, when you were in inside corporate role, the, I'm extremely cautious about any HR operations professionals, basically vibe coding something that's actually going to be used within any serious corporate use case.

[00:25:46] Yeah.

[00:26:14] Internal process that you happen to have decided, oh, you know, we could build an AI for this because you can't possibly start to guess all of the ways in which it could be broken by a skilled set of prompts. And yes, I'm really leery on the way. And yes, I'm really leery on the whole kind of idea of like, oh, HR can code now.

[00:26:33] Like I'm much more in the camp of let's try to use the tools that we procure to the very best of their ability by, by understanding what's going on under the hood to sufficient degrees and understanding what they need to be fed to sufficient degrees. But I'm, I'm really cautious about vibe coding anything, frankly, that touches sensitive data. It's a different club, a different mindset. I'm not in it. Yeah.

[00:27:02] Well, well, well, this leads to the next thing I wanted to ask you about, which was around, you know, AI, AI literacy, AI fluency, AI maturity, AI readiness. I mean, there's a lot of terminology being thrown around. I don't know that these are all synonymous, but I do know that, yeah, I do know that the word responsible should be prefixed to all of them.

[00:27:28] Because if you are not being responsible by design, if you are not, if you don't have the ability to, you know, think critically about an AI's output, questioning what data went into this in the first place and thinking about if you are building products, you know, vibe coded or not. If you're not, if you're not, if you're not thinking about being responsible by design, you're going to run into, you're going to inject unnecessary risk into the environment.

[00:27:57] And we know in the HR, HR tech space, there's a lot of risks, whether you're in the EU, obviously, or in other places where there's a lot of scrutiny. There's, these are high risk use cases and you can't just, you know, whip something up and think it's, you know, sort of enterprise grade and up to snuff for, you know, the authorities. And let's be real, people are under pressure.

[00:28:26] I was under pressure, you know, you dropped, not so long ago to just, you know, Hey, just build something, show that you can do it with AI. That, that kind of general, that general kind of slight executive panic of just like, show we can use it. Like that is real people. People are being lent upon to show that they can just knock something up and make everybody say, wow.

[00:28:52] And if anybody ever listens to me, I would just, like, just be really, really cautious. Not least because all of the tech debt and all of the kind of long tail repercussions of just whipping something up and then giving access to real data is, like, it's dangerous stuff. Yeah. And I had someone, I just had a prep call with a future guest and he reminded me that, you know, everything that I'm building for myself is, is really cool.

[00:29:20] But if I was in an enterprise, I wouldn't, I wouldn't necessarily have access to those tools, let alone the agency to actually build that. Because I'd have to go through so many checks and controls and check with cyber, check with data privacy, check with, you know, whatever.

[00:29:38] And, and of course on the CIO side, you know, they want to sort of, you know, limit the attack surface and mitigate the risk, you know, between the CIO and the CISO, of course. But, but yeah, I mean, these are all things where, look, I don't want to, I don't want to stop anyone who's, who's curious and ready to learn and, and has this sort of adaptive, you know, mindset, which is, is part of this AI readiness.

[00:30:05] I don't want them to, I don't want to put blockers up and say, you know, we're not going to do this. I want to give them the freedom to experiment and learn. But if you start with, just see if this works with AI, I mean, you're sort of putting the technology before the actual business challenge that you're trying to solve, which is, which is backwards. Yeah.

[00:30:29] And very often, you know, there's, there's, there's, there's endless reports that basically say the same thing that there's an awful lot of people building quote unquote agents for things that could just be basically automated rules.

[00:30:42] Had they started, had they started with the problem and correctly described it and then done some level of design thinking, they'd have found that it could just be a set of, if this, if then that kind of statements and rules that could easily sit within a, an automation workflow and definitely don't need to be kind of exposed to large language models.

[00:31:03] The, I think the topic of AI readiness is, is interesting insofar, so on, you know, the little practice I'm building I've described as being an AI readiness practice for workforce products. Now, my interpretation of AI readiness might be a bit narrow and maybe in a few years time I'll end up, you know, having different language to describe it.

[00:31:29] But essentially there are some people who talk about AI readiness, I think as you do, and I think Martin from Warden have talked about AI readiness effectively as a, a quality of the individual or a kind of an individual level set of skills and awarenesses and knowledge. Maybe some elements of kind of individual plasticity in there about whether people are ready to use AI within their own job. That's cool. That's a thing. Is AI readiness the word?

[00:31:59] I don't know. There is a very technical version of AI readiness, which is leaning on the side of what you were saying about evaluations and hyper kind of particularly large language model specific guardrails controls and comparisons, like performance comparisons. Is that AI readiness? I don't know.

[00:32:25] But there's certainly people who talk about that kind of hyper technical language model control as a part of being ready to use AI within your business. Totally respect that. That is definitely a thing, right? All be it, it's a thing which can only really belong to people with the right data science backgrounds, I would argue.

[00:32:53] Then there is organization or at least kind of what I think of as organizational AI readiness. Are we ready in terms of the information flows to purchase tools that are AI enabled, which can then prescribe actions to us like put Bob onto this project to do that. He's the right person. He's ready. He's available. He wants to do it.

[00:33:19] You know, it's prescribed an action about what Bob will do next or indeed if sales go up 50%, how many more Bobs are we going to need and what's the balance of skills and locations that would be most optimal according to our talent intelligence reports, etc.

[00:33:38] That is the bit which I think is in my terminology is AI readiness, which is like, are we ready to use these tools, specifically workforce tools? If the kind of, if the industry ends up settling on different terminology, you know, I'm not like, I'm not going to be, I don't have any particular dog in the fight. But I think readiness is being used to describe a lot of different things. I think that's fair.

[00:34:08] That's fair. Yes. I mean, I do have a sort of human centric, you know, slant, of course, which is, I think people who listen to the show and have seen my logo understand that that's sort of the angle that I take. But, you know, I did work for corporate for, you know, 25 years.

[00:34:27] So I get the desire and the inclination to think about the organization and its systems as well as its people, its data, its processes and things like that. And I think that, but even when I say human centric, there are different sort of levels of that, right? Yes, there's individual AI readiness.

[00:34:52] Have people gone through some fundamental, you know, literacy training that includes responsible AI, but it definitely cuts across tools, skills, mindset, your adaptability, your ability to learn and grow and recognize that there's a lot of potential here. Even though it's not perfect and it may not be the solution for every problem, which is also why I like thinking about it as, you know, necessarily need.

[00:35:21] I mean, yes, you sort of need an AI strategy, but it's really an AI plan that aligns to the business strategy. Because if those two are disconnected, then what exactly are you doing and how are you maintaining competitive advantage? How are you realizing what actually is your moat? Because it could very well, to your point, be your data.

[00:35:46] But I do also recognize, you know, we said for a long time, individual productivity, you know, these metrics, so I'm saving this time, da, da, da. It's like, that doesn't translate upward, right? It's like, we work in teams. That is sort of, I've called it a vanity metric for probably two years or so. But how are your teams more effective, right?

[00:36:12] What is the AI readiness of the team, the department, the division, line of business, and then rolling up to the organization, which is now a combination of the people readiness as well as, you know, the data readiness or data maturity and your processes. Have they been overhauled, redesigned? Have roles been redesigned? And you really start to get into broader organizational readiness kinds of attributes.

[00:36:40] But you've got to understand AI's role in all of those exercises that you're going through. And then, of course, the human centricity aspect is what should stay on people's plates, including, of course, all decisions, especially talent decisions.

[00:36:58] But any major decision that has consequence to your customers or to your people are ultimately human judgments and decisions to be made because only humans can be accountable for this decision. Yes. But if you look at something like a resourcing optimization tool, yeah, the human gets kind of some level of input.

[00:37:21] But it's not like the human has the horsepower to work out the right alternative distributions of the workforce to meet the demand, right? That's too big a bit of computation.

[00:37:40] Like, so, yeah, human is kind of, you know, the human's playing a role, but I don't think we should be under too much of an illusion that, you know, in a lot of cases, they'll be kind of rubber stamping or making tweaks around the edges. But we're beyond the point where the humans are actually making the substance of the decision. They are just acting as a kind of a failsafe on have we done anything monumentally unfair or monumentally stupid.

[00:38:09] Well, yeah, no, I agree. There's a sort of division. If you call AI part of the labor force, it is a division of labor that you need to consider. And I hear you. I just think if we're rubber stamping, I feel like that is a pretty low bar. At least we've got to have AI that is explainable, that is transparent. And we can say, I understand its logic.

[00:38:35] And thank you for going through those, you know, thousands and thousands of, you know, profiles or, you know, whatever the data is on my behalf. But if it's just like, oh, you know, this looks good, then to some degree we've outsourced critical thinking, which is a problem. Well, we probably have. I mean, when we follow Google Maps and it says, oh, we could have sent you this way. It's three minutes slower, but you'd have made five less turns. Or we could have sent you this way.

[00:39:04] And it's, you know, going over a scary bridge that you said you didn't want to go over. You know, it's giving you a kind of a series of choices. But I don't think we should be under any illusion that it has worked out all of those alternative paths. And it said, here's your three to choose from. And ultimately, the system doesn't really, you know, it's that's all just in the UI. Right. And the decision making.

[00:39:33] What's possible within a user interface in terms of explaining how a decision is come by is very thin. It's kind of a pastiche, a bit like when a map says, here's your three different routes. Like, it's very cursory. And yes, systems need to be explainable in a legal sense to a specialist lawyer in a sort of a slow and serious way.

[00:39:59] Should somebody feel like they will not feel like if somebody believes that they have been disadvantaged in some material way and they want to bring a case or they want to challenge a given decision. Like that does need to be there. But it can't all be there in the UI. Like it's too much information. So, Olivier, what about if it was driven through voice? As you know, voice is becoming a pretty dominant interface to a lot of these tools.

[00:40:26] So I don't know if you have Waze in England. It's been, yes, largely been. I used to use it. It's largely discontinued now. But yeah. Okay. Okay. So I, you're right. I mean, I trust, I generally trust Waze because, well, I use Waze versus Google Maps or Apple Maps. It's because Waze shows me where the police are. But in general, it's far from perfect.

[00:40:52] But it is, it's picking those top three choices without any input for me other than avoid tolls or like you said, avoid bridges or something like that. Well, I'm not driving a commercial vehicle. I'm driving my personal vehicle. I can pretty much go on any road, including dirt roads, technically. But, but I would not have made the choices that it made.

[00:41:16] And here in the Northeast United States, like just getting from, from New York to Boston, there's a hundred different ways I could go. So I don't, I often don't like the three choices that it gave me because I know the roads and I know some of the alternatives. I have no way to stay in sync with the mapping system if it only gives me those three choices. But if I said, I want to follow this interstate to this point and then I want to go on this road or whatever.

[00:41:45] So just map it and tell me if I'm going to lose a half hour in the process or whatever. Like right now I have no way to do that. And it's just hard for me to think that here we are in 2026, that in, in short order, there wouldn't be a way for me to just tell it exactly. Unless you have a huge objection that I can't see or predict, this is the way that I would prefer to travel.

[00:42:11] I mean, it sort of gets back to what we talked about old software versus AI built, you know, solutions. It's free form. I can go, why can't I just have the map follow me the way I want to go? You kind of, well, you kind of can because you just drive where you want to go and it works it out in real, it just works it out in real time for you. It says, oh, true. You know. True. But when we're in a fully, full, full automation, fully self-driven car, if I'm in a Waymo, right, and I want to do exactly what I just described, can I do that?

[00:42:39] Or is the Waymo just going to be like, no, no, no, this is, this is the way. I know better than you. Just sit down. You're the passenger. I'm the driver. Sit down and shut up. I don't know. Yeah. Yeah. I don't know. I think, I think, I think whether we do, however far you stretch it, but I think that's a decent kind of analogy to the fact that like we have to accept that any sophisticated enterprise system is looking at how you might match people and work.

[00:43:07] Whether it's fast or slow or over hundreds of people or thousands of people, it doesn't really matter. As soon as you're asking to make those prescriptive decisions, you are going to get a series of choices that are effectively all off the same model and all off the same set of underlying sources of information about people. The question is, is the information on the people and the work itself good enough to sustain that sort of functionality?

[00:43:36] And 99% of the time today, it's not. And that's what's hard. That's the kind of, that's the split. And I think what I've also started to realize is that the, and this is kind of one of my individual business development challenges is that, you know, if you talk to people off the record, they will say, yeah, you know, we recognize that enterprises aren't in a place where they can yet use our tools to their, to their full potential.

[00:44:04] But equally, we would be very, very silly to interrupt our sales process or any part of our deployment in order to point out the bloody obvious. So at what point, like, at what point do we put on our big, big boy pants and say, well, like, it's fun to have all of these aspirations, but it doesn't bloody work until we, until we make a check. No, that's fair.

[00:44:32] I think, you know, I know we kind of got up on a tangent there, but I think it's, you know, it's, it's, it's still one of those coordination problems, you know, humans plus AI, you know, what is, what's the optimal path and how are you defining optimal, right? Is it, is it time saved? Is it, you know, that, you know, in that case, the driving experience, I mean, there's, there's a lot of things to factor in. Yeah, no, I, well, I'll tell you what I think for what it's worth.

[00:45:00] I think it'll become normal for people in knowledge work to do a degree of self disruption or like they basically learn to work with the AI to do their traditional job a little bit better, a little bit faster. And they show levels of performance improvement, right? That's going to kind of become a new normal.

[00:45:22] But you only really get the gain of the AI system once you start to do some level of consciously designed change to the work process. And you can't do that to every job across the enterprise tomorrow. You have to start with just the processes that generate a lot of value or they generate a lot of cost, like whichever it might be. And when I was at, when I was at WPP, you know, the, it was just at the very beginning of the LLM era.

[00:45:49] But what was extremely obvious is that the, the entire value chain of how adverts are made were completely being changed by this technology, right? It used to be lots of kind of taste testing and audience segmentation and understanding followed by lots of very well-paid kind of art and creative directors determining what advert might be made.

[00:46:17] And then the, like a low quality version of the advert might be made, but ultimately millions of dollars were being spent to, to create one bit of content and never so slightly people were crossing their fingers. That entire value chain just got completely disrupted by the fact that you could pump in a, a prompt to an image generation engine that was given a few assets of a given kind of soda brand of your choice.

[00:46:44] And you could have, you know, a kind of a sexy elf sat on a tango bottle in a Christmas scene or whatever the hell it might be. But, but ultimately the whole thing's taken 30 seconds, right? It, it, it, it, it, it doesn't mean that the output is good, not necessarily, but has the value chain completely gone upside down? Yes.

[00:47:04] And there, that's only one of many, many versions of the same story whereby, you know, when the value chain has been completely changed, it's going to become entirely needed that not, sorry, let me go back.

[00:47:23] If we accept that AI is going to do some part of the value chain better than the human, you're then going to need orchestration systems that determine points of handoff within a workflow. And that is, that's likely to be much more disrupted and value generating than what we all do, trying to make ourselves 10% better accountants or 10% better lawyers or whatever it might be.

[00:47:52] But there's sort of that bottom up, how do I do my job as it stands better question is interesting, but I think it's probably also only a, almost like an entry level kind of starting point for most of corporate life in the next five years or so. Yeah. Yeah. Yeah. I mean, I, I think there's, there's that, like, how could I do my job better? But then there's the, for some people, this is a harder exercise.

[00:48:21] How do I, how do I disrupt my own role? How do I rethink how my role needs to change and my capacity to change with it? Can I, am I comfortable sort of deprecating or, or sort of outsourcing, you know, some parts of my job to some type of, you know, automation or agent or, or what have you. And then what do I pick up as a result of that?

[00:48:49] What is the, the work that I haven't been able to do because I didn't have the capacity. And of course, we've got to be careful how much we sort of repile onto our plates. But there's a lot, this is where people get tripped up and they have fears that they can't handle the, the change and the disruption and, you know, who moved my cheese kind of scenario.

[00:49:10] But I don't think there's a role, you know, knowledge work or, or otherwise that isn't going to be impacted by, by AI in some way. And it's just, how do, how do we get people to, to adapt, even if they haven't had to do that through, you know, digital waves and things like that, or at least that was more incremental. That was a slower change. And people, I think feel it's the pace of change. It's really throwing people for a loop.

[00:49:40] A hundred percent. And I look, yeah, I don't know the answer, honestly. Like I, I get the fear. We all, I think we, I think we'd all be a bit weird if we didn't get the fear. Right. We, we, we have to go through periods of, of reinvention. And so, yeah, what you describe in terms of that culturally helping people to experiment with how they change their job.

[00:50:02] Like that is a serious problem, but I think it is entirely discreet from the question of AI readiness as I have framed it. Maybe it's a bit narrow, like maybe other terms need to come up, but essentially are we at a operating system level? Is the business ready to use AI to manage its workforce, which is a sort of a, a separate and equally tricky question.

[00:50:31] Yeah, no, that's totally fair. And I think we're just coming at some of these organizational challenges from, from different angles. I do think they're, we talked about this in the, in the WhatsApp group the other day. I do think they're, they're complimentary, right?

[00:50:45] I just, I just don't want people to think that if you have a high level of data maturity and you've got the, you've got the budget and you've got tools and you've got, you know, copilot on everybody's computer and all this stuff that just like, we're, we are ready to go. And you're like putting out press releases or whatever. It's like, but your people have no clue what you're even talking about. Right.

[00:51:07] I went through this at IBM when Watson came out of the labs and whatever people you think, I mean, 300,000 people, you know, there's probably only a couple thousand that even knew what Watson was doing and what it was capable of. What about everybody else? Right. So, so you've got to have this sort of, you know, bottom up and top down, you know, movement with, with, you know, change agents and champions. And you've got to show people what's in it for them and that you're not going to leave them behind.

[00:51:36] Part of your AI investment is, is on the people side. And I just don't want people to forget that because if the people aren't ready, like what exactly are you doing? Something's going to break. Look, I've got no, I've got no argument on that. And it would be lovely if McKinsey or HBR or somebody would just kind of give us all a, give us all some like jargon that we could all just grasp onto and say, yes, this describes the problem. And, you know, get the walls up. And I, the problems you describe are very real, very real.

[00:52:06] I just, I just don't have any individual expertise in, in that side. Like, yeah, no, that's, we're not going to, we're not going to solve it here for sure. So we're going to, we're going to put a pin in that. But Olivier, I want to be respectful of your time. And I know you're probably baking in that office of yours in the hundred degree London heat. But as we wrap up, I just wanted to give you an opportunity to just talk a little bit more about what you're doing with, with Sightline and how people can find you and any just, you know, parting words of what's done.

[00:52:35] Oh, that's kind. That's really kind. So, yeah, look, Sightline is a new AI readiness practice for workforce products. We are primarily trying to build a business which is a bit like a systems integrator. But instead of looking at how you plug it in and configure it, we look at all of the client side data and knowledge that feeds systems and makes them work.

[00:52:59] It's very much a kind of built out of my own experience as a product builder and equally looking for enterprises who are struggling with these people system problems, how to actually make them perform to their potential. So Sightline-AI.co, I'm very easy to find on LinkedIn. There's not very many Olivier Vidal. So really happy to chat to anybody in this industry. Excellent. Sounds great.

[00:53:28] Yeah, I'll make sure to put your LinkedIn and the Sightline URL in the show notes for the episode. Thank you, Bob. Really appreciate it. Of course, of course. Always great to talk to you, Olivier. Thank you so much for finally getting in the studio here and recording one of our really interesting conversations. So thank you. And of course, best of luck to you. Thank you, Bob. I appreciate it. Have a great day. Absolutely. Thank you. Thanks, everyone, for listening. We will see you next time.