Lead With AI

Unlocking Human Potential With AI (Featuring Boston Consulting Group CTO Matt Kropp)

AI can drive growth beyond cost-cutting. Learn BCG's strategies for successful AI implementation to unlock hidden value in your organization and spark joy.
Last updated on
July 9, 2024 5:00 PM
15
min read
human-ai-featuring-boston-consulting-group-cto-matt-kropp
Daan van Rossum
Daan van Rossum
Founder & CEO, FlexOS
I founded FlexOS because I believe in a happier future of work. I write and host "Future Work," I'm a 2024 LinkedIn Top Voice, and was featured in the NYT, HBR, Economist, CNBC, Insider, and FastCo.

🎧 Listen Now:

Welcome to the new Lead with AI podcast.

In this series, we speak to senior leaders responsible for rolling out AI in their organizations and uncover deeply valuable insights into what success looks like.

In today’s episode, we speak to Boston Consulting Group’s X global CTO, Matt Kropp, and hear about:

  • Why 70% of AI's value comes from people, not just the technology itself
  • How to successfully implement AI, including the importance of employee engagement, co-designing solutions, and continuous education through peer sharing. 
  • How AI can drive business growth beyond mere cost-cutting, ultimately creating better jobs for more people, and why focusing on 'joy at work' might be the key to unlocking AI's full potential.  
  • What factors you should consider when choosing which AI platforms to select for your teams and building your own AI technology?

Key Insights from Matthew Kropp

Here are the actionable key takeaways from the conversation:

1. Successfully Rolling Out AI Means Focusing on Joy.

To roll out AI successfully, you need a couple of things, according to Matt:

  • You need a strategy that answers where you can use the technology and what impact you’re trying to achieve. 
  • Look at the end-to-end process you’re trying to affect. What is the work? 
  • Identify where there is toil and automatable human effort in those steps, the work people don’t like to do. At the same time, identify where there is joy, and preserve that. Use AI to create better jobs for humans, which will boost the quality of work, job satisfaction, and employee engagement and retention.

2. AI is a Change Effort.

Matt shared that BCG talks about the idea of 10-20-70 when building technology.

10% of the effort and 10% of the value come from the algorithm, 20% from the software, and 70% of the value and effort come from driving change in the organization.

The true value of AI implementation lies not in the technology itself but in organizational change.

When implementing AI, allocate significant resources to change management and employee adoption strategies, not just technical development.

3. Co-Design and Educate to Get Everyone on the AI Train.

In the example Matt mentioned, only 15% of call center workers voluntarily adopted a new AI solution. This makes sense because they knew the risk of being displaced.

Among developers, some switched on the AI only one day per week, because they didn’t know what else it could do for them.

Matt says that, therefore, the key is to co-design solutions with the people using them. Involving employees in the AI implementation leads to better adoption and more effective solutions.

Create cross-functional teams that include end-users when designing and implementing AI solutions to ensure they meet real needs and are readily adopted.

Then, ensure education about the technology's benefits. This is best done through community engagement like ambassadors who can support others, peer sharing through for example hackathons, and nudges that create the habit of using AI.

4. Selecting your LLM Starts with What Productivity Suite and Cloud Provider You Have.

With continuous product upgrades across the major LLMs like ChatGPT, Google Gemini, and Microsoft Copilot, choosing your company’s LLM isn’t easy.

Matt advises separating the questions about which tools you’ll get for your teams and what you need to build your own AI applications.

For the employee tools question, consider what productivity software you’re using and paying for. For example, Office 365 Copilot has a big advantage for companies using the Microsoft Suite for work, as it can train on that data.

For development, similarly, choose based on which cloud you’re on and which LLM is best integrated with that cloud.

5. Matt’s final point is to not just focus on costs.

Yes, we’ve heard a lot about AI being a cost saver, but also consider the tremendous opportunity.

As Matt says, 80% of jobs will be augmented by AI. For the rest, we still need humans. So, create a better job for them, even allowing them to do things they could never have done before.

For example, in the insurance industry, underwriters can now be freed from doing a lot of data input work and write policies in three hours instead of three days, significantly increasing their win rate.

Can you use AI to reinvent your offering?

🔔 Available on:

Transcript:

Daan van Rossum: I know you work in BCG and then specifically in BCG X. Can you tell us a little bit about yourself, the company, and what you're doing in the realm of AI? 

Matthew Kropp: Yeah, absolutely. I'm the CTO for BCG X, and everybody's heard of BCG, the management consulting firm. You may not have heard of BCG X, a fairly new brand, but we have an AI build division, which is BCG X, which has 3000 engineers and data scientists. We are building AI solutions and generating AI solutions as part of our work with clients. So I'm running the general AI topic, and I'm heading up the whole engineering organization there within X.

Daan van Rossum: That sounds like a lot of people—3000. So give us an idea of what those 3000 people do.

Matthew Kropp: It's sizable. BCG is 30,000. So it's actually only a fraction of the overall BCG company. So, we are a development shop, but quite different from others. You have your systems integrators out there, like Accenture, Deloitte, etc. We don't really compete with them. Really, what we're doing is building solutions that are tied to business strategy.

So when we're working with clients, we'll have both the BCG consulting team and the BCG X development team, and we're focused on how we can make an impact. What is the right strategy? What are the right solutions to put in place? Our BCG X team is building those solutions. And then our BCG consulting team is helping our clients actually get value out of them. We may want to talk about that.

We talk about this idea of 10, 20, and 70: when you build technology, 10% of the effort and 10% of the value come from the algorithm, 20% of the effort and value come from the software, the data, and the connections into other systems, and 70% of the value and effort really come from driving change in the organization.

And so that pairing, BCG plus BCG X, is really that pairing of BCG X giving you the 10 and the 20, the algorithms and the software, and BCG giving you the 70, which is the right strategy and the right change efforts to actually get the technology to create value in the organization.

Daan van Rossum: That's something that maybe people intuitively don't really get—that this is not actually a technology issue. It's mostly a change issue because there's been technology for quite a while that can automate; obviously, now things can be done on a much broader scale and in a much faster and smarter way.

The change part of it is the part that I think a lot of people are very interested in. So a lot of people that we speak to, a lot of people who are in the community, see the opportunity in AI, and maybe they have some great ideas about how to implement AI. And then, they're looking into their organization, and they're realizing that it's going to take a lot more than just launching a software product to actually get people to embed AI in their day-to-day lives.

What are some of the principles that you're either practicing internally or also sharing with clients when it comes to AI in the workplace? 

Matthew Kropp: I totally agree with your framing. AI is not new. We've had AI for at least a decade, maybe more. We've called it different things, and it has consistently been as you described. Just implementing technology or writing the code doesn't have an impact. You actually have to get the change to happen in the organization to get an impact out of it.

This new wave, so-called generative AI, is the latest evolution of AI. I would say is even more so, and the reason is because, if you think about a lot of the AI solutions that companies have been building in the past, they're really narrow solutions tailored to a specific task. I need a model that predicts something. I need to predict the next best offer to show to a customer.

I need to predict what my demand is going to be. And so, that's a very narrow application. Generative AI is a general-purpose tool, and it affects how humans work. It's something that empowers humans. So even more than in the past, you have to get the people who are working with the tools to work in a different way so that you get value from it.

We're in this new evolution. Generative AI has really only been commercially available for about a year. Yes, it's been around for a few years before that, but really, enterprises have only started being able to build solutions, starting early last year.

So we are very early in the learning curve, and so I've been out talking. I've spoken to over 200 companies in the past year, to CEOs, boards, and executive teams. And most of last year, they were really just focused on learning. What is this technology? How do I use it? They were experimenting.

They were building POCs and chatbots to do various things in their organization and not getting much or any actual value out of them. Now companies are thinking about how to get value and how to make an impact. And so they're doing a couple of things that I think are really important. The first is really having a strategy, identifying where you can use the technology, and determining what impact I'm trying to achieve.

We might want to talk about this, but the impact could be that I want cost savings. I want to reduce the amount of labor that's in a particular activity, or it might be that I want to increase the quality of people's work, or I want to drive revenue growth, or I want to change your customer experience.

There are all sorts of things that you might get out of this. But starting with the idea of what the objective is that I'm trying to solve, rather than building a chatbot and giving it to my employees, is critical. So start with the objective.

Then we think it's really important that you look at the end-to-end process that you're trying to affect. What are people doing? What is the work? If I'm an insurance agent and I'm doing underwriting, there's a whole set of steps that have to take place. I need to go get some documents. I need to read lots of reports. I need to do some research on some databases. And then finally, I get to making a decision, coming up with a price, writing a policy, etc.

What are all of those steps that are involved in that process? And then identifying where in those steps there is toil, a lot of human effort, and a lot of human work that can be automated.

Also, that may have a low value. That's not very fulfilling, which people don't necessarily like to do, and identifying, where is their joy? Where do people really get fulfillment from their work, or where do you actually want humans to be involved because they bring diversity of ideas, creativity, risk management, relationships, building, and customer experience? There are all sorts of reasons why you don't want to take people out of processes as well.

And so thinking very holistically about how to redesign the end-to-end process with AI and with AI working with humans is how we think we really need to be working in order to get not only efficiency and productivity but also better quality and, frankly, a better job for the humans that are involved in working on that process.

Daan van Rossum: And so you're actually practicing the parts that we've only been hearing in theory, which is that AI can take over the manual, repetitive part of your job—the part of your job that you probably don't really like all that much—so that you can focus on the stuff that you actually like.

When you're going into companies, is that a message that resonates when you're saying, Look, you've got to look at the entire workflow and you've got to look at, like, where are people just dreading to come in because they have to do data entry or because they have to go through manual processes or because they have to click on a bunch of buttons in an application? And where are people coming to work and really showing up for what they're uniquely good at and what they like doing? 

Is that something that resonates when you talk to companies? Obviously, when you talk to individuals, it resonates deeply. How does that go when you go into a company and say something like that?

Matthew Kropp: No, it absolutely resonates. And I would say it's been interesting, especially when I'm talking to senior leaders and boards. This was a question that everybody had. They're very excited about the technology. The technology is incredibly powerful, and in some ways, it is mind-blowing what you can do with it, but they very quickly came to the question of what this means for my people.

I think maybe we have expectations that senior leaders just care about efficiency, but they really care quite a lot about their employees and about engagement—about their employees actually being engaged and really bringing all of their capabilities to work.

And so there was a lot of anxiety about this idea: if I'm introducing this automation, am I creating an environment? We're actually going to hurt employee engagement. So, they've been very receptive to this idea, and I bring up this idea of minimizing toil in people's work and maximizing joy as the formula. Every time I talk about this, the executives’ eyes light up, and it clicks a switch. I can see how this is actually not a negative for my people, how this is actually a net positive, while I still get to take advantage of this capability.

We've done some research on this. It initially came out of COVID, work from home, and hybrid work. So, we have a think tank, the Bruce Henderson Institute, that did a bunch of research around employee engagement. So what is it that keeps people engaged, as you said, wanting to come to work and bringing all of their capabilities to work?

There are a bunch of findings, but the part that I find most interesting for the AI and generative AI topic is that when you do things that improve productivity, it basically doesn't affect employee engagement. Small 5%–3% improvement in engagement. When you do things that improve joy at work, it has a big impact on engagement, and I always joke. Maybe that's kind of obvious. If I make you happier at work, you're going to be happier at work. All of the discussions about AI have tended to revolve around productivity.

Essentially, we're saying, Hey, we're going to bring you tools that make you more productive, and what people hear is that the robots are coming from my job. And I actually give an example of this. We're working with a tech company that has implemented a chatbot to support their customer support reps. So their call center reps, and it was supposed to make them more productive.

They were very surprised that only 15% of their reps had adopted the tool, and they're asking the question of why my people are not adopting the tool. It's pretty obvious. They know exactly why you're implementing this tool. It's to reduce their jobs. So we have to take this different frame, which is that there’s a lot of work that is not value-added.

In typical jobs, up to 25% of people's work is just administrative tasks that they don't enjoy doing and that you can automate. And so we really have an opportunity to change the way people work.

Daan van Rossum: Yes. And I had a great episode with Debbie Lovich where we went very deep into that research, and it just made my day because I think it's really what this is all about. If we are going to intervene in the way that people work, let's optimize it for joy, not for being more productive. Of course, we want to be productive as well, but those two things come together quite nicely. 

So then, at an executive level, you're like telling the story, and their eyes brighten and their ears _____ that we can use AI to make people happier at work, and therefore, they will stay with you longer and they'll enjoy their job more.

How do you then get to that adoption level? Because it may sound great in theory. And then, like you said, you build a chatbot to support call center workers. And 85% say so no thanks. I'm not going to dig my own grave. 

So what are some steps that you recommend those companies take to actually go from that initial idea to people truly using it and feeling that it's actually benefiting them?

Matthew Kropp: The key is co-designing the solution with the people who are going to be using it. And so maybe I'll give an example. One of the areas I think is quite exciting and moving quickly because it's a bit more mature is software development. Generative AI tools can support coding in pretty profound ways. GitHub Copilot is one example of a product, but there are others out there that happen to be the ones that have the most market share at the moment. So we're doing a bunch of work with clients that are implementing GitHub Copilot and generative AI solutions in their software development life cycle, and it's been quite interesting.

So, we are working with a bank that has rolled this out to several thousand of their engineers, and so we were talking to the engineers about it and trying to understand their maturity journey and their learning journey to using these tools.

And what was interesting is that, in the first place, they were very excited about the tools. They said, This is great. Most of them were able to come up with examples where they said this can do things for me that I don't like to do. And so, I love it. I hate writing unit tests, but it's really good at writing unit tests. I hate writing documentation. It's really good at writing documentation.

But at the same time, most of them were not using the tool to its full potential. And so they had all self-discovered one part of their work that it was helping them with. And they were very enthusiastic about it, but they didn't see the other opportunities. My favorite example was that one of the engineers was very passionate about it. He said, I love it. I use it for writing my unit tests. But I only write unit tests on Fridays. So Monday through Thursday, I turn the tool off so that it's not in my way. And then on Friday, I turn it on and use it for my unit tests.

And then we asked the question, Well, do you think that four-fifths of the time you might also be able to get some benefit from other things that you do? And so the key learning from this was that you have to, first of all, involve people so that they feel ownership of the process. You have to give them some basic education so that they understand how it works and how it can be part of their process.

But then you have to do a lot around community engagement, peer sharing, nudges, reminders, and getting people to really be in the habit because we're not in the habit of using generative AI as part of our work. We need to get into the habit. And so a lot of this is both a reminder and seeing what my colleague is doing and hearing from my colleagues. So what we did with this client was set up a bunch of hackathons where they worked on a specific problem, but they collaborated, and we heard a lot coming out of the hackathons where one of the engineers would have some technique that they had developed and the others would say, Oh, that's great. That could work for me. And so they're sharing; they set up ambassadors. So basically, superusers who were on call spoke for others so that they could share how they did things.

So I think it's really going to be incumbent on organizations, not just to say, Here's a tool or here's the onboarding training." It really has this continuous sort of community of practice so that the people that are using the tools start to reinforce one another.

Daan van Rossum: So you're really mapping out an entire journey, which actually starts way before the rollout, which is to say, you're going to get people involved in designing maybe what tool gets used and then how that tool will be used. 

And then when it actually launches, we have a lot of touch points to make sure that, for example, the people who feel very comfortable with it, and maybe who are most enthusiastic about it, that they share some inspiration on, Oh, here's how you can also use it. And if you need any help, you can approach me. Hopefully, companies will free up time for people to do that and make sure that those ambassadors can be reached. 

What do you deal with in that process? Do you deal with people who have already adopted their own tools? So we saw from the Microsoft research the whole BYO AI thing.

People are bringing their own AI tools to work. Do you now sometimes have to retrain? Let people unlearn maybe other tools, or maybe, in the case of a Copilot for GitHub, at least some people were probably already using that. 

So in that discovery journey, do you do any surveys around what you are currently using? What are you most open to? What kind of pain points would you like to solve most? What does that part of the journey look like? 

Matthew Kropp: So I think we're pretty early in this evolution. So, I don't know that I would yet say that there are a bunch of bad habits that people need to unlearn because I think they're still learning just the first set of habits.

But I think there's something really interesting in what you're talking about, which is that we have typically had software that was, as I was saying earlier, purpose-built, that does one thing. And this is the general purpose: ChatGPT Enterprise, Bing Chat, or Google Gemini, or choose your tool. It knows about everything. It is able to do a wide range of work. If I know how to ask the right questions, if I know how to interact with it in the right way, we will certainly have very purpose-built tools that incorporate Gen AI, and that'll be.

First of all, all of our software is going to get Gen AI as part of it because every software company is busy building it into their tools. Enterprises will build purpose-built tools that bring Gen AI to different processes, but then you're going to have the ChatGPT enterprise, kinds of tools where your employees will figure out how to automate their own work.

And I think this is actually probably the place that is going to have the biggest impact, but it's going to take the most time because we all have to learn how to work differently. So at BCG, we're going through this evolution right now. We were very early to roll out ChatGPT Enterprise to all of our employees.

So all 30,000 employees have access, and we've been doing education and telling people how they can use it and what they shouldn't do with it. But really, the value comes from our individual consultants. Figuring out for themselves the work that they are doing at the moment, how can I use this to help my work?

And so, all of us humans are going through this joint discovery process of figuring out how this changes the way that we work. And it's going to take some time for us. Frankly, the tools are evolving. We will learn how to use them more effectively. We will learn when we should use them and when we shouldn't use them. That's got to be a big part of the motion that organizations go through as they bring these tools into their day-to-day work.

Daan van Rossum: Yeah, it must be a very strange time for L&D teams, for example, who have always been used to the idea that you can study something and then teach it to people, but there's no studying because you're releasing something that you also don't really know how it works, how it will work, or what the use cases are. 

I think what you said, Matt, about the way to roll it out, especially with those enterprise-wide main LLM tools, is what we saw in the Moderna case study and what we saw in the Microsoft HR case study, which is that everyone has access. You get a bit of base education, and then you go figure out where it fits in your workflow the most, what parts of your job you can now automate, and what tasks you can remove from your job. Because, as Josh Bersin said a couple of months ago, frontline workers typically know the work better than management and, typically, better than leadership. 

So in that process in your company, have you adopted certain rituals or behaviors again to let people share those use cases? So that again, if someone discovers something that is really great in their own workflow, how does it extend to their team and then maybe department geographies because you're global as well? What are some practices you have in place to make that happen? 

Matthew Kropp: Yeah, I think it's pretty important. Although we need to develop other techniques as well, we are doing what we're calling discovery sprints. And so we have set up identified teams. So this is just project teams, client projects that we're working on, but specific teams where we have made sure that they understand how to use the tools, but we're really following what they're doing and using them as the innovators; they're figuring out how to use these tools within their work, and then we're capturing that, surfacing that, and then broadcasting that out to the overall company.

So I think we're going to have to do a lot of things like that, like my example with the software engineers where they're doing hackathons, this idea of the power users with the ambassadors, it's going to be a whole host of those kinds of ways of both identifying or observing what people are doing, get catching those new techniques essentially, and then broadcasting them back out to the overall company.

Daan van Rossum: And then one other related question that I had, and you're okay to say, I don't want to answer this question, but I'm very curious because it comes up a lot in the conversations I'm having, which is that companies that are, you're one year ahead of them. So you've already decided to go with ChatGPT Enterprise, which I'm assuming was one year ago. You must've almost called up Sam Altman and said I want this for my whole company, because it wasn't quite a ready-made proposition as it is now. 

Companies that are still looking into it—maybe they're already on Microsoft, like most larger companies—should I go with Microsoft Copilot? Should I go with ChatGPT Enterprise? Are you in those kinds of discussions where companies are trying to decide which one to use? Then, obviously, you cannot give any particular advice, but what are some things that companies should be looking at as they're looking at what platforms to adopt? 

Matthew Kropp: Yeah, the platform selection question comes up all the time, and I would break it into a couple of parts. So there is the deployment of tools for your employees. So that's the ChatGPT enterprise, or Office 365 copilot, or Google's Gemini is part of GSuite, and so forth. I think I would make that choice based on, first of all, what productivity software are you already using?

What is your stance? What is your InfoSec team’s stance on? What tools are they going to allow? I think Office 365 Copilot has some advantages because of the tight integration with Office. I think what we're seeing right now is that some companies are adopting both Office Copilot and ChatGPT Enterprise, and they're letting their users pick when to use which.

Daan van Rossum: That's a nice monthly bill. Okay.

Matthew Kropp: Yeah, exactly. Presumably they will rationalize that, but then there's also the choice of what models you're using. What platforms are you using for development? And I think there's actually been more of our client conversations around that choice.

And I would say that, of course, this is all evolving very fast, so it's changing. But the conversation so far has been first around: What cloud are you on? Generally, companies have standardized on one of the three hyperscalers as their core cloud. And then, are you comfortable with the foundation of the frontier model available in that cloud?

So OpenAI is for Azure. Anthropic as being the most advanced model on AWS and then Gemini for GCP, or do you want to be able to use multiple models, and then are you comfortable with the security posture of not just using the model that's within your hyperscaler?

So that there's this question of where does my data sit? Where is my infosec team comfortable in terms of my cloud posture? What are the models I want to use? And then what do I want to use those models for? And so do I need the latest frontier model? Do I need GPT for instruction following capability for everything? Or do I have some needs where the requirements are less and I can use a cheaper model? Or maybe even I have very high volume and I need to have low costs. And so I actually want to operate Llama 3 in my own cloud because I want to control both the costs and the perimeter around the data. So those are the kinds of conversations that the clients are having.

Daan van Rossum: Where are the most people? So you obviously speak to a breadth of clients and a lot of different companies. Again, different geographies. Is there any consistency, like most companies are at this stage where, again, they're looking at custom development versus companies that are still looking at what tool to even use? 

Where do you see companies, and what's their status? And perhaps linked to that, what are some misconceptions out there as you're speaking to these companies? What are some misconceptions you're still seeing when it comes to Gen AI? 

Matthew Kropp: Yeah, it's a wide range. There are still companies that are taking a very cautious posture. In fact, I had a recent conversation where they had not yet allowed their people to use ChatGPT. And we had the conversation that you realize that your employees are using it. They're just using it at home for their own personal accounts. And so maybe you watch, and you want to bring it into your enterprise. So you have some control. So there's a range. There are companies that are still very early.

And that's more industry. If you're an industrial goods company, you're not feeling a lot of urgency to adopt this. If you're an insurance company, this feels very existential because your business is entirely text. Your business is a contract. And so the whole insurance industry has really leaned into this technology. I think they see it as both a competitive advantage and a big threat if they don't adopt it. So I would say it's a range, but you end up with choices around how much they deploy with their employees. So, what are the general-purpose tools, and then how much are they investing in building custom applications?

The leading companies are the ones that have really identified this as a key technology that they need to have as part of their organization. They're leaning in by defining: What are the key areas that they want to focus on? So what are the big rocks or the big moon shots that they want to develop?

How do they empower the whole organization to come up with ideas and give them tools so that you can have smaller departments that are coming up with their own automation or even individuals, as we were talking about figuring out how to innovate on their own process?

And then they're putting in place the right platform in terms of data, the right platform in terms of technology partners and models, putting in place a center of excellence in the engineering functions so that they can support building out these applications, and putting in place the right governance so that they're picking the right bets.

Investing in the right places, tracking that they're actually getting impact from those investments, and making sure that they're building responsible AI into everything that they're doing means that they're not creating risks with the technology.

Daan van Rossum: Taking all that into account, I can see why you said in a recent talk that this is the biggest change you've seen in your 18-year career at BCG.

It is a lot for companies. And this really goes to the DNA, to the fabric of the company. What kind of company are you? If you're thinking about it in the insurance industry, for example, you're rethinking so much of your process. You're rethinking so much of your customer touchpoint and your internal work. So, it sounds like it's very fundamental.

I know we're getting to the end of time. I'm really interested in any feedback from you personally. I know you also represent a company, but you're so deep into this AI game and you're talking to a lot of companies, and I think you're seeing that there's still so much opportunity for companies to actually be first and be ahead of others, unless maybe you're in insurance.

How are you looking at it now? Any contrarian views, just something you want to share on this topic? 

Matthew Kropp: Yeah, I think that the big thing that I would encourage all of your listeners to think about is not just focusing on cost. I think we came into this topic with the idea that this was going to be the robots coming for the jobs.

This was going to be about reducing labor costs. I think that comes from a couple of things. It comes from the media and comes from all of our stories about what AI is, the Matrix, Skynet, and all that. It also came because this technology emerged at a time when companies were very focused on cost. That is the topic that the enterprises are focused on now.

So we've come in with this mindset that says this is just about saving costs, and I think that misses probably the majority of the opportunity, and for sure, there are places where you can automate. For sure, if you have customer support and call centers, yes, you can automate a lot of that, and you're going to save a lot of costs. And yes, it's going to reduce a bunch of jobs, but in most other places, and it's actually, we have the research to back this up, that 80% of jobs are only going to be augmented by AI, meaning 10–30% of the activities in those jobs can be automated, but the rest, you still need humans for.

So it's really important not just to go looking for cost savings because you then end up saying, Yes, I can take 20% of this job out. But if I take 20% out of 5 people's jobs, it doesn't mean I can remove a person. There's all the other stuff that's being done.

But this technology can do things like massively speed up a process, massively improve the quality of the output, and allow you to do things that you couldn't even get to before. Maybe just to give a quick example, if we go back to the insurance topic, we worked with a commercial insurer on underwriting.

Underwriting is the core process in insurance. The initial thinking was, well, maybe if we automate underwriting, we can have fewer underwriters, and it's about cost savings. What ended up happening, though, was that when we looked at the process, about 80% of it was toil, with the underwriter having to read through hundreds of pages of documents and do a bunch of research. And when we automated that, we had the AI, because Gen AI is really good at reading long documents and providing really good summaries. So we automated the process of summarizing. It really slowed the time down.

And what that did was leave the underwriter, first of all, not going through the toil.

It let them focus on what they were good at, which was coming up, identifying the risk, understanding how to price the policy, and so forth. And it allowed them to write the policies in 3 hours instead of 3 or more days.

And so what this does is that it means that you're the first bid that your customer gets. So the win rate goes up in sales. And prior to this, they were not able to write 100% of the policies that were coming in because they didn't have enough capacity, but now they could. So the result was not fewer underwriters. We kept the same number of underwriters. The result was growth. They won more with a higher win rate on their proposals, and they wrote more policies.

I think we need to be thinking in this way. Yes, there's some cost savings, but how does this change? If you could take a huge amount of time out of a process, how would you work differently? How does that give you a competitive advantage? How does that change your offering?

Even further, we didn't have this technology two years ago. Can you reinvent your offering? Maybe your product and service are totally different. Maybe the way you interact with your customers is totally different. And so those are the opportunities; it's not about cost savings.

It's not about taking out work. It's not about robots coming from your job. It's about creating entirely new opportunities for companies.

Daan van Rossum: Yeah, I would say that the company would almost want to hire more underwriters because now they have such a great advantage that they can get that quote out to the customer sooner and win a lot more.

Perfect practical note to end on. Thanks so much for being on, Matt. It was a great pleasure to have you. 

Matthew Kropp: Great. Thank you so much.

🎧 Listen Now:

Welcome to the new Lead with AI podcast.

In this series, we speak to senior leaders responsible for rolling out AI in their organizations and uncover deeply valuable insights into what success looks like.

In today’s episode, we speak to Boston Consulting Group’s X global CTO, Matt Kropp, and hear about:

  • Why 70% of AI's value comes from people, not just the technology itself
  • How to successfully implement AI, including the importance of employee engagement, co-designing solutions, and continuous education through peer sharing. 
  • How AI can drive business growth beyond mere cost-cutting, ultimately creating better jobs for more people, and why focusing on 'joy at work' might be the key to unlocking AI's full potential.  
  • What factors you should consider when choosing which AI platforms to select for your teams and building your own AI technology?

Key Insights from Matthew Kropp

Here are the actionable key takeaways from the conversation:

1. Successfully Rolling Out AI Means Focusing on Joy.

To roll out AI successfully, you need a couple of things, according to Matt:

  • You need a strategy that answers where you can use the technology and what impact you’re trying to achieve. 
  • Look at the end-to-end process you’re trying to affect. What is the work? 
  • Identify where there is toil and automatable human effort in those steps, the work people don’t like to do. At the same time, identify where there is joy, and preserve that. Use AI to create better jobs for humans, which will boost the quality of work, job satisfaction, and employee engagement and retention.

2. AI is a Change Effort.

Matt shared that BCG talks about the idea of 10-20-70 when building technology.

10% of the effort and 10% of the value come from the algorithm, 20% from the software, and 70% of the value and effort come from driving change in the organization.

The true value of AI implementation lies not in the technology itself but in organizational change.

When implementing AI, allocate significant resources to change management and employee adoption strategies, not just technical development.

3. Co-Design and Educate to Get Everyone on the AI Train.

In the example Matt mentioned, only 15% of call center workers voluntarily adopted a new AI solution. This makes sense because they knew the risk of being displaced.

Among developers, some switched on the AI only one day per week, because they didn’t know what else it could do for them.

Matt says that, therefore, the key is to co-design solutions with the people using them. Involving employees in the AI implementation leads to better adoption and more effective solutions.

Create cross-functional teams that include end-users when designing and implementing AI solutions to ensure they meet real needs and are readily adopted.

Then, ensure education about the technology's benefits. This is best done through community engagement like ambassadors who can support others, peer sharing through for example hackathons, and nudges that create the habit of using AI.

4. Selecting your LLM Starts with What Productivity Suite and Cloud Provider You Have.

With continuous product upgrades across the major LLMs like ChatGPT, Google Gemini, and Microsoft Copilot, choosing your company’s LLM isn’t easy.

Matt advises separating the questions about which tools you’ll get for your teams and what you need to build your own AI applications.

For the employee tools question, consider what productivity software you’re using and paying for. For example, Office 365 Copilot has a big advantage for companies using the Microsoft Suite for work, as it can train on that data.

For development, similarly, choose based on which cloud you’re on and which LLM is best integrated with that cloud.

5. Matt’s final point is to not just focus on costs.

Yes, we’ve heard a lot about AI being a cost saver, but also consider the tremendous opportunity.

As Matt says, 80% of jobs will be augmented by AI. For the rest, we still need humans. So, create a better job for them, even allowing them to do things they could never have done before.

For example, in the insurance industry, underwriters can now be freed from doing a lot of data input work and write policies in three hours instead of three days, significantly increasing their win rate.

Can you use AI to reinvent your offering?

🔔 Available on:

Transcript:

Daan van Rossum: I know you work in BCG and then specifically in BCG X. Can you tell us a little bit about yourself, the company, and what you're doing in the realm of AI? 

Matthew Kropp: Yeah, absolutely. I'm the CTO for BCG X, and everybody's heard of BCG, the management consulting firm. You may not have heard of BCG X, a fairly new brand, but we have an AI build division, which is BCG X, which has 3000 engineers and data scientists. We are building AI solutions and generating AI solutions as part of our work with clients. So I'm running the general AI topic, and I'm heading up the whole engineering organization there within X.

Daan van Rossum: That sounds like a lot of people—3000. So give us an idea of what those 3000 people do.

Matthew Kropp: It's sizable. BCG is 30,000. So it's actually only a fraction of the overall BCG company. So, we are a development shop, but quite different from others. You have your systems integrators out there, like Accenture, Deloitte, etc. We don't really compete with them. Really, what we're doing is building solutions that are tied to business strategy.

So when we're working with clients, we'll have both the BCG consulting team and the BCG X development team, and we're focused on how we can make an impact. What is the right strategy? What are the right solutions to put in place? Our BCG X team is building those solutions. And then our BCG consulting team is helping our clients actually get value out of them. We may want to talk about that.

We talk about this idea of 10, 20, and 70: when you build technology, 10% of the effort and 10% of the value come from the algorithm, 20% of the effort and value come from the software, the data, and the connections into other systems, and 70% of the value and effort really come from driving change in the organization.

And so that pairing, BCG plus BCG X, is really that pairing of BCG X giving you the 10 and the 20, the algorithms and the software, and BCG giving you the 70, which is the right strategy and the right change efforts to actually get the technology to create value in the organization.

Daan van Rossum: That's something that maybe people intuitively don't really get—that this is not actually a technology issue. It's mostly a change issue because there's been technology for quite a while that can automate; obviously, now things can be done on a much broader scale and in a much faster and smarter way.

The change part of it is the part that I think a lot of people are very interested in. So a lot of people that we speak to, a lot of people who are in the community, see the opportunity in AI, and maybe they have some great ideas about how to implement AI. And then, they're looking into their organization, and they're realizing that it's going to take a lot more than just launching a software product to actually get people to embed AI in their day-to-day lives.

What are some of the principles that you're either practicing internally or also sharing with clients when it comes to AI in the workplace? 

Matthew Kropp: I totally agree with your framing. AI is not new. We've had AI for at least a decade, maybe more. We've called it different things, and it has consistently been as you described. Just implementing technology or writing the code doesn't have an impact. You actually have to get the change to happen in the organization to get an impact out of it.

This new wave, so-called generative AI, is the latest evolution of AI. I would say is even more so, and the reason is because, if you think about a lot of the AI solutions that companies have been building in the past, they're really narrow solutions tailored to a specific task. I need a model that predicts something. I need to predict the next best offer to show to a customer.

I need to predict what my demand is going to be. And so, that's a very narrow application. Generative AI is a general-purpose tool, and it affects how humans work. It's something that empowers humans. So even more than in the past, you have to get the people who are working with the tools to work in a different way so that you get value from it.

We're in this new evolution. Generative AI has really only been commercially available for about a year. Yes, it's been around for a few years before that, but really, enterprises have only started being able to build solutions, starting early last year.

So we are very early in the learning curve, and so I've been out talking. I've spoken to over 200 companies in the past year, to CEOs, boards, and executive teams. And most of last year, they were really just focused on learning. What is this technology? How do I use it? They were experimenting.

They were building POCs and chatbots to do various things in their organization and not getting much or any actual value out of them. Now companies are thinking about how to get value and how to make an impact. And so they're doing a couple of things that I think are really important. The first is really having a strategy, identifying where you can use the technology, and determining what impact I'm trying to achieve.

We might want to talk about this, but the impact could be that I want cost savings. I want to reduce the amount of labor that's in a particular activity, or it might be that I want to increase the quality of people's work, or I want to drive revenue growth, or I want to change your customer experience.

There are all sorts of things that you might get out of this. But starting with the idea of what the objective is that I'm trying to solve, rather than building a chatbot and giving it to my employees, is critical. So start with the objective.

Then we think it's really important that you look at the end-to-end process that you're trying to affect. What are people doing? What is the work? If I'm an insurance agent and I'm doing underwriting, there's a whole set of steps that have to take place. I need to go get some documents. I need to read lots of reports. I need to do some research on some databases. And then finally, I get to making a decision, coming up with a price, writing a policy, etc.

What are all of those steps that are involved in that process? And then identifying where in those steps there is toil, a lot of human effort, and a lot of human work that can be automated.

Also, that may have a low value. That's not very fulfilling, which people don't necessarily like to do, and identifying, where is their joy? Where do people really get fulfillment from their work, or where do you actually want humans to be involved because they bring diversity of ideas, creativity, risk management, relationships, building, and customer experience? There are all sorts of reasons why you don't want to take people out of processes as well.

And so thinking very holistically about how to redesign the end-to-end process with AI and with AI working with humans is how we think we really need to be working in order to get not only efficiency and productivity but also better quality and, frankly, a better job for the humans that are involved in working on that process.

Daan van Rossum: And so you're actually practicing the parts that we've only been hearing in theory, which is that AI can take over the manual, repetitive part of your job—the part of your job that you probably don't really like all that much—so that you can focus on the stuff that you actually like.

When you're going into companies, is that a message that resonates when you're saying, Look, you've got to look at the entire workflow and you've got to look at, like, where are people just dreading to come in because they have to do data entry or because they have to go through manual processes or because they have to click on a bunch of buttons in an application? And where are people coming to work and really showing up for what they're uniquely good at and what they like doing? 

Is that something that resonates when you talk to companies? Obviously, when you talk to individuals, it resonates deeply. How does that go when you go into a company and say something like that?

Matthew Kropp: No, it absolutely resonates. And I would say it's been interesting, especially when I'm talking to senior leaders and boards. This was a question that everybody had. They're very excited about the technology. The technology is incredibly powerful, and in some ways, it is mind-blowing what you can do with it, but they very quickly came to the question of what this means for my people.

I think maybe we have expectations that senior leaders just care about efficiency, but they really care quite a lot about their employees and about engagement—about their employees actually being engaged and really bringing all of their capabilities to work.

And so there was a lot of anxiety about this idea: if I'm introducing this automation, am I creating an environment? We're actually going to hurt employee engagement. So, they've been very receptive to this idea, and I bring up this idea of minimizing toil in people's work and maximizing joy as the formula. Every time I talk about this, the executives’ eyes light up, and it clicks a switch. I can see how this is actually not a negative for my people, how this is actually a net positive, while I still get to take advantage of this capability.

We've done some research on this. It initially came out of COVID, work from home, and hybrid work. So, we have a think tank, the Bruce Henderson Institute, that did a bunch of research around employee engagement. So what is it that keeps people engaged, as you said, wanting to come to work and bringing all of their capabilities to work?

There are a bunch of findings, but the part that I find most interesting for the AI and generative AI topic is that when you do things that improve productivity, it basically doesn't affect employee engagement. Small 5%–3% improvement in engagement. When you do things that improve joy at work, it has a big impact on engagement, and I always joke. Maybe that's kind of obvious. If I make you happier at work, you're going to be happier at work. All of the discussions about AI have tended to revolve around productivity.

Essentially, we're saying, Hey, we're going to bring you tools that make you more productive, and what people hear is that the robots are coming from my job. And I actually give an example of this. We're working with a tech company that has implemented a chatbot to support their customer support reps. So their call center reps, and it was supposed to make them more productive.

They were very surprised that only 15% of their reps had adopted the tool, and they're asking the question of why my people are not adopting the tool. It's pretty obvious. They know exactly why you're implementing this tool. It's to reduce their jobs. So we have to take this different frame, which is that there’s a lot of work that is not value-added.

In typical jobs, up to 25% of people's work is just administrative tasks that they don't enjoy doing and that you can automate. And so we really have an opportunity to change the way people work.

Daan van Rossum: Yes. And I had a great episode with Debbie Lovich where we went very deep into that research, and it just made my day because I think it's really what this is all about. If we are going to intervene in the way that people work, let's optimize it for joy, not for being more productive. Of course, we want to be productive as well, but those two things come together quite nicely. 

So then, at an executive level, you're like telling the story, and their eyes brighten and their ears _____ that we can use AI to make people happier at work, and therefore, they will stay with you longer and they'll enjoy their job more.

How do you then get to that adoption level? Because it may sound great in theory. And then, like you said, you build a chatbot to support call center workers. And 85% say so no thanks. I'm not going to dig my own grave. 

So what are some steps that you recommend those companies take to actually go from that initial idea to people truly using it and feeling that it's actually benefiting them?

Matthew Kropp: The key is co-designing the solution with the people who are going to be using it. And so maybe I'll give an example. One of the areas I think is quite exciting and moving quickly because it's a bit more mature is software development. Generative AI tools can support coding in pretty profound ways. GitHub Copilot is one example of a product, but there are others out there that happen to be the ones that have the most market share at the moment. So we're doing a bunch of work with clients that are implementing GitHub Copilot and generative AI solutions in their software development life cycle, and it's been quite interesting.

So, we are working with a bank that has rolled this out to several thousand of their engineers, and so we were talking to the engineers about it and trying to understand their maturity journey and their learning journey to using these tools.

And what was interesting is that, in the first place, they were very excited about the tools. They said, This is great. Most of them were able to come up with examples where they said this can do things for me that I don't like to do. And so, I love it. I hate writing unit tests, but it's really good at writing unit tests. I hate writing documentation. It's really good at writing documentation.

But at the same time, most of them were not using the tool to its full potential. And so they had all self-discovered one part of their work that it was helping them with. And they were very enthusiastic about it, but they didn't see the other opportunities. My favorite example was that one of the engineers was very passionate about it. He said, I love it. I use it for writing my unit tests. But I only write unit tests on Fridays. So Monday through Thursday, I turn the tool off so that it's not in my way. And then on Friday, I turn it on and use it for my unit tests.

And then we asked the question, Well, do you think that four-fifths of the time you might also be able to get some benefit from other things that you do? And so the key learning from this was that you have to, first of all, involve people so that they feel ownership of the process. You have to give them some basic education so that they understand how it works and how it can be part of their process.

But then you have to do a lot around community engagement, peer sharing, nudges, reminders, and getting people to really be in the habit because we're not in the habit of using generative AI as part of our work. We need to get into the habit. And so a lot of this is both a reminder and seeing what my colleague is doing and hearing from my colleagues. So what we did with this client was set up a bunch of hackathons where they worked on a specific problem, but they collaborated, and we heard a lot coming out of the hackathons where one of the engineers would have some technique that they had developed and the others would say, Oh, that's great. That could work for me. And so they're sharing; they set up ambassadors. So basically, superusers who were on call spoke for others so that they could share how they did things.

So I think it's really going to be incumbent on organizations, not just to say, Here's a tool or here's the onboarding training." It really has this continuous sort of community of practice so that the people that are using the tools start to reinforce one another.

Daan van Rossum: So you're really mapping out an entire journey, which actually starts way before the rollout, which is to say, you're going to get people involved in designing maybe what tool gets used and then how that tool will be used. 

And then when it actually launches, we have a lot of touch points to make sure that, for example, the people who feel very comfortable with it, and maybe who are most enthusiastic about it, that they share some inspiration on, Oh, here's how you can also use it. And if you need any help, you can approach me. Hopefully, companies will free up time for people to do that and make sure that those ambassadors can be reached. 

What do you deal with in that process? Do you deal with people who have already adopted their own tools? So we saw from the Microsoft research the whole BYO AI thing.

People are bringing their own AI tools to work. Do you now sometimes have to retrain? Let people unlearn maybe other tools, or maybe, in the case of a Copilot for GitHub, at least some people were probably already using that. 

So in that discovery journey, do you do any surveys around what you are currently using? What are you most open to? What kind of pain points would you like to solve most? What does that part of the journey look like? 

Matthew Kropp: So I think we're pretty early in this evolution. So, I don't know that I would yet say that there are a bunch of bad habits that people need to unlearn because I think they're still learning just the first set of habits.

But I think there's something really interesting in what you're talking about, which is that we have typically had software that was, as I was saying earlier, purpose-built, that does one thing. And this is the general purpose: ChatGPT Enterprise, Bing Chat, or Google Gemini, or choose your tool. It knows about everything. It is able to do a wide range of work. If I know how to ask the right questions, if I know how to interact with it in the right way, we will certainly have very purpose-built tools that incorporate Gen AI, and that'll be.

First of all, all of our software is going to get Gen AI as part of it because every software company is busy building it into their tools. Enterprises will build purpose-built tools that bring Gen AI to different processes, but then you're going to have the ChatGPT enterprise, kinds of tools where your employees will figure out how to automate their own work.

And I think this is actually probably the place that is going to have the biggest impact, but it's going to take the most time because we all have to learn how to work differently. So at BCG, we're going through this evolution right now. We were very early to roll out ChatGPT Enterprise to all of our employees.

So all 30,000 employees have access, and we've been doing education and telling people how they can use it and what they shouldn't do with it. But really, the value comes from our individual consultants. Figuring out for themselves the work that they are doing at the moment, how can I use this to help my work?

And so, all of us humans are going through this joint discovery process of figuring out how this changes the way that we work. And it's going to take some time for us. Frankly, the tools are evolving. We will learn how to use them more effectively. We will learn when we should use them and when we shouldn't use them. That's got to be a big part of the motion that organizations go through as they bring these tools into their day-to-day work.

Daan van Rossum: Yeah, it must be a very strange time for L&D teams, for example, who have always been used to the idea that you can study something and then teach it to people, but there's no studying because you're releasing something that you also don't really know how it works, how it will work, or what the use cases are. 

I think what you said, Matt, about the way to roll it out, especially with those enterprise-wide main LLM tools, is what we saw in the Moderna case study and what we saw in the Microsoft HR case study, which is that everyone has access. You get a bit of base education, and then you go figure out where it fits in your workflow the most, what parts of your job you can now automate, and what tasks you can remove from your job. Because, as Josh Bersin said a couple of months ago, frontline workers typically know the work better than management and, typically, better than leadership. 

So in that process in your company, have you adopted certain rituals or behaviors again to let people share those use cases? So that again, if someone discovers something that is really great in their own workflow, how does it extend to their team and then maybe department geographies because you're global as well? What are some practices you have in place to make that happen? 

Matthew Kropp: Yeah, I think it's pretty important. Although we need to develop other techniques as well, we are doing what we're calling discovery sprints. And so we have set up identified teams. So this is just project teams, client projects that we're working on, but specific teams where we have made sure that they understand how to use the tools, but we're really following what they're doing and using them as the innovators; they're figuring out how to use these tools within their work, and then we're capturing that, surfacing that, and then broadcasting that out to the overall company.

So I think we're going to have to do a lot of things like that, like my example with the software engineers where they're doing hackathons, this idea of the power users with the ambassadors, it's going to be a whole host of those kinds of ways of both identifying or observing what people are doing, get catching those new techniques essentially, and then broadcasting them back out to the overall company.

Daan van Rossum: And then one other related question that I had, and you're okay to say, I don't want to answer this question, but I'm very curious because it comes up a lot in the conversations I'm having, which is that companies that are, you're one year ahead of them. So you've already decided to go with ChatGPT Enterprise, which I'm assuming was one year ago. You must've almost called up Sam Altman and said I want this for my whole company, because it wasn't quite a ready-made proposition as it is now. 

Companies that are still looking into it—maybe they're already on Microsoft, like most larger companies—should I go with Microsoft Copilot? Should I go with ChatGPT Enterprise? Are you in those kinds of discussions where companies are trying to decide which one to use? Then, obviously, you cannot give any particular advice, but what are some things that companies should be looking at as they're looking at what platforms to adopt? 

Matthew Kropp: Yeah, the platform selection question comes up all the time, and I would break it into a couple of parts. So there is the deployment of tools for your employees. So that's the ChatGPT enterprise, or Office 365 copilot, or Google's Gemini is part of GSuite, and so forth. I think I would make that choice based on, first of all, what productivity software are you already using?

What is your stance? What is your InfoSec team’s stance on? What tools are they going to allow? I think Office 365 Copilot has some advantages because of the tight integration with Office. I think what we're seeing right now is that some companies are adopting both Office Copilot and ChatGPT Enterprise, and they're letting their users pick when to use which.

Daan van Rossum: That's a nice monthly bill. Okay.

Matthew Kropp: Yeah, exactly. Presumably they will rationalize that, but then there's also the choice of what models you're using. What platforms are you using for development? And I think there's actually been more of our client conversations around that choice.

And I would say that, of course, this is all evolving very fast, so it's changing. But the conversation so far has been first around: What cloud are you on? Generally, companies have standardized on one of the three hyperscalers as their core cloud. And then, are you comfortable with the foundation of the frontier model available in that cloud?

So OpenAI is for Azure. Anthropic as being the most advanced model on AWS and then Gemini for GCP, or do you want to be able to use multiple models, and then are you comfortable with the security posture of not just using the model that's within your hyperscaler?

So that there's this question of where does my data sit? Where is my infosec team comfortable in terms of my cloud posture? What are the models I want to use? And then what do I want to use those models for? And so do I need the latest frontier model? Do I need GPT for instruction following capability for everything? Or do I have some needs where the requirements are less and I can use a cheaper model? Or maybe even I have very high volume and I need to have low costs. And so I actually want to operate Llama 3 in my own cloud because I want to control both the costs and the perimeter around the data. So those are the kinds of conversations that the clients are having.

Daan van Rossum: Where are the most people? So you obviously speak to a breadth of clients and a lot of different companies. Again, different geographies. Is there any consistency, like most companies are at this stage where, again, they're looking at custom development versus companies that are still looking at what tool to even use? 

Where do you see companies, and what's their status? And perhaps linked to that, what are some misconceptions out there as you're speaking to these companies? What are some misconceptions you're still seeing when it comes to Gen AI? 

Matthew Kropp: Yeah, it's a wide range. There are still companies that are taking a very cautious posture. In fact, I had a recent conversation where they had not yet allowed their people to use ChatGPT. And we had the conversation that you realize that your employees are using it. They're just using it at home for their own personal accounts. And so maybe you watch, and you want to bring it into your enterprise. So you have some control. So there's a range. There are companies that are still very early.

And that's more industry. If you're an industrial goods company, you're not feeling a lot of urgency to adopt this. If you're an insurance company, this feels very existential because your business is entirely text. Your business is a contract. And so the whole insurance industry has really leaned into this technology. I think they see it as both a competitive advantage and a big threat if they don't adopt it. So I would say it's a range, but you end up with choices around how much they deploy with their employees. So, what are the general-purpose tools, and then how much are they investing in building custom applications?

The leading companies are the ones that have really identified this as a key technology that they need to have as part of their organization. They're leaning in by defining: What are the key areas that they want to focus on? So what are the big rocks or the big moon shots that they want to develop?

How do they empower the whole organization to come up with ideas and give them tools so that you can have smaller departments that are coming up with their own automation or even individuals, as we were talking about figuring out how to innovate on their own process?

And then they're putting in place the right platform in terms of data, the right platform in terms of technology partners and models, putting in place a center of excellence in the engineering functions so that they can support building out these applications, and putting in place the right governance so that they're picking the right bets.

Investing in the right places, tracking that they're actually getting impact from those investments, and making sure that they're building responsible AI into everything that they're doing means that they're not creating risks with the technology.

Daan van Rossum: Taking all that into account, I can see why you said in a recent talk that this is the biggest change you've seen in your 18-year career at BCG.

It is a lot for companies. And this really goes to the DNA, to the fabric of the company. What kind of company are you? If you're thinking about it in the insurance industry, for example, you're rethinking so much of your process. You're rethinking so much of your customer touchpoint and your internal work. So, it sounds like it's very fundamental.

I know we're getting to the end of time. I'm really interested in any feedback from you personally. I know you also represent a company, but you're so deep into this AI game and you're talking to a lot of companies, and I think you're seeing that there's still so much opportunity for companies to actually be first and be ahead of others, unless maybe you're in insurance.

How are you looking at it now? Any contrarian views, just something you want to share on this topic? 

Matthew Kropp: Yeah, I think that the big thing that I would encourage all of your listeners to think about is not just focusing on cost. I think we came into this topic with the idea that this was going to be the robots coming for the jobs.

This was going to be about reducing labor costs. I think that comes from a couple of things. It comes from the media and comes from all of our stories about what AI is, the Matrix, Skynet, and all that. It also came because this technology emerged at a time when companies were very focused on cost. That is the topic that the enterprises are focused on now.

So we've come in with this mindset that says this is just about saving costs, and I think that misses probably the majority of the opportunity, and for sure, there are places where you can automate. For sure, if you have customer support and call centers, yes, you can automate a lot of that, and you're going to save a lot of costs. And yes, it's going to reduce a bunch of jobs, but in most other places, and it's actually, we have the research to back this up, that 80% of jobs are only going to be augmented by AI, meaning 10–30% of the activities in those jobs can be automated, but the rest, you still need humans for.

So it's really important not just to go looking for cost savings because you then end up saying, Yes, I can take 20% of this job out. But if I take 20% out of 5 people's jobs, it doesn't mean I can remove a person. There's all the other stuff that's being done.

But this technology can do things like massively speed up a process, massively improve the quality of the output, and allow you to do things that you couldn't even get to before. Maybe just to give a quick example, if we go back to the insurance topic, we worked with a commercial insurer on underwriting.

Underwriting is the core process in insurance. The initial thinking was, well, maybe if we automate underwriting, we can have fewer underwriters, and it's about cost savings. What ended up happening, though, was that when we looked at the process, about 80% of it was toil, with the underwriter having to read through hundreds of pages of documents and do a bunch of research. And when we automated that, we had the AI, because Gen AI is really good at reading long documents and providing really good summaries. So we automated the process of summarizing. It really slowed the time down.

And what that did was leave the underwriter, first of all, not going through the toil.

It let them focus on what they were good at, which was coming up, identifying the risk, understanding how to price the policy, and so forth. And it allowed them to write the policies in 3 hours instead of 3 or more days.

And so what this does is that it means that you're the first bid that your customer gets. So the win rate goes up in sales. And prior to this, they were not able to write 100% of the policies that were coming in because they didn't have enough capacity, but now they could. So the result was not fewer underwriters. We kept the same number of underwriters. The result was growth. They won more with a higher win rate on their proposals, and they wrote more policies.

I think we need to be thinking in this way. Yes, there's some cost savings, but how does this change? If you could take a huge amount of time out of a process, how would you work differently? How does that give you a competitive advantage? How does that change your offering?

Even further, we didn't have this technology two years ago. Can you reinvent your offering? Maybe your product and service are totally different. Maybe the way you interact with your customers is totally different. And so those are the opportunities; it's not about cost savings.

It's not about taking out work. It's not about robots coming from your job. It's about creating entirely new opportunities for companies.

Daan van Rossum: Yeah, I would say that the company would almost want to hire more underwriters because now they have such a great advantage that they can get that quote out to the customer sooner and win a lot more.

Perfect practical note to end on. Thanks so much for being on, Matt. It was a great pleasure to have you. 

Matthew Kropp: Great. Thank you so much.

FlexOS | Future Work

Weekly Insights about the Future of Work

The world of work is changing faster than the time we have to understand it.
Sign up for my weekly newsletter for an easy-to-digest breakdown of the biggest stories.

Join over 42,000 people-centric, future-forward senior leaders at companies like Apple, Amazon, Gallup, HBR, Atlassian, Microsoft, Google, and more.

Unsubscribe anytime. No spam guaranteed.
FlexOS - Stay Ahead - Logo SVG

Stay Ahead in the Future of Work

Get AI-powered tips and tools in your inbox to work smarter, not harder.

Get the insider scoop to increase productivity, streamline workflows, and stay ahead of trends shaping the future of work.

Join over 42,000 people-centric, future-forward senior leaders at companies like Apple, Amazon, Gallup, HBR, Atlassian, Microsoft, Google, and more.

Unsubscribe anytime. No spam guaranteed.