The next phase of Industrial AI

Matt Kempson, Chief Operating Officer, AI, at IFS, discusses how the company is scaling Industrial AI adoption, the role of enterprise data and agentic capabilities in improving operational performance, and why customer and partner engagement is accelerating the transition from experimentation to deployment.

IFS has introduced several AI capabilities over the past year, including digital workers and IFS Loops. How well positioned is the company to accelerate Industrial AI deployments among existing and new customers?

We’re in a genuinely strong position. We’ve got a bedrock of fantastic data and context, and we’ve taken a very deliberate approach to controls that allow us to scale over time. We’re prepared to disrupt ourselves and invest in different technologies and different ways of building products. But the most important thing is that our customers and partners actively want to work with us on these use cases. We can move very quickly from an idea or research into market testing and building solutions with clients. That’s why, a year on, you’re seeing dozens of capabilities that didn’t exist a year ago, on top of the dozens we launched last year.

 What foundations have enabled the company to build its current Industrial AI capabilities?

There are three advantages that we have. The first is that we’ve been working with machine learning, predictive technologies, and simulation for more than a decade. That builds the skills and knowledge around emerging technologies, which means that when something new comes along, we’re really well positioned. We’ve got the people, and we can attract people who want to work in this kind of environment.

Another important advantage is that businesses trust us with really complex industrial operational data that they wouldn’t trust anywhere outside a highly controlled, secure environment. And then there’s the context we’ve built up over 30 to 40 years. That’s the jet fuel for AI that’s actually going to take off. So we’re very well placed.

When did IFS begin investing in agentic AI, and how much of its current capability has been developed internally versus through acquisitions?

Our existing AI team was already extending beyond machine learning and predictive capabilities, so the investments have been ongoing. Agentic AI has really been around for four to four and a half years, and we’ve been investing throughout that period. What happened about two years ago was that the market started accelerating to such an extent that we had to think about how we wanted to build our capabilities. Do we acquire? Do we acqui-hire? Do we train our own people? Do we hire new people? In the end, we decided to do all of the above.

We acquired Loops and acqui-hired the teams that predominantly made up Nexus Black, although we also brought IFS people into those teams. So it’s been a journey 35 to 40 years in the making, with AI over the last decade and agentic AI really over the last three years. This time last year, we already had dozens of customers using the technology. Now it’s hundreds. So we’re in a good place.

How are tools such as Claude and IFS’s partnership with Anthropic changing the way your teams develop AI solutions?

On a practical level, they help. We give our developers access to the best technologies available in the market. We use Anthropic’s technology in our development processes, and we have the option to embed its models in any of our solutions. We’re not exclusive with any model provider, for several reasons. The primary one is that we need the independence to drive the technology we think will have the biggest impact. We also build our own intellectual property, so we’re not going to be locked into anything.

That’s the practical development and engineering side. But the bigger aspect is the human psychology around AI. If people are using AI in their day-to-day lives, they’re less afraid of it and more likely to adopt it. For example, I connected the latest release of ChatGPT to my personal Gmail, and within 30 seconds it told me I had two health insurance policies that I would have completely missed. As a new parent with a young baby, these are things that can easily get overlooked. Every time somebody has that moment where AI saves them hundreds of pounds or dollars, they become more willing to accept the technology. I think that’s one of the biggest changes over the last year. Market forces are increasingly dictating that companies using AI will be more successful than those that don’t. Equally, employees who embrace the technology and find ways to use it will have more enriching, engaging roles and become more valuable to their organisations. Companies that fail to adapt will lose market share or grow more slowly. Those that adopt it successfully will grow faster and hire more people. So this narrative that AI simply means fewer jobs is a very binary way of thinking about it. The real question is: what do those jobs turn into?

IFS has highlighted the considerable gap between current industrial productivity and its full potential. What prevents businesses from operating closer to that potential, and how can Industrial AI help?

You have to understand why businesses are operating at around 60% of their productivity potential. In fact, some of the better-performing businesses are at 60%. I come from a manufacturing background, so think about the supply chain. Are you running a just-in-time supply chain where there’s no latency whatsoever and everything moves straight through? Of course not. You’ve got to plan for bad weather, traffic, and material shortages. So every company builds in a buffer to allow for that uncertainty. Then you get into the warehouse, where you’ve got another set of dependencies, so you build in another buffer. On the production floor, you know your machines can’t run 24/7, with a few exceptions. There’s maintenance, planned and unplanned downtime, and worker shortages. So again, you build in a buffer. The same applies to product design and engineering.

Every part of the business has these built-in buffers because of people, technology, processes, and wider disruptions. You need them to create predictability and give the business a degree of control. With AI, you have the ability to cut through a number of these constraints. Whether you achieve a 10% or 20% improvement in one area, the point is that these improvements compound. A supply-chain disruption has a direct impact on production. You can’t manufacture something if the raw materials haven’t arrived. Too many businesses have siloed different types of data and context. Industrial AI allows them to draw inferences, make decisions, and delegate tasks to digital workers in real time, rather than relying on people in back offices to constantly reconcile all that information.

But some operational uncertainties, such as transportation delays, will remain. How much of that disruption can AI realistically address?

There are many dependencies, but some of them are controllable. If you’re already using traffic data in your operations planning, if you’ve got scheduling and optimisation of drivers, if you have visibility into adverse weather conditions and shipping schedules, you can plan well ahead of the moment a lorry is actually late.

You can’t necessarily control whether that lorry is going to be late. But you can control how far in advance you recognise that the delay could become disruptive. The alternative is to hold more stock, which means less cash available to your business. The more control you have, the more cash you can unlock.

How does IFS use the operational context already available within customers’ existing environments, and can that intelligence extend to third-party systems?

Depending on what the customer is using, we’ve got the contextual data across finance, supply chain, maintenance, service and operations. What intelligence allows us to do is use that information more effectively, while drilling down in a safe and secure way at both the user and agent levels.

It also allows us to extend into third-party data and other sources of context that enrich the experience. For example, IFS isn’t running shipping schedules within our solution, but there are shipping-schedule solutions and even open-source information that we can draw from. That additional context helps us make the intelligence much more useful.

All the ambition we showed last year with those early-adopter and pioneer customers has come good. We’re seeing really strong demand. The thing that excites me most, and this might sound a little strange, is actually the excitement among our partners. AI is an ecosystem game. To do it well, you need multiple companies bringing different perspectives together to achieve outcomes.

You never quite know how your ecosystem is going to react to really fast-moving technology. Typically, when new technologies emerge, partners tend to be cautious, questioning how they will work and where the commercial opportunities lie. This time, however, there’s a much clearer understanding of the potential. We’re also seeing strong enthusiasm from customers. The conversation has shifted from why organisations should adopt AI to how they can implement it. That’s a significant change from what we experienced a year ago.

Where do you see Industrial AI heading next, particularly in terms of the operational challenges it can address?

There are some big, complex problems that are now being solved. Supply-chain resilience is one that immediately comes to mind. Another major development is the consolidation of rich, trusted data from different sources within controlled environments. Bringing together operational intelligence from IFS solutions with design and engineering data could be a game-changer, giving businesses a more comprehensive view of their operations.

Imagine an executive who can now plan and run that whole business with access to that combined intelligence. That’s a game-changer. I think the businesses that really crack this are going to be more profitable. Those that lag behind will become very motivated to catch up. We’re already seeing that demand. There are thousands of people here for a reason.

You’ve mentioned the strong response from partners, but what about actual deployments, investment commitments and customers’ readiness to invest in these solutions?

The investment is significant, and it is a friction point. Whenever we’re bringing a new product to market, which is happening all the time, we have to invest in those first two, three or five customers to make sure it works. And not everything we do is going to work.

Sometimes we reach a point with a customer where we conclude that something isn’t commercially viable, or there’s an easier way of doing it that doesn’t necessarily require agentic AI or Industrial AI. And that’s okay. If we’re open with customers, we build trust over time. If something isn’t viable for long-term investment, it’s better to validate that early. AI tooling is also helping us make those decisions much faster.