The AI-driven company: a maturity model

After discussing the challenges that companies seeking to become AI-driven experience, the aim for the second part of this series is to explore how companies evolve through the stages of transitioning from operating traditionally to becoming fully AI-driven. The best way to present our findings is in the form of maturity models, where we identify the stages companies move through, based on the results of our interview study. We’ll discuss three such models: the first is at the organizational level, the second is concerned with AI in the R&D process and the third is concerned with AI in products.

When focusing on the organizational level, we found that companies tend to evolve through several stages, which we capture in five steps (see the figure). As companies start to explore what AI and especially LLMs and agentic AI can do for them, the first step is what we refer to as “playtime.” Here, individuals start to use AI for personal productivity. This means that the tasks employees used to conduct by hand are now supported or fully automated by AI. Examples include generating response emails, automatically generating documents or summarizing material that was sent to them. The focus is squarely on personal productivity.

In the second stage, which we refer to as “automation 2.0,” business process owners seek to use AI to automate individual steps in their process that they haven’t been able to automate with traditional approaches. It often means that some level of pattern matching and human-interpretable generation is required that couldn’t be achieved with traditional means. Examples mentioned in our interview study include candidate selection and prioritization in HR based on application letters and CVs, transaction classification in finance and product planning tasks. Although not all tasks are completely automated, AI can provide significant help and reduce the amount of human effort by an order of magnitude or more.

It’s in the third step, “local AI-first,” where things become more disruptive. Here, the AI experts and a business process owner jointly engage in a zero-based thinking approach to reinventing the process in an AI-first fashion. Rather than organizing it around the limitations humans exhibit in their information processing capabilities, the business process is organized around the limitations of AI. This means that process steps or phases may be merged as an AI agent, which can easily combine multiple disciplines and quickly access all relevant information. Also, some steps may not even be required and can be removed entirely, such as certain quality assurance steps. Finally, of course, there are steps where the AI is unable to conduct everything independently or at all and this is where there are tasks that are assigned to humans.

In this context, the roles of humans tend to fall into two categories: supervisory and executing. In the supervisory case, the AI generates output that needs to be reviewed and, where necessary, changed before it can be sent forward in the process. In the executing case, it often means manipulating objects in the real world, such as packing orders in a warehouse or driving a vehicle from one location to the next.

Once the first processes have been reinvented from an AI-first perspective, the next stage is concerned with super-agents. That sounds perhaps ominous, but the intent is to connect earlier disconnected business processes to achieve optimization across the board.

A typical example is between the R&D department and the purchasing organization. In most companies, R&D is in the lead, conducts product design and tells purchasing which design components need to be acquired from suppliers. In this case, a super-agent could develop a back-and-forth loop between R&D and purchasing where it suggests adjustments in the product design to lower cost and increase the number of parts that can be sourced externally.

Although this activity of going back and forth between R&D and purchasing occasionally happens in companies, it’s a highly manual process that’s very effort-consuming as it goes over organizational boundaries and involves individuals with quite different incentives. Using AI agents, this will become systematic, continuous and fast. As a result, outcomes can be improved significantly.

The final stage, the AI-first organization, is our definition of Nirvana, where every activity in the company has been redesigned from an AI-first perspective. Here, every activity that can be conducted using AI has been identified and realized. Also, the dependencies between activities and processes are managed dynamically and a potentially large set of agents ensures global optimization of processes.

In the context of AI-driven companies, we can distinguish between “the algorithm” and humans. The algorithm, in this context, refers to the layer of automation and AI that forms the key value in the company. However, there will still be humans involved in the business. These humans can be organized into those who work for “the algorithm” and those who work on it. Those who work for the algorithm are like Uber drivers in that the automation and AI functionality will tell them what to do and how to do it and will give continuous feedback on performance. Humans working on the algorithm are concerned with continuously finding improvements and implementing changes in it. Although AI has many strengths, innovation and creativity tend to fall more in the human realm and this is where we’ll continue to rely on humans for at least a while longer.

We’ve presented a maturity model for how companies evolve and develop through five stages: playtime, automation 2.0, local AI-first, super-agents and the AI-driven company. Many of the companies we interviewed are in the first two stages; some are climbing to stage three and are exploring stage four. To end with a quote from Fei-Fei Li: “AI is everywhere. It’s not that big, scary thing in the future. AI is here with us.”