Business units that choose their own AI tools often skip the innovation team’s proposals, until they see a hard number on what they can save.
Philips invests €1.7 billion a year in R&D and €1 billion in AI. In its last reorganization, the company moved most decision power from the center to its business units. Since then, each unit manages its own R&D budget and picks its own tools.
Rob de Graaf, Director of Innovation Consulting at Philips, works in the central innovation and design organization that supports those units. His team builds AI and digital tools that speed up R&D work, from managing requirements to shaping new value propositions. The units decide whether to use those tools. So Rob’s team has to show each unit, one by one, that AI is worth its money and its engineers’ time. The team does that in four steps:
1. Connect the data before adding AI
Philips runs innovation on three platforms- one for the portfolio and business cases, another one for design and product documentation (12,000 to 14,000 engineers work in it), and a third one for parts, suppliers and production. Each system keeps its own data, and engineers write the documents by hand. Rob describes most Philips engineers as technical writers who produce the documents the FDA and other regulators read when something goes wrong.
AI can speed up that writing, but the output can look correct and still be “plausible nonsense,” in Rob’s words. A reviewer can catch a doubtful sentence in one document. A wrong field in the data, however, ends up in every document built from that data. So the data has to be connected and checked before AI writes anything.
The central team therefore moved the data from all three systems into one data lake. It started where Philips already had plenty of data, in requirements management and configuration management, and both areas delivered results quickly. Documents now come out of the data, and engineers' jobs are shifting from writing documents to keeping the data correct.
2. Give every unit its own euro figure
With the data connected, the central team had working tools. The next step was getting the units to pay for them. Unit leaders judge projects on business results, and their budgets lacked a line for AI in R&D. So the team had to show each unit what the tools were worth in money.
To do that, the central team built a value tree. The tree starts from one objective and links every AI initiative to a business gain:
Objective: save a set percentage of R&D spend.
Engineering time saved: requirements management, configuration management.
Cost of poor quality avoided: documents generated from data, connected data from idea to market.
Faster time to market: digital value proposition creation, a smooth handover to downstream teams.
The team puts a euro value on each initiative and splits that value across the business units, phased from year one to year three. As a result, every unit leader sees one number for their own unit, year by year.
Value proposition creation shows how large the gains can be.
“Normally it could take us up to seven months to even decide whether we will pursue a value proposition,” Rob says. “Basically, you can do that in a day now.”
A faster decision also speeds up the handover to development, so products can reach the market sooner.
Once unit leaders saw their own number, they understood the value quickly. “Then somehow people start to understand,” Rob says. The leaders looked at the benefits first and handled the cost through their regular yearly budget. As a result, the units now invest more in AI for R&D, from a budget line created for it.
3. Build the tools once and let each unit set its pace
Before the reorganization, the central team could roll a new tool out across Philips. Once decision power moved to the business units, the central team lost that option, and each unit now chooses for itself which tools to use and when.
With the decision now in the units’ hands, Rob’s team focused on the part it still controls: building the tools. The central team provides standards, platforms and expertise, and builds each tool once so every unit can reuse it. The choice to use those tools stays with the units. Because each unit carries a target to get more value from its R&D budget, each unit picks the projects where the tools help most and moves at its own speed.
Inside each unit, the final test is whether engineers use the tools. “For engineers to really accept the tooling, it’s key for them to see: okay, my work gets more fun,” Rob says. So the tools they build target the work engineers want to get rid of - in the case of Philips, documentation tops that list.
4. Be honest about the problems AI creates
Before AI, a product manager developed about 3 value propositions a year, chosen from 10 to 20 starting ideas. With AI, the same product manager can create 300, when Philips still has the budget to develop 3. Choosing 3 from 300 takes far more work than choosing 3 from 20. Rob is open about the limitations of that.
“I can’t go to a product manager and say, ‘Here are 300 value propositions. Can you pick the best ones, please?’ I know that I can’t do this.”
The answer, in Rob’s words, is to be “very purposeful.” Before the team generates more options, it goes back to the questions that narrow the field: what customers say they need, what matters most to them, and where Philips can make a real difference against competitors.
