Better AI won't win you the next decade.

Board of Innovation's AUTONOMOUS summit brought 10,000 people together this month. I left convinced most 2027 plans still aren't ambitious enough.

Philippe De Ridder, BOI's founder and CEO, expects as much progress in the next ten years as we saw in the last hundred, because invention and adoption are both speeding up. He opened with a challenge:

“Better AI won't win you the next decade, but rebuilding your company will.”

Rebuilding a company is the hardest job most of us will ever take on.

AI will push more change through businesses, faster, than the internet did. We'll be rethinking what we sell at the same time customers and competitors are changing what it's worth, and the business still has to make its numbers while all of that is going on.

The 2030 company doesn't get built by writing a bolder vision statement. It gets built by changing what you fund, what you approve, who has authority, and what work you treat as valuable, starting with the 2027 plan. Ambition alone won't move any of that. The operating choices you make next year will.

Three implications stood out to me:

Each one lands in the same place: a change you have to build into how the company runs in 2027.

Laura Stevens from BOI warned that waiting for the technology to settle down only puts off learning how to run a different kind of company. By 2030, I want to be running one that changes as fast as its customers and the market do. Our 2027 plans have to pay for that while we still have the money and the customer relationships to build it.

Replace revenue before the market does

Amir Ouki from BOI described a meeting that's starting to happen at services firms. A client shows up with an AI-generated first draft of the work they used to pay the firm to do. The firm can see the problems in it right away. The client thinks part of the job is already done and wants to know what the firm is going to add.

Services fees have always bundled two things: the expert's judgment and the hours it takes to produce the deliverable. Clients now have a reason to pull those apart. Bill by the hour, and faster work means fewer hours to bill. Amir expects fixed fees to protect margins for a while, until a competitor or the client's own team pushes the price down.

His answer is to rethink what the firm takes responsibility for. The recommendation that used to end an engagement becomes the start of an ongoing service.

Say you run a market-research firm. Your client has its own draft report now. What it still needs is help deciding whether a shift in demand calls for a price change or a new product. You keep that advice current as the market moves, and you help the client test its response. AI does the ongoing analysis. Your experienced people decide what it means and work through it with the client.

The same question applies inside your company. Once business teams run more of their own analysis, your insights group has time to help them decide what to build and to stay involved as the results come in. That means changing what you hold the group responsible for. Leave the old request-and-report routine in place and you waste the time you just freed up.

The change that decides whether the new offer ever happens is letting it compete with what you sell today. A manager paid on billable hours has every reason to defend the old model. In 2027, that means writing the incentive change into how you pay and measure that manager, and giving someone the authority to launch the new service knowing it will take revenue from an existing one. Wait until the market forces the call and you'll be making it without the customer's trust or the room to build. I'd rather make it now, while we have both.

Shoppers are starting to ask the same question of brands. Nick Lafferty from Profound showed what happened when someone asked ChatGPT how much Nike's Alphafly 3 costs. It gave the price, then spent most of the answer on competing shoes. The shopper had picked a product and got handed reasons to reconsider it.

An AI assistant can't try on a shoe, so it leans on people who do. Nick pointed to RunRepeat, a review site that weighs running shoes and cuts them in half. In a sample Profound tracks, it gets cited almost three times as often as Nike.com. A shopper comparing options needs a reason to believe your product is better for them. That kind of evidence also helps a product that does one job especially well reach people who never would have searched for your brand.

So the 2027 marketing plan has a decision in it. If AI is cutting what you spend on production, the savings can flow straight to the bottom line, or you can move part of them into the product and into proof that it does what you say it does. That's a budget line, and a change in what the marketing team is responsible for producing.

Speed creates new bottlenecks

At ExxonMobil, Sameer Khan's team cut a testing cycle from eleven weeks to five. They started by asking what was holding it up. The answer was writing thousands of test scripts by hand, so they pointed AI at that step and made sure it had the right data.

Jochen Kokemueller from Bosch talked about a developer who finishes a piece of work in an hour and waits three days for someone to review it. Checking has to speed up along with producing, and a person still has to answer for what reaches the customer.

That gap between a one-hour build and a three-day review is an operating decision hiding as a delay. Speed up production and leave the approval untouched, and the bottleneck moves to the review. The 2027 work is redesigning the review itself: who signs off, how fast they can, and what they're accountable for when the work reaches the customer.

The market-research firm runs into the same wall. Ongoing advice is worth paying for when it reaches the client while there's still time to act. Approvals built for one-off projects would kill the service before it got going. Fund the new offer and keep the old sign-off process, and you've funded something that can't run.

In a product launch, work sitting between departments eats the window when the opportunity is worth chasing. We own that waiting time as leaders, including the approvals we've kept for ourselves.

Before you turn time saved into headcount cuts, look at the work the business can now go after. The people who used to prepare the report could be helping customers act on it. Cut them too early, and you won't have the team to deliver the higher-value service you're trying to build. That's a staffing decision the 2027 plan makes explicitly or by default, and the default is usually the cut.

The bottleneck will keep moving. Faster analysis means decisions pile up. Once those get made, the pressure lands on the people delivering the service. The operating answer is to give teams the authority to fix problems across departments and to stay close enough to see where the work is waiting, with clear responsibility for what the customer ends up getting. That's a change to decision rights, and it belongs in next year's design.

Judgment gets more valuable, not less

Eric Porres, Logitech's chief AI officer, described a customer who turns down a proposal because they can't spare the people to implement it. An AI helping with the next proposal needs to know that, and it needs a way to check whether it's still true. If it is, building implementation support into the offer gives that customer a reason to say yes.

Eric writes down why he made each call and keeps his corrections, so the next piece of work starts from what he already learned. An established business has paid for that kind of experience through years of serving customers. For a brand, it's knowing why people tried a product and why they didn't buy it again. The next product team should start with those lessons in hand.

Capturing that judgment is now an operating task with an owner and a place to live. In 2027, that means deciding whose reasoning gets written down, where it's kept, and how the next team picks it up. An experienced researcher gets to set the questions the AI works on and the bar its answers have to clear. Their judgment ends up in work they never touch, and their own time goes to the unfamiliar problems where a wrong call costs the customer. Treat that as part of the job and staff for it.

Eric also warned that an AI that knows your opinions gets good at agreeing with them. When fresh customer evidence cuts against something senior leaders have believed for years, the team holding that evidence needs the standing to say so and change the plan. That standing is a decision right you either grant or withhold. Withhold it, and the AI quietly protects yesterday's assumptions.

Dominik Heinrich, Coca-Cola's design intelligence chief who I wrote about last week, argued for paper and pencil before AI, so people understand the work well enough to judge it. A brand team working on margins has to know which qualities customers are paying for before it approves a cheaper version.

Jochen also pointed out that the junior roles going first are the ones where people learned by doing the work badly and getting feedback. A junior colleague can still own a customer recommendation, with an experienced person helping them check the reasoning and see what happened after the client acted on it.

The people who'll run the business in 2030 need chances now to make decisions they can learn from. So the 2027 change is in what you measure a senior person on. I'd make developing colleagues part of how we judge that contribution, and give them the time to do it. Leave the scorecard as it is, and coaching stays the thing people do after the real work, which means it doesn't happen.

Budget for learning, not just execution

Paulo Oliveira's team at IKEA set out to cut the time from idea to market pilot from about eighteen months to eighteen weeks. The biggest holdup was deciding which ideas deserved more work. Without the resources to explore them all, teams killed ideas before they had the evidence to choose well.

They built an AI pipeline that runs ideas past simulated customers, so they can chase all of them. An AI facilitator asks the team hard questions about each idea, and the people have to work out the answers together. The eighteen weeks still end in a market pilot with real customers.

What it costs to learn about an idea decides which businesses you ever get to build. As that cost falls, you can take an unfamiliar opportunity a lot further before committing to development. So the 2027 move is practical: reopen the ideas you ruled out because the early work was too expensive, look at them again at today's costs, and get customers involved before anything launches. That's a change to how the pipeline runs, with someone accountable for keeping killed ideas in view.

Laura explained why those ideas still struggle to get funded. Past success turns choices about how a company operates into beliefs about how it has to operate, and those beliefs get built into approval processes and performance targets. A leadership team can ask for a different future while its investment rules keep rewarding improvements to the business it already has. Fixing the rules is the 2027 work, and it's harder than approving the vision, because the rules are where the old company defends itself.

Back to the market-research firm. A paid trial would show that clients will buy ongoing support. Whether they renew, and how much expert time it takes to serve them well, only shows up after months of doing the work. The team needs funding that lasts long enough to find out. Expecting a young business to supply the same evidence as an established one makes it harder to fund the work that would produce that evidence.

Jon Chen from BOI pointed out that a three-year plan usually gets funded a year at a time, and the CFO or the board can pull the plug at any review. His advice was to build the multi-year case with finance, and have finance set aside the first year's savings for the next stage. That's the funding-system change that makes the rest possible, and it happens in the 2027 budget or it doesn't happen at all.

That agreement has to cover the time you free up, too. A team given room to build a new offer needs some of its other work taken off its plate. Any savings you plan to reinvest have to show up as money the business can spend.

Put part of the 2027 budget behind a business you want the company earning from in 2030. The team building it should know what it has to learn to earn the next round of funding, and it should have permission to come back with findings that change the idea. Teams get better at judging opportunities by putting offers in front of customers and working through what comes back. Leaders need the same practice deciding which bets get more time and when to stop funding one. Give that skill several years to develop, with repeated chances for people inside an established company to build something new.