Research
Most companies don't have an AI strategy. They have activity.
Buying tools and running pilots is not the same as deciding where AI should change how the business works, and who owns proving it did.
Over the last two years, most companies have done the visible parts of AI adoption. Budgets are approved. Pilots are running. A few copilots are rolled out. From the outside it looks like progress.
Then a quieter question starts moving through the room: the spending is real, so where is the return?
That question is not a sign of failure. It is the moment a company discovers the difference between activity and strategy.
Activity is easy to mistake for progress
The evidence is consistent across the serious research. MIT's The GenAI Divide: State of AI in Business 2025 found that around 95% of generative AI pilots deliver no measurable impact on the profit and loss statement, and only about 5% reach production with real value. McKinsey's State of AI survey reaches a similar place from another angle: most organisations are still experimenting or piloting, and only about a third have begun to scale. The UK government's AI adoption research found that even among companies already using AI, only about half feel ready to scale it.
Read together, these say one thing. Adoption is now common. Value is not. A company can point to dozens of use cases, a wall of dashboards, and real productivity anecdotes, and still be structurally unable to turn any of it into a result the board can see.
Pilots flatter progress. That is their danger. They prove the technology can do something. They do not prove the business decided what that something is worth.
What a strategy is not
When leaders say they have an AI strategy, they usually mean one of three things.
01
A list of tools they have bought or plan to buy.
02
A portfolio of pilots, each owned by whoever was curious enough to start one.
03
A deck full of the word transformation, with no line that says which number moves.
None of these is a strategy. They are inputs at best. A tool is not a decision. A pilot is not a commitment. Ambition is not a plan until someone can say where it lands.
What an AI strategy actually is
A workable AI strategy is smaller and harder than most decks. It is a short set of decisions, made before the build starts.
If a company can answer those four for one workflow, it has more strategy than most competitors running twenty pilots. If it cannot, the problem was never the model. It was that no one decided what the model was for.
We wrote separately about why AI ROI so often turns out to be an organisational problem rather than a technical one. This is where that begins.
Efficiency is not the same as value
There is a trap worth naming, because most companies fall into it. A pilot saves people time. Everyone agrees it is useful. It gets counted as a win.
But time saved is not return realized. Unless that saved time changes throughput, decision quality, customer outcomes, or margin, it stays an operational anecdote. It never reaches the profit and loss statement. The same government research found exactly this pattern: many businesses reported better workforce productivity from AI while seeing no change in revenue.
Efficiency at the task level and value at the business level are different claims. A strategy is what connects them.
Without it, a company can feel more productive every quarter and grow no faster.
Where the value shows up
The MIT research points somewhere useful for anyone deciding how to build. The pilots that succeed tend to share a shape. They pick one real pain point. They integrate into an actual workflow instead of sitting beside it. They improve as they run.
Two details are worth holding onto. Mid-market companies often move faster than large enterprises here, reaching full implementation in around ninety days, because they carry less bureaucracy and can redesign a process without a committee. And tools built with an external partner succeeded roughly twice as often as internal builds, largely because an outside team is forced to be specific about the workflow and the outcome before anyone writes code.
That last point is not an argument for outsourcing. It is a lesson about discipline. The teams that win are the ones made to answer, out loud, which decision changes and who owns it. That discipline is the thing we build every engagement around, and it is described in more detail in how we work.
A better starting move
If you are early, the useful next step is not another tool review. It is narrower than that.
Pick one workflow where AI could change the economics. Name the number it should move. Put the outcome on the person who owns that number. Agree in advance what proof would justify scaling and what would justify stopping. Then build only that, and redesign the process around the capability instead of laying the capability on top of the process.
That is unglamorous next to a platform-wide rollout. It is also the version that tends to reach production.
Adoption got most companies into the game. It does not decide whether the game was worth playing. The companies that pull ahead from here will not be the ones making the most noise about AI. They will be the ones who can point to one workflow, one number, and one name, and say: here is where it changed, and here is who proved it.
Everything else is activity.
One workflow, one number, one name.
That is the shape of every engagement we take. If you can name the decision, we can tell you within a fortnight whether it is worth building against.