AI doesn’t create value. Better decisions do.

Most companies do not have an AI problem.
They have a decision problem.
AI activity is now everywhere: copilots, proofs of concept, new data platforms, innovation teams, agentic workflows. But for many leadership teams, the uncomfortable question remains: where is it showing up in the P&L?
The gap is stark. Individuals report substantial productivity gains (>30%), while organisations capture only a fraction of them (<10%). The missing value is rarely a model-performance issue. It leaks away in approval layers, fragmented accountability, poor process design and decisions that still move at committee speed.
Value capture is a decision problem in disguise.
The clock is ticking
This matters most for private-equity-backed businesses.
Brynjolfsson’s productivity J-curve applies to AI as it did to electrification: first investment, then experimentation and disruption, and only later returns. But the hold period is finite. If AI is still a collection of pilots when the business comes to market, the buyer will not pay for the promise.
Buyers pay for evidence, not intent. They will ask what AI did for EBITDA, whether it scales, and whether it survives diligence.
A better question
That changes the question data and AI leaders should be putting to their CEO and board.
Not: “Where can we use AI?”
But: “Which decision, if we made it materially better and faster, would move enterprise value?”
Think pricing. Bid/no-bid. Working capital. Demand planning. Credit decisions. Workforce deployment.
Choose one decision that matters. Then redesign it end to end.
That means looking beyond the technology:
What data does the decision need?
Who has the right to decide?
Where do approvals slow it down?
What should AI recommend, automate or execute?
Should a human be in the loop, on the loop, or out of it?
Who is accountable when the decision is wrong?
What this looks like in practice
I recently worked with a leading global professional services business facing a familiar threat. AI was commoditising the work it was paid for, and expensive, inefficient operations were already eroding profitability.
The original request was for a strategy to make better use of its data assets and exploit AI more effectively. That was the right instinct but the wrong scope.
A strategy alone would have produced a roadmap of use cases bolted onto an operating model that was already under pressure. What the business needed was a transformation that built Data and AI into the method and lifecycle the practice ran on: how work was scoped, designed and delivered. That meant creating new areas of value, and deliberately reinventing the work most exposed to commoditisation before the market did it for them.
The outcome was Board investment in a three-year transformation programme, a team built to deliver it, and three significant GenAI and agentic AI pilots, each proving the opportunity in a different part of the design lifecycle.
Those were pilots, but not the kind I warned about above. Each was chosen to prove a specific piece of the redesigned lifecycle, with a funded path to scale behind it. A pilot that tests a tool proves very little. A pilot that tests a redesigned way of working proves the investment case.
The seventy per cent
BCG’s 10/20/70 rule captures why this matters, though I would adjust what sits in the ten per cent.
Ten per cent is AI strategy, governance and tool selection. Twenty per cent is data, platforms and technology. Both are important, but neither is sufficient.
The other seventy per cent is how work gets done: decision rights, roles, management routines, incentives and leadership behaviour.
Put rocket boosters on a horse and cart and you do not get a spaceship. You get a very expensive accident.
The same is true of organisations. The old maxim holds: new technology plus an old organisation equals an expensive old organisation. AI cannot compensate for unclear accountability or an inability to make decisions. In fact, it usually exposes both.
Governance at machine speed
There is also a governance dimension that boards cannot treat as a separate workstream.
As AI agents take action at machine speed, AI, cyber, risk and compliance stop being four functions and become one design problem. Governance designed for human-speed decisions will not hold when an agent can trigger an irreversible decision in seconds.
The answer is not to slow everything down. It is to design deliberately:
Where does autonomy create value?
Where is human judgement essential, and should it sit in the loop, on the loop, or step out of it?
What controls are proportionate to the risk?
How are outcomes monitored?
Can the organisation explain its AI-enabled decisions to a buyer, regulator, customer or employee?
Someone has to own the decisions
If decisions are the unit of value, then someone has to own the decision portfolio. That means deciding which decisions to redesign first, how they are governed, and whether the value is real.
Today that ownership is usually spread across the CDO, the CIO, the COO and the business. That fragmentation is exactly where value leaks away. This is the case for the Chief Decisioning Officer: not another technology role, but the executive accountable for turning better decisions into enterprise value.
Start with a decision
The companies that create real value from AI will not necessarily be those with the most pilots or the largest technology budgets.
They will be the ones that identify the decisions that matter, redesign the work around them, assign clear accountability and prove the economic benefit quickly.
The prize is not an AI strategy. It is a better business: one that decides faster, allocates capital and talent more intelligently, and converts technology into measurable enterprise value.
That is what an Intelligent Enterprise looks like.
The right place to start is not with a tool.
It is with a decision.

Eddie Short
View on LinkedIn
