The board had your AI strategy in hand. You presented it last quarter; the deck was good. Then, a few minutes into this quarter’s meeting, a director set it aside and asked a different question.
Where is the EBITDA?
Not what is the strategy. Not how many pilots are running. Not what percentage of the company has a Copilot license. Where, specifically, in the numbers the CFO reports every quarter, is the money AI was supposed to make.
If you have felt that question land in the room, you are not imagining a shift. The question your board is asking changed sometime in the last few quarters, and it changed more than it sounds. Most of the consulting industry is still answering the old one.
The question changed, and most of the industry didn’t notice
For roughly three years, the right answer to a board’s AI question was a strategy. The technology was moving too fast to commit, the failure rates were ugly, and a credible plan with a few funded experiments was exactly what a prudent board wanted to see. Presenting an AI strategy was the responsible move, and it bought time while the ground settled.
That posture has expired. Boards have stopped grading the plan and started grading the P&L. Kyndryl’s 2025 readiness report found that 61% of senior leaders feel more pressure to prove AI returns now than they did a year earlier. Teneo’s Vision 2026 survey of CEOs and investors found a majority of investors expecting positive returns within six months or less. The trade press has converged on the same sentence from different directions: the era of AI as an experiment is ending, and the era of AI as an accountable line item has begun.
The phrase that keeps surfacing in finance circles is the cleaner tell. Boards have stopped counting pilots and started counting dollars. The verb in the room is no longer “explore.” It is “show me.”
Here is the part most firms have missed. This is not a tougher version of the old question; it is a different question entirely, addressed to a different organ of the company. “What is our AI strategy” is a question for the strategist. “Where is the AI EBITDA” is a question for the operator, and ultimately for the CFO. A firm built to answer the first one will produce a beautiful, useless artifact when handed the second.
Why your strategy deck can’t answer the new question
A strategy deck answers what AI could do. The board is now asking what AI did, in the only ledger that counts. Those are not the same document, and no amount of polish converts one into the other.
The trap is that most companies have an enormous pile of evidence that feels like an answer and isn’t. Seats deployed. Hours saved. Tickets deflected. The percentage of employees who touched a model last month. These are activity metrics, and activity is not earnings. A CIO interviewed by Computerworld put it more bluntly than any consultant would dare: telling the CFO that most employees use AI is, in his words, <q>like saying 100% of employees use email</q>. Finance does not care about adoption. Finance cares about profitability, revenue, and risk; everything else falls flat in that room.
The numbers behind the discomfort are real. IBM’s own CEO study found that only about a quarter of AI initiatives deliver the returns expected of them, and only around one in six have scaled across the enterprise at all. That is not a story about bad models. The models are extraordinary and getting cheaper by the month. It is a story about the distance between deploying a capability and banking it, and that distance is exactly where the new board question lives.
So the CEO walks in with a deck full of momentum and walks out having answered a question nobody asked. The activity was real. The earnings statement did not move. And the board noticed.
The AI dividend has a leak
Why doesn’t all that genuine productivity show up as EBITDA? Because the gain is real but the capture is not automatic, and most companies have never asked where the money actually goes once AI creates it.
We have written before that there is an AI dividend moving through your business that you are probably not the one banking. The same mechanism explains the missing EBITDA, and it has two main leaks.
The first leak is the vendor. When the productivity gain is delivered by a tool you rent, the vendor sets the price, and the vendor’s pricing is designed to recapture a healthy share of the value the tool creates. You feel faster; your software bill explains where the speed went. The last technology wave trained a generation of mid-market leaders to wait for their vendors to package the capability and sell it back to them, and that habit quietly routes the AI dividend onto someone else’s income statement.
The second leak is the market. When a capability is available to you, it is available to your competitors, and competition is a machine for handing efficiency gains to customers in the form of lower prices. If everyone in your sector can now produce the same work for a fraction of the cost, the savings do not stay as margin; they get competed away into price. That is the uncomfortable engine underneath the observation that, within a few years, you may be able to buy competitors for pennies. The same force that lets you acquire a rival cheaply is the force that erodes the easy margin you assumed AI would hand you.
The conclusion is the thing most strategy work skips entirely. AI EBITDA is not a byproduct of AI activity; it is a deliberate act of capture, performed against two forces actively trying to drain it. The answer was always going to be cheap. Keeping the money is the expensive part.
What an AI EBITDA narrative actually looks like
Here is the artifact the new board question is actually asking for. We call it an AI EBITDA bridge, and it has nothing in common with a strategy deck except the staples.
A bridge starts not with a technology but with a metric. Every mid-market company runs on a countable unit of economics; the claim, the load, the ticket, the lease, the underwriting file, the service call, the closed deal. Pick the metric that drives your P&L, and the rest of the narrative builds itself.
The bridge then walks the board across five planks, in order.
Establish the pre-AI per-unit economics. What did it cost to process one unit before AI touched it, fully loaded; labor, time, error rate, rework. This is the baseline a CFO will trust, because it is the number the company already lives on.
Name the specific intervention. Not “we deployed AI.” A named change to a named step in the handling of that unit: the first-draft underwriting memo now arrives in four minutes instead of forty; the routing decision that took a dispatcher six minutes now takes ten seconds with a human approving on the loop.
Show the per-unit delta. The change in cost, cycle time, or quality on a single unit, measured against the baseline, ideally with a holdout or a before-and-after the finance team can audit. One unit, one honest number.
Multiply by real volume and net the true cost. The per-unit delta times the units you actually process, minus the fully loaded cost of the capability; licenses, infrastructure, integration, and the change management nobody budgets for and everybody pays for. What survives that subtraction is the gross gain.
Trace where the gain lands. This is the plank the entire shift hangs on, and the one strategy decks never include. Of that gross gain, how much reaches the P&L as margin, how much is recaptured by the vendor’s price, and how much will be competed away into customer pricing over the next several quarters. The honest bridge shows all three. The number at the far end, the part that actually stays, is your AI EBITDA.
That last figure is smaller than the headline productivity number, and it is supposed to be. A board does not want the headline; the board has seen the headline. It wants the number that survives contact with vendors, competitors, and the cost of running the thing. A narrative that produces that figure, plank by plank, passes a CFO’s smell test. A narrative that stops at “hours saved” gets the polite nod that means the room has moved on.
This is also why the AI ROI problem is so persistently misdiagnosed. As one of our consultants has argued, the bottleneck is not the model; it is the operating model that the unit flows through. You cannot bridge from a capability to EBITDA without redesigning the work around the unit, and that redesign is operating-model surgery, not a software rollout.
Three questions a defensible AI EBITDA narrative answers
Strip the bridge down and it answers three questions, in this order. If your current AI reporting cannot answer all three, you are still holding a strategy deck no matter what the cover says.
- Which unit of our economics did AI change, and by how much per unit? If the answer is a percentage of employees or a count of licenses, you have measured activity, not impact.
- Did the gain reach the P&L, or did it leak? A board that has read the room knows the dividend can be intercepted by a vendor’s price or surrendered to a competitor’s. The credible narrative names the leak and quantifies what survived it.
- Is the change durable, or does a rival erase it next quarter? A one-time efficiency that every competitor will match is a customer discount with a delay timer. A capability built into your operating model, on data and workflows competitors cannot trivially copy, is a margin you keep. The board is asking which one you have.
The discipline these questions demand is unglamorous and decisive. The right time to figure out how to attribute earnings to a technology initiative is before you need to, not in the forty-eight hours before a board meeting. The companies that will answer the new question well are the ones instrumenting the per-unit ledger now, while the question is still merciful.
Why most consulting is one beat behind
There is a structural reason your advisors are still handing you the old artifact, and it is worth naming plainly because it tells you who to trust with the new one.
A strategy deck is a deliverable a firm can produce from the outside, bill for, and walk away from. It requires no standing inside the operating model and no accountability for what happens after the engagement ends. The AI EBITDA bridge is the opposite kind of work. It requires sitting inside the business long enough to know its unit economics, redesigning the workflow that unit moves through, and staying present while the gain either reaches the P&L or leaks away. That is leadership work, not advisory theater, and most of the industry is neither built nor paid to do it.
This is why so much AI consulting feels one beat behind the board. The firms still selling the strategy are not behind on intelligence; they are behind on business model. They produce the artifact they know how to monetize, which happens to be the artifact your board stopped accepting last quarter.
It is also why a scattered pile of employee-built AI tools, each impressive in isolation, almost never shows up in EBITDA. A rack of reinvented wheels, every one a private experiment, produces enormous activity and no consolidated economics, because no one owns the bridge from any of it to the P&L. Capture is an act of leadership, and leadership is the thing that has to be present, not delivered and invoiced.
The handoff
The board’s question is not going to revert. The strategy phase was real and it is over; what replaced it is a standing demand for the number that survives. The CEO who walks into next quarter’s meeting with an AI EBITDA bridge instead of a strategy deck does not just answer the question better. He changes who is asking it, because a board that sees per-unit economics it can audit stops interrogating and starts compounding.
That bridge is built from inside the operating model, by someone whose job is the earnings line and not the software contract. It is the work our Contract CIO+ engagements and vCAIO service exist to do; vendor-neutral, sector-matched, and accountable for the number at the far end rather than the deck at the front. It is also why we are increasingly willing to put our own compensation on the table against the outcome. A firm that believes its EBITDA bridges should be willing to stand on them.
Your board already changed the question. The only choice left is whether your next answer is a strategy deck or real impact.
