How to Justify an AI Investment With No ROI Yet

The playbook for justifying AI's return is already written, and it gets repeated in every proposal that reaches a board. The problem isn't the metric. It's the director who already watched two pilots die and won't be talked into another slide of indicators.

Carlos Andrés Ramírez · 2026-09-12

The playbook for justifying AI's return is already written, and it gets repeated in every proposal that reaches a board. The problem isn't the metric. It's the director who already watched two pilots die and won't be talked into another slide of indicators.

Sit in the room where someone defends an AI investment that isn't producing anything yet, and the pattern repeats almost every time. The sponsor arrives with a slide of future indicators, a transformation narrative, and a well-chosen use case. Fifteen minutes in. Then someone at the far end of the table, the same person who signed off on two pilots that went nowhere, asks one thing: why would this time be different? That is where the indicator deck ends and the real conversation starts.

The playbooks circulating today solve half the problem. They teach how to pick a measurable indicator, tell a story instead of throwing out a bare number, anchor the case in one concrete process. All correct, and none of it lands with a director who stopped looking at the indicator months ago. What that director is weighing is whether to trust the person bringing it, because trust got extended once already and it cost something.

Not every AI investment with no return yet is the same thing, even though a board tends to treat them alike. Building the data foundation that makes any AI project possible does not pay back in year one, and it should not have to. A pilot still hunting for its use case does not either. Scaling something that already worked in one team and still shows nothing in the next one should be showing something by now. Putting all three under the same question, when do we see numbers, is exactly why the room never leaves satisfied with the answer.

The symptom

How do you justify an AI investment to the board when it has no return yet?

The answer taught in most playbooks is the same one: pick an indicator before starting, choose a use case with visible impact, tell the story of the problem it solves. It works the first time someone asks for AI money. It stops working the second time, once a record of unmet promises is already sitting at the same table.

• The committee repeats the same language of future indicators it used a year ago, when it asked for the last budget.
• Nobody in the room can say, without pulling up the minutes, what actually happened to the pilot approved last time.
• The sponsor presents a complete business case and skips the question of what happened to the previous money.
• The skeptical director stops arguing the proposed indicator and asks instead who answers if this one fails too.
• The budget gets approved anyway, with less money and more conditions than last time, without anyone calling it what it is: accumulated distrust.

The problem underneath

The board isn't voting on the indicator. It's voting on the memory of the last time it trusted one.

Credibility in front of a board works like its own budget, and it spends the same way. Every promise that goes unmet withdraws something from that account, and the account does not refill by telling next quarter's story better. It refills by showing that what got said last time actually happened. A sponsor asking for money a third time without ever closing the loop in public on the previous two is not presenting a business case. They are asking the board to ignore its own record, and an experienced board will not.

The board doesn't reject the indicator. It rejects whoever can't say what happened the last time they asked for the same thing.

BECOME

The framework

What do you bring into the room when the result doesn't exist yet?

Five pieces separate a budget request a skeptical board approves from one it merely tolerates.

Named stage

Say whether this is infrastructure, a pilot, or scaling, because each one carries a different, reasonable window before it should show a return. Putting all three under the same question is what runs out the board's patience early.

Closing the last chapter

What exactly happened to the last approved AI budget, said in the same room, without dressing it up: what worked, what got shut down, what got dropped. Without that close, the new request inherits the previous one's debt.

What can be shown today

If there is no return yet, what intermediate evidence exists: what got learned, what risk got removed, what decision no longer has to be made twice. A skeptical board accepts no number yet if something concrete stands in its place.

Conversion date

The day this investment has to start showing a measurable result or get shut down, said in the same meeting where the money is requested, not months later once nobody remembers saying it.

Who owns the hard question

The name of the person who comes back into that room if this doesn't work either, and answers in person, not a committee or a new slide. A board trusts a name more than it trusts a plan.

None of the five require a new indicator. They require something less comfortable: saying out loud what happened last time, before someone else in the room says it first.

Before the next board presentation, write in one sentence what happened to the last approved AI budget: what worked, what got shut down, what got dropped. If that sentence doesn't exist yet, that is this week's work, not the indicator slide.

Let's prepare your next board presentation