The Three Questions That Decide Whether a Martech Investment Pays Off

Most martech investment decisions fail the same way: the tool gets bought on capability, owned by everyone, and defended past the point it stops working. Three questions, borrowed from AI project triage, catch the bets that won’t pay off before the money moves.

Eric D. Brown kills technology projects for a living. He’s a fractional CTO who gets called in when a leadership team is staring at a build-or-buy decision they can’t afford to get wrong, and a fair share of his job is telling them which projects to stop. Before he greenlights any AI initiative, he runs it through three questions (1. Brown, 2026). They look modest. They’re brutal.

Apply them to your martech stack and most of what you’re about to buy, renew, or “finally turn on” doesn’t survive. That’s the point.

Brown presents the questions as a checklist. The more useful read is that they’re a chain, where each one only works because the one before it held. Skip the order and the discipline falls apart, which is exactly how martech investment decisions go sideways.

Does it touch a number you already measure?

Start with the question that sounds too simple to matter. Does the tool move something you already count, today, in dollars or hours or customers who leave?

If you can’t name that number before you buy, you’ve already lost the ability to judge the purchase. You’ll evaluate it on the demo, the capability list, the roadmap, the feeling that you should have this. None of those are measurable, and all of them are how the slickest line item on your license gets approved.

The failure pattern is visible in AI spending right now. KPMG’s 2026 survey of 2,500 technology executives found investment decisions for new AI tools are routinely made on “indirect and hypothetical benefits” rather than anything the organization already tracks (2. KPMG, 2026). The same reflex runs through martech. A CDP gets bought to “unify the customer view.” A DXP gets renewed because personalization is “strategic.” Ask which existing metric either one is supposed to move, and the room goes quiet.

Pick something boring and already on a dashboard: cost per qualified lead, email revenue per send, the hours your ops team spends reconciling data. If the new capability doesn’t touch a number you already report, you’re buying a hypothesis and calling it an investment.

Who specifically owns the outcome?

Now the number needs a person. Not a committee, not “marketing ops and IT,” not the vendor’s customer success manager. One name.

Here’s where the stack stops being a collection of tools and starts being an operating model. Every capability you switch on changes who does what, who’s accountable for the result, and who has to change their Monday morning to make it work. Owning the outcome means your job changes when the capability goes live. If nobody’s job changes, nobody’s actually running it, and the demo was the high point.

Shared ownership is the quiet version of the same problem. When a capability belongs to two functions equally, it belongs to neither the moment the number stalls. Marketing points at IT’s integration backlog. IT points at marketing’s unclear requirements. Both are partly right, which is why nothing gets fixed.

The owner you want is the person who already owns the number from the first question. If lead quality is the metric, the demand-gen lead owns the lead-scoring model in your automation platform, and owns it on their performance review. That’s the test. Could this person’s review include whether the tool delivered? If it can’t, the ownership is fictional, and fictional owners can’t kill anything.

Can you walk away in 90 days?

The last question is the one nobody wants to answer while they’re excited about a purchase. If the number doesn’t move, can you kill this in 90 days without a political fight?

If you can’t, the bet was too big to start. The reason you can’t is usually that you skipped the first two questions. With no baseline metric, you can’t prove it failed. With no single owner, no one has the standing to pull the plug. So the tool stays. It gets defended, re-licensed, and folded into next year’s “optimization” plan, because walking away would mean admitting the original call was wrong.

That describes most of the AI spending wave already. MIT’s Project NANDA studied enterprise AI initiatives and found 95% delivered no measurable return, with the barrier being how the work integrated into daily operations rather than the technology itself (3. MIT NANDA, 2025). A bet with no measurable return and no owner doesn’t get killed. It gets renewed, because killing it would require someone willing to say out loud that the number never moved.

A 90-day exit is what lets you make the bet at all. You can move fast precisely because walking away stays cheap. The capabilities that can’t be killed are the ones that should never have been funded.

The order is the argument

Run the three questions in sequence, because the sequence is the point. A measurable number gives you the basis to judge. A single owner gives someone the standing to act. A real exit gives you permission to try. Drop any one and the other two stop working.

The bets that pass all three look small, specific, and a little boring. They’re also about the only martech investment decisions that reliably pay for themselves.