# Why 'we'll figure out ROI later' kills AI budgets

_Later never comes with the data you needed. It comes as a budget meeting where you are asked for a number you no longer have any way to produce._

Deferring ROI measurement feels reasonable and quietly guarantees you cannot prove value when the budget is on the line. An anonymized founder story on measuring from day one, and how to avoid the trap.

# Why 'we'll figure out ROI later' kills AI budgets

"We'll figure out ROI later" sounds responsible. Get the thing working first, prove value once we see it. In practice it is how a promising AI pilot loses its funding. Later arrives as a budget review, someone asks what it returned, and the honest answer is that nobody set up the measurement, so there is no number. No number, no renewal.

## The founders who measure from day one do not have this problem

The founder of an equity-valuation platform that now serves more than 22,000 companies told me his rule for new founders is almost aggressively simple: prioritize cash flow and paying customers first, and treat everything through that lens from the start. His whole business is putting a defensible number on things that used to be vague, and he runs his own company the same way, with the return measured in real time rather than promised for later.

That is the mindset that protects a pilot. Not more optimism, more measurement, early. When value is tracked from day one, the budget conversation is a formality, because the number already exists. When it is deferred, the conversation is an ambush, because the number has to be invented under pressure and everyone can tell.

## Deferring measurement makes the claim unfalsifiable

There is a hard mechanical reason "later" fails. If you set the baseline before deployment, you can prove ROI. If you do not, the ROI claim is unfalsifiable, and "later" almost always means the baseline was never captured. Once the AI feature is live, the old cost, the old handling time, and the old error rate are gone. You cannot go back and measure a before-state that no longer exists. The window to make the pilot provable closes the moment it launches, and "we'll figure it out later" is a decision to let that window close.

This is a big part of why so many pilots die measurable-less. MIT's 2025 State of AI in Business report found roughly 95% of enterprise GenAI pilots deliver no measurable return. Some of those genuinely did not work. Many of them worked fine and simply were never instrumented, so when the budget question came, they had a story and not a result.

## Measure the leading indicators now, the return on schedule

The counter to "later" is not "prove full ROI in week one." A pilot measures feasibility, not ROI, and real return follows a curve of roughly 0% during the pilot, 10 to 30% by month 12, and 50 to 150% by month 18. What you do now is capture the baseline and the leading indicators that will roll up into that return, so when month 12 arrives you are reading a number, not reconstructing one. The instrumentation is cheap while the pilot is small, and it never gets cheaper than it is on day one. Later is fine for the return. It is fatal for the baseline.

## How we approach it at Density Labs

Our AI Readiness Assessment is a fixed two week engagement priced at $2,500. We capture the baseline before a line of production code exists, because that is the only moment it is available cleanly. We define the leading indicators to track during the pilot and the horizon on which real return should show, so the eventual budget conversation is backed by data collected from day one. It is the difference between walking into a review with a number and walking in with an excuse.

Later is where budgets go to die. Measure the before-state now, while you still can, and the pilot gets to keep its funding.