Retail boards are very good at funding proof. They are terrible at funding scale.
That gap is why your AI portfolio looks busy and your P&L looks unchanged. You have a wall of successful pilots and almost nothing in production. The problem is not the technology. It is the shape of the money.
A pilot is a cheap, comfortable commitment. One store. One category. One clean dataset. A fixed end date. Finance can approve it without flinching because the downside is capped and the demo always lands. The second deployment is the opposite. It is bigger, messier, harder to bound, and it is the only budget that actually returns anything.
So we systematically fund the step that cannot pay back and starve the step that can.
The economics only start working on deployment two
Here is the part most business cases get backwards. The cost of an agent is front-loaded. You pay for the integration, the data plumbing, the governance, the human-in-the-loop design, and the change management mostly once. The first site absorbs nearly all of it. That is why the first deployment looks expensive and underwhelming on a spreadsheet.
Deployment two is where the curve bends. The build is already sunk. Each new store or category is close to pure contribution. The agent that looked marginal at one site looks obvious at fifty, because you are now spreading a fixed cost across a growing base of nearly free rollouts.
Fund only the pilot and you book all the cost and none of the compounding. You have, quite literally, paid for the hard part and then walked away before the payback.
BCG named this directly in their 2026 work on retail and CPG AI returns. The pilot trap is real: pilots succeed in controlled environments, but complications compound at scale. The complications are real, but they are a reason to budget for scale deliberately, not a reason to stop at proof.
The prize is a rollout number, not a pilot number
Look at what scale actually unlocks. BCG analysis suggests that scaling the full set of relevant AI initiatives across the demand value chain can deliver 220 to 350 basis points of cumulative earnings before interest and taxes for CPGs and 180 to 360 bps for retailers. And the ceiling moves as agents mature. As agentic capabilities mature and companies build stronger transformation muscles, the opportunity could expand to as much as 1.7x for each segment.
Read those numbers carefully. Every one of them is a deployment-at-scale number. None of them is a pilot number. The margin lives in the rollout you keep deferring.
Then look at why boards cannot see it. Measurement is a weak link, as more than half of survey respondents say that they do not measure the ROI of consumer AI investments. And it gets worse at the handoff. Many pilots track data but do so without a clear baseline or threshold for scale.
That is the real failure mode. Not a bad model. A pilot that was never designed to answer the only question that matters: what has to be true to fund site two? If you do not set the baseline and the scale threshold before the pilot starts, you will finish it with a nice demo and no decision. So it stalls, and everyone quietly moves on.
Price the Cost of Doing Nothing into the case
This is where CODN earns its place in the business case. A stalled pilot is not a neutral outcome. It is a loss. You have spent the integration and governance money, captured none of the compounding, and left the margin on the table for a competitor who did fund the rollout.
So the Cost of Doing Nothing on the second deployment has three parts. The sunk build cost you will not recover. The margin at scale you chose not to book. And the widening gap, because the retailers who are scaling are already pulling away on growth and returns while the rest run their fortieth pilot.
Here is the operating change I would make at board level. Stop approving pilot budgets as standalone line items. Approve a deployment business case that funds the scale step up front, with the pilot as gate one inside it. If a use case cannot survive that case, it does not deserve the pilot money either, because a pilot that was never going to scale was theater from the day you signed it.
Three questions to run every AI request through before you fund it:
First, what is the baseline and the explicit threshold that triggers deployment two? If nobody can answer, you are funding a demo.
Second, who owns the P&L for the scaled version, not the proof? An agent with no production owner will die at the handoff every time.
Third, is the integration and governance cost in the business case from day one? If it appears only at scale as a nasty surprise, the rollout will be killed by the very cost the pilot was supposed to have retired.
The retailers who win this cycle will not be the ones with the most pilots. They will be the ones who funded the second deployment before the first one finished. Proof is cheap and scale is where the margin lives. Fund accordingly, or keep paying for demos your competitors are turning into EBIT.