If you ran a portfolio of operating businesses in 2024 or 2025, you have already lived this. A pilot got funded. The team demoed something impressive. The slide deck made the rounds. Nine to fifteen months later, the pilot is either in indefinite extension, has been quietly scoped down into a chatbot, or has disappeared from the executive review entirely.
You are not alone. The pattern is consistent enough across operating companies and PE portfolios to name.
The shape of the stall
There are four common landing places for an AI pilot that did not reach production.
The chatbot collapse. The pilot was scoped against a complex operational workflow. By the time the production-readiness review came around, the scope had been quietly reduced to a chat interface that answers questions the staff already knew the answer to. The operating gap is unchanged; the team installed an interface.
The dashboard pivot. The pilot produced model output. The team did not have a path to write the output back into the workflow that downstream staff depended on, so they built a dashboard for the operations director to look at instead. The dashboard is rarely opened.
The indefinite extension. The pilot stays in pilot mode. New use cases are scoped on top of the original pilot. The original pilot never reaches production but is no longer up for review. Innovation budget keeps flowing.
The vendor swap. The first vendor did not produce the production result. A second vendor is brought in with a new pilot scope. The cycle repeats. The shape of the work that did not get done is identical.
What the financial impact actually looks like
For an operating business in the $5M to $50M revenue range that ran an AI pilot in 2024:
- Pilot cost: $40,000 to $250,000 in vendor fees plus internal team time.
- Operational change since pilot: typically less than 2 percent on the metric the pilot was supposed to move.
- Compounding cost: every month of unrecovered operational leak that the pilot was supposed to close.
For a PE-backed portfolio with 4 to 12 operating companies, the math is multiplicative. The portfolio operating partner is now sitting in front of a sponsor review that requires either a real production result or a documented write-down.
What changes the outcome
The structural answer is that the work nobody scoped during the pilot is the work that produces the production result. Specifically:
Integration depth, not model selection. The model is a commodity. The wedge is the integration into the system of record. AMS360 for P&C insurance. ServiceTitan for HVAC. Dentrix or Open Dental for dental. The wedge is owning the integration code, not the model.
Operational measurement, not pilot metrics. A pilot is measured on model accuracy. A production deployment is measured on revenue recovered, hours reclaimed, retention lifted, contact velocity gained. The measurement system is operational, not technical.
Vertical specialization, not horizontal sprawl. The firm that goes deep in one vertical can install Layer 1 in two weeks against the existing CRM. The firm that goes wide across twenty verticals is always custom-building the integration. Customization is where pilots stall.
The pause clause. A retainer that pauses if the operating layer does not produce the recoverable revenue is structurally different from a retainer that bills regardless of operational outcome. The pause clause realigns the incentive of the firm with the incentive of the operator.
The investor read
If you are a sponsor sitting on top of a pilot that stalled, the correct question is not "should we replace the vendor." It is "was the work that needed to happen ever in scope."
Most of the time it was not.
The integration work that produces the production result is its own engagement. It is sized against the operator's specific recoverable revenue figure. It is governed by the pause clause. It is delivered by a firm whose mandate is the integration, not the model.
STRATA is an AI Operations Integration firm. Book a Revenue Audit at /audit.