Guides

Why most AI pilots never reach production.

8 August 2026 · 5 min read

AI pilots stall because the prototype is the easy fifth of the work. What stops them reaching production is integration with the systems people actually use, cost that scales unpredictably, no owner once enthusiasm fades, and no agreed definition of good enough to ship.

The prototypeRoughly 20% of the work
Most common blockerIntegration
Most expensiveUnmodelled cost

01

A demo and a product are different objects

A demo runs on one laptop, with one person watching, who is forgiving and knows which inputs work. Production runs unattended, on real data, for people who will not tolerate a wrong answer and will not report it either, they will simply stop using it.

Almost every stalled AI project we see is a good demo that nobody costed the distance from.

02

The four blockers, in the order they bite

Prototype works100%Integrated with real systems62%Cost per user modelled41%Owner and support agreed27%Live with real users18%
Where AI projects lose momentum between demo and production
01Integration. The pilot used a spreadsheet export. Production needs to read and write the system people actually work in, and that is where the real effort sits.
02Cost. Nobody modelled what it costs per user per month, so the finance conversation arrives after the build rather than before it.
03Ownership. The enthusiast who built the pilot has another job. Without a named owner and a support arrangement, it decays quietly.
04Definition of good enough. Without an agreed accuracy threshold and a plan for what happens when the model is wrong, there is no moment at which anyone can say it is ready.

03

What to do if you already have a stalled pilot

Do not restart it. Scope the production version specifically: which system it must integrate with, what it costs per user, what accuracy is acceptable and what happens when it falls short, and who owns it afterwards.

Then treat the pilot as what it was, a sales tool that won the decision internally, rather than as a foundation to build on. Rebuilding the working part is usually faster than adapting a prototype that was never meant to carry load.

Common questions

We have a working prototype. Can you take it to production?

Usually, and the first step is a short review of what production actually requires: integration, cost per user, accuracy thresholds and ownership. Sometimes the prototype is a good foundation and sometimes rebuilding the core is faster. We will tell you which.

How accurate does an AI feature need to be?

It depends entirely on what happens when it is wrong. A feature that drafts something a human reviews can tolerate far more error than one that acts unattended. Agreeing that threshold in advance is what makes shipping possible.

Why does cost stop projects specifically?

Because it scales with usage rather than headcount, so a feature that is cheap in a pilot with ten users can be alarming at a thousand. Modelling cost per user before scaling turns that from a surprise into a decision.

Get a stalled pilot moving

Tell us what you built and where it stopped. We will tell you what production actually requires and what it costs to get there.

Start a conversation →