# What is a six-week AI pilot, and what should it produce?

A six-week AI pilot is a fixed-length engagement that builds one working version of an AI system on a client's own data and systems, tests it with the people who do the job today, and ends with a decision: go to production, change the approach, or stop. It is the second stage of FetchSky's AI process, after a free consultation that picks the task and before roll-out. The output is not a slide deck or a demo on sample data. It is software that ran on real records, a measured result, and a production plan with scope and cost.

By Muhammad Ali Khoja, CSO. Published 2026-10-17.

## Why six weeks?

Six weeks is long enough to build on real data and short enough that a wrong bet costs little. Proofs of concept that run for months tend to drift into production work without the controls production needs. A six-week window forces the use case to be narrow, which is also what makes it safe to put live. FetchSky agrees scope and cost before the pilot starts, so the six weeks has a fixed price and a fixed question to answer.

## What happens in each week?

| Week | Focus | What exists at the end |
| --- | --- | --- |
| 1 | Pick the use case and secure access | One task chosen; service account and data access granted |
| 2 | Map the workflow and choose the approach | A step-by-step map of how the work is done today; model and integration approach selected |
| 3–4 | Build on real data | A working version processing actual records from the client's systems |
| 5 | Test with the people who do the work | Measured accuracy, time saved and failure cases, judged by the operators |
| 6 | Production plan and go/change/stop | A plan for roll-out with scope and cost, and a decision |

Week 1 is where most pilots are won or lost. If access takes three weeks to arrange, there is no pilot. Week 2 produces the workflow map that the build is checked against. Weeks 3 and 4 build against the client's own systems, not a copy. Week 5 puts the working version in front of the analysts, agents or clerks who do the task, because they will find the failure cases nobody else can. Week 6 turns what was learned into a roll-out plan.

## What should a six-week AI pilot produce?

Four things. First, one working version of the AI system, running on the client's data inside the client's environment. Second, a measured result against the baseline: hours per day, error rate, turnaround time. Third, a log of everything the system read, decided and did during testing, because logging is a FetchSky control and the pilot is where it gets proven. Fourth, a production plan covering scope, cost, access, monitoring and who reviews what.

The brokerage trade-confirmation automation shows the shape of a measured result: 10 agents spending 2 hours daily became three AI agents, with 98% less manual effort and roughly 20 person-hours saved a day. A pilot should be able to state its result that plainly.

## Who is needed from the client?

A pilot needs three people, not a committee. A sponsor who can approve access in week 1 and make the go/change/stop call in week 6. A data or systems owner who can grant scoped access and answer questions about where fields come from. And two or three of the people who do the work today, available for part of week 2 and most of week 5. FetchSky's engineers do the building, often as forward deployed engineers working alongside the client's team.

## What does "stop" mean at the end of a pilot?

Stop means the measured result did not justify roll-out, and the pilot has shown why. Perhaps the data was less consistent than expected, or the time saved was smaller than the cost of review. A stop is a good outcome compared with a roll-out that fails in month three. Because the six-week pilot comes before any larger commitment, stopping costs the pilot and nothing more. "Change" means the task is worth doing but the approach needs adjusting, for example switching models for cost or data-residency reasons, and the plan reflects that. The full process is laid out in [the AI in production guide](/insights/ai-in-production-guide).

To scope a pilot for one task in your business, start with the free consultation on [AI transformation](/ai-transformation).

## Read next

- [Enterprise AI in production: the complete guide](https://fetchsky.com/insights/ai-in-production-guide)
- [AI transformation](https://fetchsky.com/ai-transformation)
- [How three AI agents cut a four-day report to three minutes](https://fetchsky.com/insights/how-three-ai-agents-cut-a-four-day-report-to-three-minutes)
- [Why most enterprise AI pilots never reach production](https://fetchsky.com/insights/why-enterprise-ai-pilots-never-reach-production)

Source: https://fetchsky.com/insights/what-is-a-six-week-ai-pilot