Client work · Expert assessments
A specialist check that took an expert hours now takes 6–18 minutes.
Illustrations with invented values — not the client’s software or data.
The problem
Before a customer can move to the cloud, every custom report they rely on has to be assessed: what it does, whether the standard product now covers it, and whether to migrate, rebuild, redesign or retire it. One expert did this by hand, taking hours each — and there were 189,127 of them across 3,267 customers.
What we built
An AI analysis platform that prepares each assessment for the expert: what the customization does, how it is used, and a recommendation with its reasoning and sources. The expert reviews it and makes the call.
What changed
Hours of an expert’s time per assessment6–18 minutes per assessmentKnowledge stuck on one expert’s laptopA shared platform their consultants are pilotingReasoning that was hard to checkEvery result checked against 62 expert-reviewed examples
The expert still makes every final decision. Nothing changes itself.
Recorded pilot runs: 6–18 minutes and $1–$11 per analysis. This is per-analysis usage cost, not the project price or a guaranteed result for another workflow.
The starting point
Thousands of their customers each carry years of custom code and reports that must be assessed one by one before moving to the cloud product: what it does, whether the standard product now covers it, and whether to migrate, rebuild, redesign or retire it.
That assessment lived on one expert's laptop. Hours of specialist effort per customization. Inputs arrived as spreadsheets dropped into folders, and credentials were shared over chat.
With 189,127 report customizations indexed across 3,267 customers, the backlog was never going to be cleared by hand.
The delivery record
What we found, what we built, and where an accountable person stays in control.
- What we found
One expert, one laptop, hours per assessment, and a six-figure customization inventory
observed- What we built
An Azure platform: portal, job queue, and a container fleet running headless AI agent sessions with 13 hand-written analysis skills
working- Evidence discipline
Every verdict validated against a 60-field contract derived from 62 expert-reviewed examples, with full run traces kept for audit
working- Human control
Machine-proposed patterns need human approval before they can steer a run. Nothing self-modifies. Only a person may declare code unused.
human- Measured outcome
Hours of expert effort became 6 to 18 minutes and $1 to $11 per analysis. Zero throttling across 892 recorded requests.
measured- Status
In live pilot with the company's consultants. Four testers' feedback already folded into shipped improvements.
in pilot
Client and employer names, identifying details and proprietary screens are withheld. The illustrations above use invented values. This larger engagement is separate from the five-day free proof.
Read the factory records story →
Your project doesn’t need to start this big.
We start with one workflow and a working proof in five days — $0 for selected workflows, no commitment.