Service · AI consulting
AI consulting that measures before it recommends
A two-week AI workflow audit that shadows the real process, records a baseline and ranks every AI opportunity, from the team that then has to build it.
Leaders under pressure to “do AI”
Who want evidence before they fund a programme.
Teams with scattered experiments
People using AI on personal accounts, with no inventory and no baseline.
Operations with obvious manual work
Where the question is which process to automate first.
Buyers comparing vendors
Who need a number of their own to judge proposals against.
What you get,in writing.
- A time-and-error baseline for the department
- Every AI opportunity scored on return, feasibility and data readiness
- A ranked plan with the reasoning attached
- A draft you review before it is final
- Accuracy thresholds tied to the baseline
- A recommendation of what not to automate
- 1
Scope
Two days on scope and access.
- 2
Shadow
Four days shadowing and measuring the real process.
- 3
Data
Two days on data and systems.
- 4
Scoring
Two days scoring every opportunity on return, feasibility and data readiness.
- 5
Report
Two days on the written report and a walkthrough.
Why measure first
The word audit gets used loosely in AI sales. Most of what is sold under that name is a day of whiteboarding and a proposal at the end. Nothing gets measured, so nothing can later be proven.
We record minutes per case, cases per week, how often work has to be redone and what a redo costs. Measured, not estimated in a meeting. That baseline is the most valuable thing in the report, even if you never hire us to build anything. It lets you judge every proposal you receive afterwards, including ours, against a number your own team recorded.
- Return. How much time and error the change would actually remove.
- Feasibility. Whether today's AI can do the task reliably enough.
- Data readiness. Whether the data exists, is clean and can be reached.
What we tell you not to automate
Integration cost is largely fixed and does not shrink with volume. A process worth ten hours a month rarely repays the weeks of engineering and the maintenance after it, however painful those ten hours feel. A good audit says so plainly.
We also look at whether a workflow is ready to be automated at all. Is it documented? Does it run often enough to measure? Is there one person who owns the outcome? These questions block more AI projects than the model ever does, and they are the least likely to be asked.
From audit to a working system
If you move to a build, we scope the first one to a single workflow, so a result arrives while the budget and attention are still there. A single workflow typically reaches production six to ten weeks after the audit, with most of that time going to integration and evaluation rather than the model.
Every automation has a line: above a set confidence it acts, below it a person decides. We set that line against the error rate the audit recorded, so it is calibrated against how often people get it wrong today. That is the same principle behind PreOp Screen, where the practice's own rules decide and the doctor signs, and Brands10x, where every plan waits for the owner's approval.
Scoping done by the people who then have to ship the thing stays honest, because an unrealistic recommendation becomes our problem a few weeks later.
Built with this service
All case studies →Common questions
How long does the AI workflow audit take?
About two weeks inside one department: two days on scope and access, four on shadowing and measurement, two on data and systems, two on scoring, and two on the report and walkthrough.
How long is an AI readiness assessment?
Typically two to three weeks, depending on how many business units are in scope.
What does it cost?
A fixed price agreed after a short scoping call.
Do we have to hire you for the build?
No. You keep the report and the baseline either way, and can use them to judge any vendor.
How soon could a first automation be live?
A single workflow typically reaches production six to ten weeks after the audit ends.
Will you tell us if AI is the wrong answer?
Yes. Some processes are not worth automating, and the report says which ones and why.