How we work

Scope, then Build, then Run.

Three engagements. Each one ends with something you can check and a decision that is yours: whether to spend more money, add more people, or give the system more autonomy.

Book a call Get the Diagnose toolkit

The three engagements

Select · Diagnose

1 Scope

Two weeks. We help you choose the workflow, map it as it runs today, and measure it.

Ends with: An agreed baseline and a recommendation

Design · Prove

2 Build

We redesign the workflow, build the system, connect it to the tools you use, and test it.

Ends with: A working system and its test results

Pilot · Decide

3 Run

Your team uses it on live work, with training, monitoring, and named owners.

Ends with: Results against the baseline and your decision

What each engagement includes →

The six stages underneath

Step	Stage	Question answered	What must be true to move on
Scope	1 Select	Which work matters most, and is it ready?	Named owner, measurable outcome, usable data
Scope	2 Diagnose	How does the work really happen, and where is the bottleneck?	Baseline agreed with the business owner
Build	3 Design	What is the lowest level of automation that solves it?	Design approved by the owner
Build	4 Prove	Does it meet the acceptance criteria agreed in advance, and beat the non-AI option?	Criteria met on the test set, zero unacceptable errors
Run	5 Pilot	Does it work for real users under enforceable controls?	Used independently, no unresolved critical incidents
Run	6 Decide	Did the value arrive, and what happens next?	Scale, iterate, or stop, with reasons

Each gate is a decision point. You can stop after any stage, with the evidence in hand.

Setting the standard

The baseline and the acceptance criteria are agreed together.

During Scope we measure how the work performs today and agree on that baseline with the person who owns the result. Before any testing, we propose acceptance criteria from it: what counts as good enough, and which errors are unacceptable. You know the work and the risk. We know what can be tested. Neither side sets the standard alone.

What a pass means. A pass shows the system met the agreed standard on the agreed test set. It is not a guarantee that it will never be wrong on live work. AI output can be inaccurate, which is why our builds keep a person in review on consequential actions, and why monitoring continues through Run.

Three principles

Rules before AI

Every step is tested against four questions before any AI is considered. Most stop at the first or second. Across our four documented builds, 59% of steps run on rules.

Criteria agreed first

What counts as good enough is written down before testing starts, so the test result means something.

Named owners

Business, AI operations, engineering, and risk each own a part. Any of them can pause the system.

Key terms

AI operations consulting: Improving how operational work gets done: redesigning a manual, document-heavy workflow, building the software to run it, and helping the team adopt it. It is not IT monitoring or infrastructure management.
Baseline: How the work performs today, measured during Scope and agreed with the business owner, so results can be compared against it.
Pass mark: The acceptance criteria a system must meet before it goes live. We propose them from the baseline and agree on them with you before testing starts.
Unacceptable error: An error named in advance as one the system must not make in testing. A build that makes one has not passed.
Test set: The examples a build is tested against. Passing shows the system meets the agreed standard on those examples. It is not a guarantee of error-free operation afterward, which is why review and monitoring continue in Run.
Rules before AI: Every step is tested against four questions before AI is considered. Most stop at the first or second.
Scope, Build, Run: The three engagements: a two-week fixed-fee sprint, a fixed-fee build, and a flat monthly run.

What you receive along the way

When	What lands on your desk
End of week 2	Ranked shortlist, map of the work, agreed baseline
Design	Signed-off design, with each step marked rules, AI-assisted, or human
Prove	Test results against the agreed acceptance criteria
Pilot	Your team running it, with training and monitoring in place
Decide	Results against the baseline and a decision memo

Diagnose toolkit

Understand the work before changing it.

A working example of what Scope produces. Five tools take you from how people say the work happens to how it actually happens, and then to a step-by-step call on rules, AI, or a person.

Interviews. The intended process, and who is involved
Observation. The real process, workarounds included
Data inventory. Where each input lives and whether it can be trusted
Step map. Each step defined precisely enough to automate
Classification. Rules, AI-assisted, or human, step by step