How we work
How we work as your fractional AI team
Large companies run AI through an internal team: they collect ideas, rank them, build, train and measure. We run that for you.
Diagram: the six stages of the cycle, Intake, Prioritise, Build, Review, Adopt and Measure, drawn as upright threads labelled Your team. A second thread, labelled The AI, runs across all six, passing over Intake, under Prioritise, over Build, under Review, over Adopt and under Measure.
The cycle we run each month
Intake
Led by our business analyst
Our business analyst collects the tasks your team finds slow, repetitive or error-prone, from the people who do the work. Ideas go into one backlog, however small, with who owns each one.
Prioritise
Led by our business analyst, decided by you
Each item is ranked by value, risk and effort. Your owner decides what goes first. Some items won’t need AI at all, and we’ll say so.
Build
Led by our AI engineer
Our AI engineer builds the chosen item and connects it to the systems you already use, with guardrails and a person in the loop for anything that matters.
Review
Led by our AI engineer, with your champion
Before anything goes live we test it against examples from your own work and agree a pass mark with the person who would normally check the output.
Adopt
Led by our business analyst
Our business analyst trains the team, writes the short guide and works with your champion until people are using it as part of their normal week.
Measure
Led by the whole team
We track usage, quality and running costs, tune what isn’t working, and bring the next item from the backlog.
Who does what
Three roles. The first two are ours; the third is yours, and it is what makes adoption stick.
- Business analystDiscovery, backlog, adoption and training
- Maps how the work really happens, keeps the backlog, trains your team and checks that people are using what we build.
- AI engineerBuild, guardrails, evaluation and running
- Builds the systems, tests them against your examples, keeps models current and runs them after go-live.
- YouOne owner and one champion per team
- One person who decides priorities, and one champion in each team who tests what we build and helps colleagues use it.
The monthly report
Each month you get a short written report, so you can see what the retainer is doing.
- What shipped
- What’s in progress
- How much it is being used
- Issues we found and what we did about them
- Running costs
- Next month’s pick from the backlog
How we test accuracy
We describe accuracy in terms you can check, using your own examples, before anything goes live.
A test set from your own work
Past examples, de-identified where possible, with the answers your own people would give.
Your reviewer’s checklist, as scoring
The checklist your reviewer already uses becomes the scoring criteria for every output.
Grounding and source checks
Outputs are built to link to the document, email or page they rely on, and anything without a source is flagged.
A shadow run before go-live
The system runs alongside your usual process until it meets the agreed pass mark.
A monthly quality report
The same test set, run every month, so any drift shows up early.
What it costs
Discovery has a fixed fee, agreed before it starts. The monthly retainer is quoted once we know what is in your backlog.
- AI discovery
- A fixed fee agreed before it starts, credited toward your retainer. It usually takes 2–3 weeks and ends with a ranked backlog.
- Monthly retainer
- Quoted after discovery, based on how much we take on each month.
- Running costs
- AI model usage and AWS hosting costs are estimated before we build, and how they’re billed is set out in your proposal.
- What you keep
- The process map, the backlog, the test sets and the documentation we produce for you are intended to be yours, as set out in your client agreement.
An illustrative first quarter
A 40-person distribution business. Weeks 1–3: discovery, ending with a ranked backlog. Month 1: supplier-invoice checks. Month 2: customer-enquiry triage. Month 3: a staff knowledge assistant for policies and procedures.
Questions about working with us
Can we stop after discovery?
Yes. Each step is agreed separately. The process map and the backlog are intended to be yours whether or not you go ahead, as set out in your client agreement.
Who owns what you build?
The process map, test sets and documentation are intended to be yours, and ownership of code and configuration is set out in your client agreement.
What if our data is messy?
Most businesses’ data is. Discovery shows where it matters, and each build is scoped around documents as they really are, not as they should be.
Do you work with our IT provider?
Yes. We agree access, accounts and security settings with them, and give them the system description and runbook.
What happens if someone on the team is unavailable?
The whole team works on every engagement, and each system comes with a system description and a runbook, so it doesn’t depend on one person.
Next step
Start with AI discovery.
A fixed-fee discovery, credited toward your retainer. A 15-minute call first, with no preparation and no client data.