How we work

Audit, pilot, scale, then operate

An engagement runs in four stages: a fixed-price audit of about two weeks, a fixed-scope pilot of four to eight weeks on one process, a phased scale-up, and an ongoing managed-operations retainer.

The pilot is the decision point. It is deliberately small, priced up front, and measured against metrics agreed before we write code — so continuing or stopping is a decision made on evidence from your own processes rather than on a vendor's confidence.

01The four stages

Small commitment first

Each stage ends with something you can hold: a document, a working system, or a report.

01AI audit~2 weeks02Fixed-scope pilot4–8 weeks03ScalePhased04Managed opsOngoing
The pilot is the decision point. Nothing scales until measured results from your own processes say it should.
  1. 01

    AI audit

    About 2 weeks · fixed price

    We interview the teams doing the work, map the processes, count the hours each one consumes, and check whether the data behind them is usable. You get a ranked shortlist with an estimated payback for each candidate and a costed plan for the first pilot. The document is yours to keep whether or not you continue.

    Outputs

    • Process map and measured baseline
    • Data and systems readiness assessment
    • Ranked opportunities with estimated payback
    • Costed pilot plan
  2. 02

    Fixed-scope pilot

    4–8 weeks · fixed price

    One process, built and running in your environment, measured against success metrics agreed before we write any code — hours saved, cycle time, error rate, cost per transaction. The scope does not move mid-build. If the pilot misses its metrics you have the evidence to stop, and the cost of stopping is small by design.

    Outputs

    • Success metrics agreed before build
    • Live integration with your systems
    • Human approval steps where they matter
    • Measured result against the baseline
  3. 03

    Scale

    Phased

    We extend what worked to adjacent processes and connect the shared operating layer. Access control, logging, evaluation, and rollback are in place before volume increases, not retrofitted after an incident. Each new process is still scoped and measured individually.

    Outputs

    • Adjacent processes onboarded in phases
    • Shared access control and audit logging
    • Evaluation suite covering each process
    • Documented rollback for every automation
  4. 04

    Managed ops

    Ongoing retainer

    The running of what is live: monitoring, accuracy checks, model and prompt changes, cost control, and improvements as your processes change. You receive a monthly report of what ran, what was escalated to a person, and what it cost.

    Outputs

    • Monitoring and scheduled evaluation runs
    • Change management for models and prompts
    • Cost per process tracked and reported
    • Monthly performance report
02Ground rules

What the contract actually says

The commercial terms behind the four stages, stated plainly.

Scope is fixed before the build starts

The audit produces a written scope and a firm price. Changes are handled as an explicit change request, not absorbed silently into a timeline that then slips.

Metrics are agreed before any code

Hours saved, cycle time, error rate, cost per transaction. We measure the baseline during the audit and repeat the same measurement after go-live so the comparison is honest.

Stopping is cheap by design

Each stage is small enough that abandoning it costs little. You keep every artefact produced up to the point you stop, including the process map and the code.

You work with the engineers

There is no sales team followed by a rotating delivery bench. The people in the audit interviews are the people who build and operate the system.

Stage one

Start with the fixed-price audit

About two weeks, a firm price, and a ranked list of what to automate first — yours to keep whether or not you continue.