For when
“This recurring process consumes too much time.”
Usually commissioned by: Department head, process owner, CTO.
What you get
- Deployed workflow integrated with the agreed systems
- Evaluation suite and acceptance tests that run again after every change
- Access controls, logging and a human-review step where needed
- Operating runbook, hand-over and a defined support boundary
Acceptance condition
The agreed quality, security, operational and usability checks pass in the buyer's environment.
How it runs
- 1
Written scope with acceptance criteria and a change process
- 2
Build in short increments, demoed on real data
- 3
Acceptance testing with the process owner
- 4
Go-live, hand-over and a hypercare period
It may conclude…
If the scoping shows deterministic automation is enough, we build that instead. It is usually cheaper to run.
Not included
- Open-ended product development
- Run-time hosting costs (passed through at cost or billed to you directly)
Read before you decide
Free guides that cover the same ground. Plenty of teams get what they need from these alone.
What should require human approval for AI agents?
Decide which AI agent actions need human approval by mapping reversibility, blast radius and confidence, then choosing an approval pattern and controls.
Writing an AI implementation brief, and the questions to ask suppliers
What an AI project brief must contain, a skeleton you can copy, and the questions that separate credible AI suppliers from confident ones.
How to automate recurring client reports and packs (and check AI-written summaries)
Split report automation into data assembly, narrative drafting and review, trace every figure to source, and check AI-written summaries before release.
Discuss a Workflow Implementation
Send a short project brief. A person reads it and replies with whether this engagement fits, a different one would, or none is needed.