Engineering for brands whose products carry regulated claims
Find what your business thinks is working — but isn’t.
Your storefront says one thing. Your email platform says another. I inspect the joins by hand, prove what is broken with a reproduction path you can run yourself, and fix what is worth fixing — with your people approving every change.
From one real inspection · 2026
The shop said 5,490 customers could be mailed. The email platform could reach about 150. Nothing anywhere said a word.
One of three findings published here, reproduced before reporting. No campaign has been sent to any of those records, and no customer outcome is claimed.Prefer to describe a stuck workflow instead? One sentence by email is enough.
One named builder: David Bentley · Built in your accounts, not mine · Nothing changes state or sends without your named approver · A documented handover
Evidence you can reproduce
One inspection. Six verified failures, invisible from inside.
One private engagement with a small independent British supplement business, inspected in 2026. Six findings were reported privately, each reproduced before it was reported; five sat between two systems, in a join nobody owned. Three are published, with the path to check each one yourself. The business is not named and never will be: a client’s private weakness is not my marketing asset.
Observed · Reproduced · Not proven — stated on every finding. Read the findings and how they were checked
The method
How an engagement runs
- Analyse. £150 fixed. An agreed public surface of your business, examined from outside. No credentials, no access to your accounts.
- Inspect. £750 fixed. One workflow you name, followed read-only across the systems it crosses, agreed in writing first.
- Build. From £2,500. Scoped and priced in writing before it starts, and only where the evidence justifies building something.
- Operate. From £750 a month. Only where there is a real recurring workload worth operating. Your business keeps the system.
Every step is complete in itself and refusable, with a fixed price stated in writing before any work begins. If the inspection does not justify the work, I decline it. What each step costs, and what it does not buy · how the work runs, in detail
Selected work
Also built: three systems for a €3 million regulated training business
Sofema Aviation Services trains engineers under EASA regulation, where a wrong answer carries a regulatory cost. A training ROI calculator, a course finder and a cited EASA answer system, all built agent-assisted and human-directed. None is in service: all three are platforms in beta, in review with Sofema, and no deployment decision has been made on any of them. The answer system may never ship.
Platforms in beta show what I can build inside a regulated operation. They do not show use, adoption or a result, because nobody is working in them. See the work
The model
Agents produce. One engineer answers. Your people decide.
A directed team of AI agents gives one senior engineer unusual capacity: they read, test, draft and propose inside agreed boundaries; I brief them, verify their work against each other and merge what survives. Judgement does not delegate. A named person in your business approves anything that changes state or reaches a customer. The system of record, source, credentials, workflow definitions and operating records live in infrastructure your organisation controls; where an approved external provider processes data, including an AI model, I document what leaves the environment, why, and who processes it. No dependency is hidden.