A signature on a safety record should mean a person stood behind it. For as long as inspections have been written down, that has been the deal: the record carries a name, and the name carries the responsibility.
As AI arrives in inspection and compliance, that sentence is going to be tested.
The fluent wrong answer
Across the industry, software is getting better at producing text that looks like a finished report. Fluent. Formatted. Plausible. In most lines of work that is a convenience. In a safety-critical environment it is exactly the risk.
A confident wrong answer inside a compliance record is not a glitch. It is a liability. When a record is challenged, nobody asks how advanced the software was. Regulators, insurers and auditors ask who the inspector was, where the inspection was carried out, when the record was captured, whether it has changed since, and who approved it. “The model said so” is not an answer any of them accept.
A record nobody reviewed is a record nobody can defend.
Start with evidence, not data
Most AI starts with data and asks what it can infer. We start at the beginning.
Every TickAudit record is captured at the point of work, by the person doing it, with GPS position, timestamp, photograph and signature attached as it is made. That is not a feature we added for AI. It is how the platform has worked since the first inspection, because it is what makes a record defensible when someone questions it two years later.
It also turns out to be the only honest foundation for intelligence. The sequence runs in one direction: real-world activity, then verified evidence, then an operational history that grows around every asset and site, then context, and only then anything that deserves the word intelligence. Skip a step and what you have is a fluent guess with a logo on it.
AI is only as trustworthy as the data beneath it. We built the data layer first, and we have published that sequence on our homepage before anything else ships.
Four principles, published before anything ships
TickAudit AI™ is built on the evidence TickAudit already captures: verified real-world activity, operational history and context. These are the principles it is built on, and we are putting them in writing now so that we can be held to them later.
Trust before intelligence. The data foundation and the provenance of every record come first. No user-facing AI ships before they exist and are provably reliable.
AI drafts, humans commit. Nothing generated becomes a defensible record without logged human approval. A draft is a draft until a person with their name on the line has read it, changed it if needed, and approved it. The original evidence is never altered.
The client’s model, on the client’s data. Your records are yours. We never sell, share or monetise operational data, and your records are never used to train a model that serves another customer. Anything beyond delivering your own service is opt-in only. In regulated accounts the data often cannot legally leave the client’s control, so we designed for that before building anything. The details are on our Trust page .
Attack the return we already deliver. The job of intelligence here is not novelty. It is more reporting hours eliminated and risks surfaced earlier, on the same records teams already capture. On one Scottish Water programme, structured capture alone generated 2,259 reports automatically and eliminated 4,000 reporting hours, with no AI involved. Those are reported outcomes from a single client programme , and they set the bar for what intelligence on top of that foundation has to beat.
The inspector’s judgement is the point
I spent the first part of my career making critical systems reliable: Unix clusters that could not be allowed to fail, complex systems that needed to be simple to manage and to use. The lessons were the same every time. Reliability is not a layer you add at the end. It is a property of how something is built. Simplicity is not a nice to have, it is what makes a system work in the real world.
The same is true of trust in a safety record. The people who capture it, at height, in a tank, on a bridge deck with no signal, are the ones whose judgement makes it worth anything. Any tool that sits beside them should protect that judgement, not replace it, and should make it easier to show, years later, that a person stood behind the record.
That is what we mean by Trusted. Actionable. Intelligence. Trust comes first. Usefulness comes second. Only then does intelligence mean anything.
The foundation is live today. The decision-support is what comes next, and it only ships once that foundation is provably trustworthy.
Where the physical world becomes digital truth.
Gerald Peoples is Founder and Chief Executive Officer of TickAudit. See the foundation it is built on , or request a demo on one of your own inspections.