Field-work documentation: how the workflow works

Three sources, two machine numbers

A coordinator receives a photo, a voice message and a job card in PDF from the site. The card gives one machine number, and the technician says another in the recording. The photo shows the machine and liquid beneath it, and the nameplate is not visible. An ordinary summary would pick one number and hide a conflict that may matter for safety.

In our demonstrator each file enters the case file separately and keeps its source, time and author. The system compares the shared fields and shows the differing values side by side. An open conflict blocks closing the case file. The coordinator chooses a value present in the evidence, picks the basis from a closed list and signs the reasoning. The system does not identify the machine or the source of the liquid, and it does not release the machine for work.

What a field evidence app costs and when it pays back

The count covers the work around a case: searching, reading, writing and passing it on. What stays is the time to check the output, 30% of today's work in the scenario. Before we quote, we measure both times on 20 to 100 cases your team has already handled, in a €0 test that takes 5 working days.

We leave 30% of today's work in the count because every conflict goes to the coordinator. According to the European Commission (2026), 90% of companies using AI say people take the decisions, and 2% of decisions are fully automated. In field cases this split follows from safety.

Why the transcript stays a suggestion

Ten transcription samples of the same recording returned two texts that differed in punctuation. That is why the case file keeps the audio and the original file, and the text is an aid for the coordinator. No reading of the photo settles the machine number when the nameplate is not in the picture. We write such limits down before the pilot, together with the rule that a model's suggestion never starts a machine and never replaces an inspection. Anthropic's guide to evaluating agents recommends running the same task many times, because a model's output changes between trials.

Where to start

The pilot covers a few types of event with an agreed result, a decision owner with acceptance criteria and material from real phones. To prepare: rules for storage and access. We build this kind of rollout as part of AI apps. For the free process review, a description of one case from recent weeks and the tools it went through is enough. You do not need to send confidential documents, and you get a written plan within 2 working days.