How to review an electrical thermography report before approving repairs

7 min read · Updated 30 July 2026

Somewhere on your desk is a thermography report, and stapled to it — sometimes literally — is a quotation for repairs. The report is the evidence; the quotation is the decision. This article is about making sure the first actually supports the second, using checks any manager can run without owning a thermal camera.

One thing first, in fairness to the trade: most reports are honest, and a good one passes everything below in five minutes. The checks exist because the incentive problem is real — when the surveyor and the repair contractor are the same company, the report is also a sales document — not because every vendor abuses it.

Check 1 — Was the load recorded?

Heat at a connection rises with the current through it. A loose joint carrying 30% of its normal load can read almost nothing on camera and still be a serious fault at full load; the reverse error also happens, where a fully loaded but healthy circuit gets flagged because nobody noted it was working hard. So the first thing to find in the report is the operating condition at scan time: what was running, and ideally the measured current.

If load is not recorded anywhere, every severity rating in the report is standing on air. That alone does not make the findings wrong — it makes them unverifiable, which for an approval decision is the same problem.

Check 2 — Compared against what?

A temperature means little alone; the diagnosis lives in the difference — usually against the same component on a neighbouring phase under similar load, sometimes against ambient. A defensible report states which reference each finding used, and states the classification logic: what temperature rise puts a finding in "monitor," what makes it "serious," what makes it "act now."

The logic should be printed in the report, not held in the vendor’s head. Our own published case study shows what that looks like in practice: a severity table with its delta-T bands stated, so the client can check the classification themselves. Any competent provider can do the same; the question is whether yours did.

Check 3 — Can you find the component?

Each finding should carry two images of the same point: the thermal image and an ordinary photograph, labelled well enough that your own electrician could walk to the exact terminal. A thermal image alone is close to useless for the repair step — rainbow blobs do not carry asset numbers. Reports that show dramatic thermal images without their visual twins are showing you heat, not findings.

Check 4 — Is the recommendation proportionate?

A rated finding should map to a proportionate next step: monitor and re-scan, repair at the next planned shutdown, or act immediately — and in an honest report those categories are all used. The pattern to distrust is uniformity: when every anomaly, whatever its rating, points to the same conclusion — urgent, quoted, and attached — the report has stopped rating findings and started justifying the quotation.

Run the trace both ways: every quoted repair item should point back to a rated finding, and the money should follow the severity. A repair scope noticeably wider than the findings that justify it is the single strongest signal to pause.

Check 5 — Are the limits stated?

No survey sees everything. Covers that stayed closed, circuits that were not energised, areas skipped for access or safety — a defensible report says so, because the gaps define what the report can and cannot support. A report with no stated limitations is not claiming a perfect survey; it is usually just not claiming anything carefully.

The same goes for the person behind it: someone identifiable should take responsibility for the findings, with some stated basis for their competence. Evidence your own auditor or insurer would accept has a name on it.

The five checks, on one card.

What to do when a report fails a check.

Failing a check does not mean the findings are fabricated — the most common cause is a rushed template, not fraud. It means the report cannot yet carry the decision stapled to it. Three reasonable moves, in escalating order: ask the vendor for the missing element (load data and comparison logic usually exist even when they were not printed); ask for the raw radiometric images, which any professional camera stores; or put the report and the quotation in front of someone with no stake in the repair work.

Our interest, declared: that last one is a service we sell — a written second view of the report, images and quotation, with an honest boundary on what a desk review can conclude. But the five checks above are free, they are yours, and most of the time they are all you need.

And approving the repairs is only half the discipline. Once the work is done, the finding needs closing with the same standard of evidence — we have written separately about how to confirm a hotspot was actually repaired.