How to measure a market nobody counts

When we set out to size the application engineer market, we hit the problem that defines it: there is no number to borrow. No occupation code captures the role, so no official count exists, and the analyst reports that quote market sizes for adjacent software categories are built on those same codes. If we wanted a number we could stand behind, we had to build it, and show the working.

Three legs, one estimate

The sizing triangulates three independent ways of looking at the same workforce. None of them is sufficient alone. Together they bound each other.

1. Top-down: start from official employment pools for the sixteen candidate occupation codes, then apportion each pool by the share of its live adverts that are genuinely this role. This leg sets the central estimate.
2. Bottom-up: count and classify live job adverts directly, domain by domain, after a manual hygiene review. This leg is noisier but independent of official weighting.
3. Register check: bound the result against the population of relevant companies on the register, asking whether the implied engineers-per-firm is plausible. This leg catches runaway estimates.

Rules that keep the numbers honest

The mechanics matter less than the discipline around them. Six rules govern every figure we publish from this research.

• Grade every figure. Each number carries a confidence grade from A to D, and the grade is printed beside the number wherever it appears. A blend of two grades takes the lower one.
• Report the envelope, not just the centre. 27,200 means little without 11,550 to 30,550 beside it.
• Never average independent measurements. Our demand research reports a flow of about 900 adverts a month and a stock of 186 live adverts as two layers. The pair is the result.
• Date every snapshot. All figures in this body of research are stamped 4 June 2026 and will drift from that day onward.
• Pin the conversions. Currency uses the European Central Bank reference rate of 1 GBP = 1.34582 USD on the snapshot date, recorded with the data.
• Write the scope decisions down. 286 of 645 sampled adverts, 44%, were excluded after review, and the exclusion rules are documented so the cut is repeatable rather than convenient.

Why a software company publishes its workings

Partly because we would want the same from anyone selling to us. But mostly because the discipline is the same one our product runs on. Kabaido's AI is built to cite the source of every extracted value and to abstain rather than guess, and a company asking customers to trust that behaviour should hold its own market claims to the matching standard. A graded number you can check beats a confident number you cannot.

The method is also portable. Because it starts from a national classification and a job-advert feed rather than anything UK-specific, the same three legs can size the role in any country that has both. Each new market becomes a dated, reproducible measurement rather than a rebuild, which is how we intend to grow the map.