Methodology
One rule governs everything below: every number in the score is a count of a record we stored. Nothing is estimated, modelled or projected. If we cannot count it, it does not go in.
The score totals 100. Each component is capped, so no single signal can carry a category on its own.
| Component | Max | What it counts |
|---|---|---|
| People here already buy software | 35 | How many people bought a tool for this job and left a review. Counted only over tools specific to this category, never general business software. |
| Room left on the shelf | 35 | Inverted count of established tools already serving the job. A crowded shelf scores low here. |
| How easy the buyer is to find | 30 | How spread out the hiring is across distinct employers. One giant employer scores lower than forty small ones, because forty small ones are reachable. |
It used to be five. Two were removed on 5 August 2026 and it is worth saying why, because both were removed for failing rather than for tidiness.
Complaint volume and freshness counted reviews of the software people use. An audit found that in roughly 40 of 68 categories we were collecting reviews of the wrong software: patient-facing apps rather than the clinician's chart, exam prep rather than the practitioner's tool, in one case business magazines. Those two components were 35 of the 100 points. Complaints are still shown on category pages as evidence. They no longer move the number.
Market size multiplied how many people hold a job by what they earn. It measures how big an occupation is, not whether anyone buys software for it, and it pushed the largest and most crowded categories to the top of the list. Removing it moved specialist categories back above general management.
Public reviews of the software people already use for the job, collected from app stores and software directories. We keep the verbatim text, the star rating, the product name and the date. Every quote published on this site names its product, because a complaint without a product attached is just a mood.
Live listings where an employer is paying a salary to have this work done by a person. This is the strongest signal in the whole system. A company that already pays someone to do a thing manually has proven both that the problem is real and that it has a budget for it.
Posted contracts with a stated price, which tell us what the work is worth when bought piecemeal.
Products listed under the software category on Capterra, with their rating and review count. We use these two ways: as the crowding measure in the score, and as the named list of who you would be competing with.
US Bureau of Labor Statistics figures for the occupation. These size the market. They are not part of the score.
Two numbers on the category pages are not counts: monthly search volume and cost per click. They come from Google’s advertiser data and are modelled estimates.
They are labelled as estimates everywhere they appear, and they are not part of the score. We show them because what a company will pay for a single click is the clearest available sign that the person clicking has a budget.
BUILD requires two things to be true at once: money is measurably moving toward this work, and the buyer is findable. It does not mean the category is empty, and it is not a prediction that you will succeed.
Stated exactly, BUILD means a score of 70 or above and a reach sample of at least 10 job postings. Both are required. A category that scores 75 on 6 postings is not called BUILD, and its score is published unchanged beside the verdict it did not get.
The second condition exists because the reach component is distinct employers divided by postings, and that ratio inflates when the sample is small: more postings means more chances to see the same employer twice, so the ratio falls as evidence accumulates even where the real hiring spread is identical. Rarefaction over the categories holding a full 25-posting sample puts the inflation at roughly 3 to 5 points out of 30 for samples between 5 and 10. It is a bias with a known direction rather than random noise, so a wider confidence interval would not have fixed it. Added 7 August 2026; no score changed, three categories lost the label. Finding 5 of the Demand Score Index shows the measurement.
NOT YET means the pain is documented but the money is not visibly moving, or the buyer is concentrated in a handful of large organisations. You would be creating the demand as well as the product, which is two jobs instead of one.
It does not tell you what you would earn. It does not tell you whether you can build the thing, whether you can reach the buyer, or whether you would enjoy the work. It is not investment advice and it is not a business plan.
It answers exactly one question: is money already moving toward this problem? That is the part you can check before you build, and it is the part most people skip.
The next step after a score is never more research. It is ten conversations with people who hold the job, and one of them saying yes to paying you.
Scores recompute as new records arrive. Every category page carries the date it was last updated. As of the index covers 68 software categories serving 94 US job titles, built from 5,279 collected signals and 3,298 counted tools.
If a number here is wrong, we want to know, and we will say so on the page when we fix it. Every figure traces back to a stored record, so a correction is checkable rather than a matter of opinion.