TL;DR: We searched 1,064,011 scored discussions for pain statements that describe work still being done by hand. 32,688 of them do — 3.07% of the corpus. Those complaints carry payment evidence at 1.7 times the corpus rate. But the ones worth building are not where the volume is: developer tooling produces the most manual-work complaints and the least paying intent, while e-commerce operations, CRM and marketing automation produce a fraction of the volume at two to three times the rate.
What we counted, and how carefully
The claim "people still do this by hand" is easy to assert and easy to fake with a loose keyword search, so here is the method.
We matched pain statements — the extracted problem sentence, not the surrounding analysis — against phrases that describe manual labour: manually, by hand, copy and paste, copy-paste, spreadsheet, Google Sheets. That returns 32,688 discussions, 3.07% of the corpus.
A first, looser pass matched the same phrases anywhere in the document, including the analysis prose, and returned 44,661. Reading a random sample of fourteen of those, four turned out not to describe manual work at all — the phrase had matched a sentence about something else. So we tightened to the pain statement, re-sampled eighteen at random, and read the specific statement that matched in each one. All eighteen described actual recurring manual work. The 32,688 figure is the one we use; the 44,661 is not.
We mention this because the cheap version of this article uses the bigger number.
Manual work carries more payment evidence — but not much more
| Share scoring 7+ on willingness to pay | |
|---|---|
| Discussions describing manual work (32,688) | 23.6% |
| Corpus baseline (1,055,233) | 13.6% |
A 1.7x lift is real but modest. "It is done by hand" is a useful filter, not a qualifier on its own. Roughly three in four manual-work complaints still show no evidence that anyone would pay to remove the labour — usually because the person doing it is the person who would have to buy the tool, and they are absorbing the cost in their own time.
The narrower cut is more interesting. Of the 32,688, 2,192 also score 8 or above on willingness to pay. That is the list worth reading.
Where the volume is, and where the money is — different places
Grouping the 32,688 by product category:
| Category | Manual-work discussions | Share scoring 7+ on willingness to pay |
|---|---|---|
| E-commerce Operations | 122 | 48.4% |
| CRM | 176 | 46.6% |
| Marketing Automation | 540 | 40.4% |
| Workflow Automation | 471 | 34.4% |
| AI Productivity Tools | 308 | 27.3% |
| DevTools | 3,954 | 18.2% |
| Social Media Management | 418 | 9.8% |
| — corpus baseline — | 1,055,233 | 13.6% |
DevTools contributes more manual-work complaints than the other six categories combined and has the second-lowest paying rate on the table.
The explanation is not that developers do not value their time. It is who pays. A developer complaining about manually setting up local environments, manually parsing pod logs, or manually maintaining seventeen dependency-update pull requests is describing their own working day — and developers are famously willing to spend a weekend automating something rather than expensing it. The same task shape inside a marketing or e-commerce team lands on someone whose employer buys tools, and whose manager can see the five to ten hours a week going into it.
Boring beats exciting, and back-office beats developer-facing. Both of those are visible in the same table.
What the complaints actually describe
The vocabulary in the 32,688 pain statements is consistent. Across 4,000 of the matching statements, the most common substantive terms are time-consuming (928), developers (617), multiple (516), error-prone (438), spreadsheets (372), tracking (290), copy-paste (219), hours (178), repetitive (168) and tedious (157).
Read those together and the shape is specific: a recurring task that spans multiple systems, where the human is the integration layer. Not one tool being bad — several tools that do not talk, with a person and a spreadsheet in the gap.
Representative shapes from the scored material, paraphrased:
- Compiling client reports every week from an ads platform, an SEO tool and three social accounts.
- Reconciling purchase records against a tax filing in a spreadsheet, line by line.
- Re-tailoring the same résumé for every application because no two job descriptions use the same words.
- Requesting company lists by hand because the sales stack has no live source.
- Keeping files in sync between a laptop and a desktop by moving them manually.
- Comparing costs across twenty-plus model providers in a sheet that goes stale immediately.
Each one names a task, a frequency and a workaround. That triple is what separates a manual-work signal from a generic complaint.
How to use this
- Filter for the workaround, not the feeling. Search for "manually", "by hand", "copy-paste" and "spreadsheet" inside the problem statement. A complaint without a named workaround is usually a product review.
- Then filter for who pays. Prefer categories where the person doing the work is not the person who would have to build the alternative. That single distinction explains most of the table above.
- Look for the multi-system seam. The recurring pattern is a human bridging two or three tools. The product is the bridge, not a replacement for either tool.
- Treat a spreadsheet as a scoreboard. It means the job is real, recurring, and currently unowned by any vendor — the clearest possible evidence that a budget-holder has already decided the task matters.
- Check the paying rate before the volume. The biggest pile of complaints in this dataset is also the one least likely to convert. More on that method in how to tell if a pain point is worth building for, and on category selection in what a solo developer can actually build and sell.
You can run these filters against the dataset in the validator, or browse categories in the dashboard.
Honest caveats
- 3.07% is a floor, not a census. It counts discussions whose extracted pain statement uses the vocabulary of manual work. Plenty of manual work gets described without those words and is not in the 32,688.
- The categories are model-generated labels, not a fixed taxonomy — the corpus contains 220,073 distinct ones. Categories with a few hundred discussions, like E-commerce Operations at 122, have correspondingly wide error bars. The direction of the gap is robust; the exact 48.4% is not.
- Willingness to pay is inferred from text. It tells you where to look. It is not revenue.
- Manual work is sometimes the correct answer. Some of these tasks run once a quarter and take twenty minutes. Frequency is doing as much work in this analysis as the word "manually" is.
Where the data comes from
PainHunt processes discussions from 22 active sources — developer forums, app store reviews, product communities, job boards and social platforms — and scores each for pain intensity, commercial intent and willingness to pay. Figures here are from 1,064,011 processed discussions as of 2026-09-20. We link back to every original discussion and never republish its text.