12 One-Person Businesses Claude Backs for 2027
Benchivo asked Claude for one-person businesses worth starting in 2027, under a hard rule: no employees, no outside capital, and it survives the owner taking two weeks off.
A model was asked, and this is what it said. Its opinion, not a measurement. Asked .
- Model
- Claude Opus 5
Benchivo asked Claude which one-person businesses are worth starting in 2027. Claude imposed its own filter before answering: no employees, no outside capital, and the owner can stop for two weeks without the business dying.
That last condition does most of the work. It eliminates the majority of solo advice currently in circulation, including nearly every version of “start an AI agency”.
Quick answer
Claude’s top five one-person businesses for 2027 are: a narrow evaluation consultancy, a model migration contractor, deposition-grade media forensics, vertical dataset licensing, and an automation retainer for a single trade. The common thread is that each one sells a judgement a client cannot safely make alone.
How Benchivo ran this
Claude was asked one question in one conversation, with no examples to react to. It was told to rank, be specific, say who each business suits, rate the difficulty, explain why it might be underestimated, and give the reason it could fail.
Claude was not shown ChatGPT’s answer and ChatGPT was not shown this one. Both were asked the same question separately on 28 August 2026. The companion piece is ChatGPT’s 12 one-person businesses for 2027.
These are opinions, not measurements. What Benchivo measures lives in the tests.
The 12 one-person businesses Claude picked
1. Narrow evaluation consultancy
What it is. One person who can prove whether a model is fit for one regulated task. Why Claude picked it. Day rates stay high because the alternative to a proper evaluation is a lawsuit. It ranked this first for that reason alone. Best for. Domain experts with methodological discipline. Moderate difficulty. Main risk. Engagements are lumpy, and lumpy income is what kills solo practices.
2. Model migration contractor
What it is. Watching for vendor deprecation notices and selling the port. Why Claude picked it. The trigger is external and repeats forever, so the business never has to create its own demand. Best for. Technically capable generalists. Moderate. Main risk. The migration gets automated by the vendor causing it.
3. Deposition-grade media forensics
What it is. Certifying whether a recording is authentic, in a report that survives cross-examination. Why Claude picked it. Courts and insurers need a named human expert, and the supply of credible ones is thin. Best for. People who can build genuine forensic credibility. Involved. Main risk. Detection tooling commoditises and the expert premium goes with it.
4. Vertical dataset licensing
What it is. One person, one narrow domain, one maintained dataset, sold annually. Why Claude picked it. It is the rare solo business with real recurring revenue and almost no delivery cost after the first year. Best for. Operators with existing access to a niche. Moderate. Main risk. A handful of buyers, which means no pricing power.
5. Automation retainer for one trade
What it is. Being the only person who knows the full stack of, say, dental labs or veterinary clinics — on a monthly retainer. Why Claude picked it. Referral-driven, recurring, and defensible through specialisation rather than technology. Best for. Anyone who can pick one trade and stay in it. Easy to start. Main risk. You become an employee with extra steps and worse benefits.
6. Technical archaeology contractor
What it is. Reading a legacy system and writing the specification nobody has. Why Claude picked it. Generation got cheap; comprehension did not, and replacement projects stall without it. Best for. Senior engineers. Involved. Main risk. Burnout, which Claude named directly — it is unpleasant work.
7. Assessment design for certifiers
What it is. Freelance psychometric work rebuilding exams that a model can now pass. Why Claude picked it. Every invalidated assessment is a funded project, and there are a lot of them. Best for. Educators and psychometricians. Moderate. Main risk. Procurement cycles measured in years, which a solo operator cannot finance.
8. AI-resistant physical service with a booking moat
What it is. Mobile repair, in-home care coordination — work that happens in a place. Why Claude picked it. Demand rises as digital work devalues, and the moat is the calendar rather than the skill. Best for. Anyone willing to do physical work. Easy. Main risk. It is a job rather than an asset, unless the owner subcontracts.
9. Paid research letter for one industry
What it is. A subscription publication covering one industry properly. Why Claude picked it. AI collapsed the cost of reading everything, which makes judgement about what matters the scarce input. Best for. People with a genuine information edge. Moderate, and slow. Main risk. Distribution is brutal, and Claude was blunt that this is the hard part rather than the writing.
10. Compliance documentation, productised
What it is. Templates plus a review call for one regulated niche. Why Claude picked it. Fast to launch and immediately useful, which is rare together. Best for. Anyone with existing compliance experience. Easy. Main risk. Trivially copied once it works.
11. Robot commissioning subcontractor
What it is. Freelance installation and calibration for one robot manufacturer. Why Claude picked it. Certification is the barrier, and barriers are what protect solo rates. Best for. Mechatronics and field-service backgrounds. Moderate, certification-gated. Main risk. Single-customer concentration — one manufacturer can end the business.
12. Prompt and policy libraries for regulated professions
What it is. Maintained, profession-specific libraries updated as the rules change. Why Claude picked it. The maintenance is the product; a static library is worthless within a year. Best for. Practitioners inside a profession. Easy to moderate. Main risk. Model vendors ship domain packs for free.
Where Claude and ChatGPT agreed
This topic produced the strongest convergence of the whole series. Asked separately, with neither shown the other’s answer, both models arrived at six of the same businesses.
| Theme | Claude’s version | ChatGPT’s version |
|---|---|---|
| Being paid to check AI output | Narrow evaluation consultancy | Human verification service |
| Moving things between systems | Model migration contractor | Niche data cleanup and migration |
| Running one profession’s stack | Automation retainer for one trade | AI operations manager for one profession |
| Selling filtered information | Paid research letter | Tiny vertical intelligence publisher |
| Translating the rulebook | Compliance documentation, productised | Regulatory change translator |
| Certifying competence | Assessment design for certifiers | Micro-certification business |
Six overlaps out of twelve is close to agreement. Where they parted, Claude went physical and forensic — robots, recordings, legacy systems — while ChatGPT went toward ownership and access, including buying abandoned software and brokering introductions. Read ChatGPT’s list next.
What this means
Verification is the solo business of 2027, on both models’ reading. The single most repeated idea across the two lists is a person paid to confirm that machine-produced work is correct.
Claude’s filter is the useful part. Requiring that the business survive two weeks of absence rules out consulting dressed up as a company. Applied to most published solo advice, very little passes.
Specialisation beats capability. Nearly every pick is defensible because of what the owner knows about one narrow domain, not because of what tools they use. That is the opposite of the usual framing.
FAQ
Did Claude actually choose these twelve? Yes. The list, the order, the difficulty ratings and the risks came from one Claude conversation on 28 August 2026, run for this article. Benchivo edited for structure, not substance.
Which needs the least money to start? By Claude’s own ratings, the automation retainer, productised compliance documentation and the physical service business are the easiest to begin. All three are also the easiest to copy.
Is any of these genuinely passive? Vertical dataset licensing comes closest, because the delivery cost after the first year is close to nothing. Claude still rated the buyer pool as its fatal weakness.
What did ChatGPT say to the same question? It agreed on six of twelve. See ChatGPT’s 12 one-person businesses for 2027.
Conclusion
Claude’s answer is narrower and stricter than most solo-business advice, mostly because it refused to count anything that collapses when the owner takes a holiday. What survives that test is a list of businesses selling judgement, certification and maintenance — published here dated and attributed under Benchivo’s methodology, so it can be checked against what 2027 actually looks like.