Seven models people actually use to sell AI work. All seven are real. None is passive. This page is what each one requires — the companion to the arithmetic piece, which shows why the pitched versions do not add up.
- Five of the seven are services. You are selling your time with a faster tool, which is a business, not an escape from one.
- Two are trading. Different risk entirely, and the failure mode is losing money rather than losing a client.
- Client acquisition is the constraint in all five service models, and it appears in none of the pitches.
The five service models
Task farming
Find proven demand on an existing marketplace, list a service AI can help fulfil, keep the spread between what the customer pays and what delivery costs.
What it requires: a marketplace account with standing, which takes months. The demand is proven precisely because others already serve it — you are entering a priced market, not discovering one.
Where it breaks: quality-checking every order. Skip that and the ratings go, and the ratings were the asset.
Creation as a service
Enhance a client's real assets — their photos, products, locations — rather than replacing them with obviously generated media.
This is the strongest of the seven, and the reason is in the framing: it uses AI to raise the production value of something that already exists, so the output does not read as synthetic and the client's brand stays theirs.
What it requires: taste, and a client who has assets worth enhancing. Camera grammar and colour and light matter more here than prompt tricks.
Social media marketing
Audit a brand's current content, regenerate it at higher quality, turn it into repeatable posts and campaign variations.
What it requires: judgement about what to post, which is the part clients actually pay for. Generation is the cheap half. A retainer runs $500–1,500 basic, $3,000–8,000 premium.Verified 22 Aug 2026 against Autocalls, Aircall, Retell, Trillet, Kommunicate and WhiteLabelAI pricing analyses, Mar–Aug 2026
Media channels
Build a channel where the content engine, narration and visual identity scale without filming daily.
What it requires: the same things any channel requires — hooks, retention, consistency, distribution. AI removes the production constraint and none of the others. Most channels fail on distribution, and this model does not touch distribution.
It is also the slowest to pay. Ads, affiliates and sponsorship need audience first, and audience takes the same year it always did.
Automation as a service
Map a client's workflow, find the step wasting the most time or money, build something reliable for that specific process, charge for the outcome.
What it requires: access to how the business actually runs, which is a trust problem before it is a technical one. And it carries the heaviest obligation of the seven — see what you owe a client whose system you automated.
This is the model most likely to still be paying in three years, and the one most likely to wake you at 2am.
The two trading models
Task farming and automation earn a fee for work. Arbitrage and resale earn a spread on a position — you are holding inventory or capital at risk.
A failed service engagement costs you a client. A failed trade costs you money. The pitches present them side by side; they should not be.
Agentic arbitrage
An agent monitors marketplaces for price gaps and flags what clears your rules on margin, fees and risk.
What it requires: rules that survive contact with costs. The spread has to cover fees, transfer time, and the cases where the gap closes before you execute. Most apparent gaps are fee structures you have not fully counted.
And an agent monitoring markets is an agent with money-touching permissions. Tier five, without exception.
Resale research
Research resale values, track underpriced stock, buy on strict rules, use AI for listings and customer responses.
What it requires: capital, storage, and category knowledge. The AI writes the listing — it does not tell you the market for that item is thinner than the historical data suggests.
What all seven share
- Client acquisition, or its trading equivalent. None of the pitches address how you find the customer, and in five of the seven that is the binding constraint.
- An API meter running at 30–50% of margin on the service models.
- A quality check you cannot skip. Every model has a step where unreviewed output destroys the asset — ratings, brand, client trust, or capital.
- Disclosure, where clients are involved. See the handover page.
None of these is passive. Five are agencies and two are trading desks, and both of those are real jobs with known failure modes.
That is not a reason to avoid them. It is the difference between choosing a business and being sold one — and the arithmetic behind the second is on the repo analysis.