There is a great deal of noise around making money with AI, most of it selling a course. This hub takes the opposite approach: for each model, what you actually need, what it costs to start, who you are competing with, and how it fails.
What you will not find here
No income figures, no earnings guarantees, no “passive income” framing. Anyone promising you a number does not know your market, your skills or your capital. What follows is a description of business models, not a promise about outcomes.
The models, honestly assessed
AI consulting
What you sell: judgement. Which tools a business should adopt, how to roll them out, what policy to put around them.
Skills needed: real domain experience in the client industry, plus enough technical literacy to be credible. This is the differentiator, not the AI knowledge.
Startup cost: low — tool subscriptions and your time.
Competition: heavy at the generic end, thin in specific industries.
Main risk: being one of thousands of undifferentiated “AI consultants”. Without a sector you know deeply, you have nothing to sell.
AI automation services
What you sell: working automations, built and maintained.
Skills needed: a platform such as Make, n8n or Zapier, plus the process-mapping skill to work out what should be automated.
Startup cost: low, though hosting matters if you self-host for clients.
Competition: growing fast.
Main risk: maintenance. Automations break when APIs change. Price a retainer or you will be doing free support forever.
AI freelancing
What you sell: a specific deliverable produced faster because you use AI well.
Skills needed: the underlying craft. AI-assisted editing still requires an editor.
Startup cost: minimal.
Competition: intense, and price pressure is real as clients learn what the tools cost.
Main risk: selling “AI” rather than an outcome. Clients buy the finished thing, not the method.
AI agencies
What you sell: delivery capacity across a team.
Skills needed: everything above, plus hiring, quality control and cash-flow management.
Startup cost: materially higher — people are a fixed cost against variable revenue.
Main risk: the classic agency failure, which has nothing to do with AI: winning work you cannot deliver profitably.
AI digital products
What you sell: templates, prompt packs, workflows, courses.
Skills needed: genuine expertise plus an audience. Without the audience, distribution is the whole problem.
Startup cost: low to build, high to market.
Main risk: the market is saturated with low-effort products, so the bar for anything that sells twice is higher than it looks.
AI SaaS
What you sell: software, on subscription.
Skills needed: product, engineering and distribution — or the capital to buy them.
Startup cost: highest here, and the runway is measured in years.
Main risk: building a thin wrapper around a model API that the model provider absorbs in its next release. Durable products own the data, the workflow or the distribution — not the model.
How to choose between them
Work backwards from what you already have. Deep industry experience points to consulting. A technical bent and patience points to automation services. An existing audience points to products. Capital and a co-founder points to SaaS. Starting from “which is most profitable” rather than “which am I equipped to do” is the most common mistake.
Frequently asked questions
How much can I earn with an AI business?
We will not answer that, because nobody can honestly answer it for you. Earnings depend on your market, your skill, your pricing and your ability to win work. Anyone quoting a figure is selling something.
Do I need to be technical?
For automation services and SaaS, effectively yes. For consulting, content and productised services, domain expertise matters more than technical depth.
Is it too late to start?
The generic end is crowded. Specific applications in specific industries are not — which is why sector knowledge is worth more than AI knowledge right now.