To start an AI services business UK firms will actually pay for, sell outcomes they already understand: hours saved on admin, faster replies to customers, cleaner data, and staff who can finally use the software they are licensed for. The work that pays reliably is unglamorous. It is process automation, pulling data out of documents, sorting inbound queries, producing marketing copy at volume, making internal knowledge searchable, and training people. Skip the impressive-sounding “AI strategy” deck and lead with a problem a business owner can name.
You do not need to build a large language model or raise funding. Most successful one-person and small AI services firms in the UK resell and configure existing tools, wrap them in a clear process, and take responsibility for the result. Your value is judgement: knowing what to automate, what to leave alone, how to handle the client’s data safely, and how to price so that the third party costs never sink you. This guide covers the services, the pricing, the legal obligations, and how to win the first few clients when you have no portfolio.
The services small firms actually buy
Pick a short menu. Owners buy things they can picture, not capabilities. These six lines cover most of the paying demand right now.
Process automation
Connecting the tools a business already uses so a task runs without a human copying and pasting. Think new enquiry to CRM to first reply, or order to invoice to reminder. Much of this is achievable with no-code platforms, and our guide to building a workflow automation without coding shows the shape of the work. The buyer is usually the owner or an operations manager drowning in repetitive steps.
Document and data extraction
Turning PDFs, scanned forms, delivery notes and emails into structured data a system can use. Accountants, logistics firms and property managers pay for this because manual keying is slow and error-prone. The related work of automating invoices and payments often starts here.
Customer support triage
Sorting inbound messages, drafting replies for a human to approve, and answering common questions from an approved knowledge base. Done well it cuts response times without letting a bot loose on angry customers. Our guide to AI chatbots for customer service sets sensible expectations.
Content and marketing production
Producing product descriptions, blog drafts, ad variants and social posts at volume, then editing to a house style. The buyer is a marketer or an owner with no time. Pair it with a system like marketing on four hours a week so the output has somewhere to go.
Internal knowledge search
Making a firm’s own documents, policies and past quotes answerable in plain English so staff stop asking the same colleague the same question. This is high value and high risk, because it touches sensitive internal data.
Training staff on tools they already pay for
Many firms hold Microsoft 365 or Google Workspace licences with AI features they never use. A half-day session that gets a team using them safely is often the easiest sale you will make. Point clients to five AI workflows a ten-person firm can run this quarter as a starting map.
Positioning: pick a sector or a process
Generalists struggle to get referrals because “I do AI” gives no one a sentence to repeat. When a happy client is asked “who do you know”, they can only pass your name on if they can describe what you do in one line. “She automates quoting for electrical contractors” travels. “AI consultant” does not.
Choose one axis. Position by sector (dental practices, letting agents, wholesalers) or by process (invoice processing, inbound enquiries, reporting). Sector positioning wins referrals within a trade and lets you reuse solutions. Process positioning lets you sell the same fix across industries. Either beats being a generalist. You can widen later, once you have proof and cash flow.
Pricing models, honestly
There is no single right model. The risk sits in different places depending on which you choose. Here is how the common options behave in practice.
| Model | Typical UK range (approx, subject to change) | Main risk | Best for |
|---|---|---|---|
| Day rate | £400 to £900 a day | Income capped by hours; client watches the clock | Discovery, training, uncertain scope |
| Fixed scope project | £1,500 to £15,000+ | Scope creep and messy data eat your margin | Well-defined builds with clear boundaries |
| Monthly retainer | £500 to £3,000 a month | Proving ongoing value so it is not cancelled | Maintenance, iteration, support after launch |
| Outcome based | Share of saving or per-result fee | Attribution disputes; you carry delivery risk | Measurable, isolated outcomes you control |
Ranges are approximate and vary with region, complexity and your track record. Newer providers sit at the lower end.
Pass through or cap the usage costs
The tools you build on charge by usage, often per unit of text processed. That cost moves with how much the client actually uses the system, and it is not yours to absorb. Either pass it through at cost with a small handling margin, or set a monthly cap and pause the service when it is hit. If you fold unlimited usage into a fixed fee, a client who suddenly doubles their volume turns your profit into a loss overnight. Write the usage terms into the contract in plain figures.
A sensible default
For most first engagements, quote a fixed price for a scoped project, run a short paid discovery before you commit to that number, and offer a retainer for support afterwards. This gives the client a predictable figure while protecting you from quoting blind. To show the value, agree the numbers you will measure up front, using the approach in how to know automation has paid off.
The awkward commercial realities
Three problems will show up on almost every deal. Plan for them and you look professional. Ignore them and you work for free.
The free pilot expectation. Clients often ask you to “prove it first”. A demonstration on your own examples is reasonable. Building a working prototype on their data is real work and should be paid, even if only at cost. Frame it as a paid discovery or pilot with a fixed, modest fee and a clear deliverable.
Messy data. Projects stall because the client’s spreadsheets are inconsistent, their documents are unlabelled, or the same customer appears five ways. This is the single biggest cause of overrun. Inspect a real sample before you quote, and make data quality the client’s responsibility in writing, with a day rate for any cleanup you end up doing.
Discovery before quoting. Never quote a fixed price from a single conversation. A paid half-day or day of discovery, where you see the actual files, tools and volumes, lets you scope accurately. Clients respect it because it protects them from a project that balloons.
Your professional and legal obligations
Handling other firms’ data raises real duties. Get these right early; they also become a selling point.
UK GDPR and third party model providers
If you process personal data on a client’s behalf you are a processor under UK GDPR, and you need a written contract that says so. Be explicit about what leaves the client’s systems and goes to a third party model provider, and check whether that data is used to train the provider’s models. Business tiers of the major tools generally do not train on your inputs, but you must confirm the tier and settings and put it in the privacy notice. The ICO guidance on AI and data protection is the primary source, and our guide to using ChatGPT and Copilot without breaking UK GDPR translates it into steps.
Insurance and company setup
Carry professional indemnity insurance before you take paid work; it covers claims that your advice or build caused a loss. If you work from home, check the tax and insurance points in our guide to running a business from home in the UK. Whether you trade as a sole trader or a limited company, register correctly, and if you incorporate at Companies House you will also need to complete the new director identity verification. Choosing the right activity code matters too, as our guide to SIC codes explains.
Contracts that limit liability for model output
AI output can be wrong, and you cannot guarantee it. Your contract must say the client remains responsible for reviewing output before relying on it, cap your total liability (commonly to the fees paid), and exclude liability for the model provider’s errors and downtime. State clearly who owns what, including the configurations and prompts you write. A solicitor-drafted template pays for itself the first time a client disputes an outcome. The privacy lessons in our piece on the Hims and Hers lawsuit apply to your promises too.
Getting the first clients without a portfolio
No case studies is a chicken-and-egg problem. Break it deliberately.
Do one job at cost for proof
Find one business with a clear, contained problem and do the work at cost in exchange for a named case study and a testimonial. Agree the deliverable and the numbers you will report before you start. One credible before-and-after result opens more doors than any amount of marketing.
Partner with people who hold the relationships
Accountants, bookkeepers, marketing agencies and IT support firms already have the trust and the client list. Offer to deliver the AI work behind their brand or on referral for a fee share. This is the fastest route to a pipeline, because the hard part, the relationship, is already done. Firms exploring the best AI accounting tools often need someone to implement them.
Demonstrate on the client’s own data
Nothing sells like showing a prospect their own worst spreadsheet answered in seconds, or their own inbox triaged. With a signed short agreement covering confidentiality, a live demonstration on real data converts far better than a generic pitch. Keep the sample small and delete it afterwards unless they engage you.
Be findable locally
Most early clients come from referral and local search, so set up a Google Business Profile and follow our local SEO guide. A single sharp page saying who you help and how beats a broad brochure site.
Mistakes people actually make
- Selling “AI” instead of an outcome. Owners buy fewer hours on invoices, not “intelligent automation”. Name the task and the saving.
- Quoting before seeing the data. The demo data is always tidy. The real data is not. Always run paid discovery first.
- Absorbing usage costs. A generous fixed fee becomes a loss the month a client’s volume spikes. Pass through or cap.
- Automating a broken process. Speeding up a bad workflow just produces bad results faster. Fix the process, then automate it.
- No liability cap or review clause. One confidently wrong output the client acted on, and no contract, is a nightmare. Get the contract right first.
- Ignoring security. You will handle sensitive client data. Consider Cyber Essentials, and be aware of emerging threats like the Copilot AI worm hidden in documents.
Frequently asked questions
Do I need to be technical?
You need enough skill to configure tools reliably, test edge cases, and explain limits honestly, but you do not need to write production software. Many strong providers come from operations, marketing or finance and win because they understand the business problem. If you cannot judge whether an output is safe to rely on, though, you are not ready to charge for it.
Should I build or resell?
Resell and configure existing tools for almost everything. Building your own model is expensive, slow and rarely what a small firm needs. Your margin comes from judgement, process and accountability, not from owning the underlying technology. Reserve custom development for cases where no existing tool fits, and price it accordingly.
What should I charge for my first project?
Quote a fixed price for a scoped piece of work, typically in the low thousands for a contained automation, after a short paid discovery. If you take a first client at cost for a case study, still put a real number on the invoice and mark the discount, so the value is visible. These figures are approximate and depend on complexity and your experience.
What if the client wants to own the prompts?
That is a reasonable ask, and often fair, because the prompts encode their process. Decide your position before quoting: you might hand over prompts and configurations on final payment while keeping your general methods and templates. Whatever you agree, write the ownership terms into the contract so there is no dispute later.
Do I need to register for VAT or a limited company?
You can start as a sole trader and register with HMRC; a limited company adds protection and credibility but more admin. VAT registration is only mandatory once your turnover passes the threshold. Our guide to what counts as a small business in the UK explains the bands, and if you take on help, follow the checklist for hiring your first employee.
What to do next
- Choose one axis and write your sentence. Decide the sector or process you serve and draft the one line a client would repeat about you. Everything else follows from it.
- Line up your legals. Buy professional indemnity insurance, get a contract template with a liability cap and review clause, and prepare a processor agreement and privacy wording using the ICO guidance.
- Land one proof project. Approach an accountant or agency partner, or a business with a contained problem, and deliver one job at cost with agreed before-and-after numbers.
- Price to protect yourself. Run paid discovery before every quote, pass through or cap usage costs, and keep a retainer offer ready for the work that follows.
