Most businesses frame this as AI consultant vs AI agency, as if they were choosing between two kinds of shop. They're really choosing who thinks, who builds, and who owns the result once the invoices stop. Get those three wrong and the label on the supplier won't save you.
The short answer: hire an AI consultant when you don't yet know which problems AI should solve, or you need someone senior to design, scope and govern the work. Hire an AI agency when the problem is clear and you need things built and maintained. Build in-house when AI is core to what you sell and you can recruit and keep the skills. Many businesses end up using two of the three.
Key points:
- The labels are loose. Plenty of "agencies" are one person, and plenty of consultants build.
- Most UK businesses using AI bought ready-made tools rather than building or fully outsourcing, according to the government's AI Adoption Research.
- Hiring AI talent is hard: 35% of organisations in a DSIT-commissioned survey had AI vacancies they found hard to fill.
- Whatever you choose, decide up front who runs the system after go-live.
AI consultant vs AI agency: what's the actual difference?
In theory it's simple. An AI consultant diagnoses and advises: which processes are worth automating, what data you have, what the risks are, what a sensible first project looks like and how you'll know it worked. An AI agency delivers: it builds the chatbot, the automation, the agent or the integration, and often charges a monthly fee to keep it running.
In practice the line is blurry. A good consultant will often build the first version themselves, because a working prototype answers questions faster than a slide deck. A good agency will push back on a bad brief, which is consulting in all but name. If you want a fuller picture of the advisory side, I've written up what an AI consultant actually does day to day.
The more useful distinction is incentive. A consultant is usually paid for judgement and time, so the risk is that they over-analyse. An agency is usually paid for things built and kept running, so the risk is that everything looks like a build. Neither is sinister. You just need to know which way the current pulls.
My take: ask any supplier what they would tell you not to build. A consultant who can't answer is selling hours. An agency that can't answer is selling builds.
What does an AI automation agency actually sell?
"AI automation agency" has become a business model in its own right, with courses teaching people to start one. That tells you something about the range of quality. At one end you'll find experienced engineering teams who build properly tested systems. At the other, someone who finished a course last quarter and is wiring together no-code tools for the first time on your customer data.
The typical offer looks like this:
- A discovery call, often free, that turns quickly into a proposal.
- A build: lead handling, document processing, an inbox assistant, a voice agent, a reporting workflow.
- A monthly retainer for hosting, monitoring and changes.
That retainer is where you need to pay attention. If the automations live in the agency's accounts, use their API keys and are documented only in their heads, you don't own a system. You rent one. The cheapest build in the world is expensive if you can't leave.
None of this means agencies are a bad choice. For a well-defined job, such as routing inbound enquiries into a CRM, a capable agency is usually the fastest route to something working. Just insist that accounts, credentials and documentation sit with you from day one.
When does building AI in-house make sense?
When AI is part of what you sell, or when the workflows are so specific and so central that you'll be changing them every month. In those cases, the knowledge needs to live inside the business.
For most small and mid-sized firms, though, the honest answer is "later". Look at how UK businesses actually adopt AI. In the government's AI Adoption Research, based on 3,500 interviews in 2025, most AI users had bought ready-to-use external solutions. For text generation and language tools, 71% bought, 14% developed in-house and 9% outsourced development. Even for machine learning, where in-house work was most common, it was only 24%.
Hiring is the other constraint. In the AI Labour Market Survey 2025, 35% of the 119 organisations surveyed had AI vacancies they found hard to fill, mostly because candidates lacked work experience (31%) or technical skills (30%). Senior roles were the hardest. And that's a sample made up largely of AI-active firms, who know what they're looking for.
There's also a structural risk in the first hire. One AI engineer in a 40-person business becomes a single point of failure. If they leave, the systems leave with them in every way that matters. Budget for the full cost too: in the UK, employer National Insurance is 15% on earnings above £5,000 for 2026 to 2027, before pension, tools, recruitment fees and management time.
How do the three options compare?
This is a generalisation, and individual suppliers will break it in both directions. It's still a useful starting frame.
| AI consultant | AI agency | In-house team | |
|---|---|---|---|
| Best for | Deciding what to do, scoping, governance, vendor choice | Building and running clearly defined automations | AI that is core to your product or operations |
| What you get | Diagnosis, roadmap, business cases, sometimes a prototype | Working systems, usually with a support retainer | Capability that stays in the business |
| Main risk | Recommendations nobody implements | Lock-in and builds you didn't need | Slow hiring and key-person dependency |
| How it's usually priced | Hourly, day rate, fixed-fee projects or retainers | Project fee plus monthly retainer | Salaries, on-costs, tools and infrastructure |
| Who owns the knowledge | You, if handover is part of the scope | Often the agency, unless contracted otherwise | You, for as long as the people stay |
The hybrid is common and often sensible: a consultant to pick the problems and write the brief, an agency or off-the-shelf tool to build, and a named internal owner to run it. If you're weighing the build question specifically, see build vs buy AI.
Which option costs less?
Over a single project, a consultant usually costs less than an agency because there's less to build. Over three years, an agency retainer can quietly outgrow both. In-house is the most expensive to start and can be the cheapest per unit of work once you have enough of it. That's why "which is cheaper?" has no clean answer.
For context on scale, the same government survey asked businesses what they spent on AI in 2024. Among those who could give a figure, the median was £2,000 and the mean was £19,000. A big chunk, 35%, didn't know at all. Most UK businesses are still at the subscriptions-and-experiments stage, so be wary of anyone proposing a six-figure programme as your first step.
For current market rates, I've pulled them together in how much an AI consultant costs, and for full project budgets, what AI implementation costs. The pattern that matters more than the day rate: the expensive option is the one you pay for twice. That usually means a strategy nobody builds, followed by a build nobody scoped.
What I've seen
I've spent over 20 years on the buying and building side of technology, including as co-founder and CTO/COO of Telenor Health. These days I run my own working lab of AI agents, Ascendancy IQ, which gives me a fairly unsentimental view of what's easy and what isn't.
The common mistake in the AI consultant vs AI agency decision is hiring for the wrong stage. Businesses that haven't picked a problem hire an agency, and the agency builds whatever was mentioned first, which is usually a chatbot. Businesses that have a clear, boring, high-volume process hire a strategist, and get a roadmap for something they could have bought off the shelf in a week.
The second risk is no owner. The system goes live, the supplier moves on, and six months later the automation is failing silently because a form changed. Nobody inside was told it was theirs.
The research backs up the value of outside help when it's done properly. MIT's NANDA team found that external partnerships reached deployment about 67% of the time, against about 33% for internal builds. Their sample was mostly larger organisations and the figures are self-reported, as the authors themselves note. What I take from it: teams who have shipped this kind of system before tend to finish. Building your own is fine once you've got that experience in the room.
How to decide which route is right for you
- Write down the problem in one sentence. If you can't, you need a consultant or a short discovery exercise first. My AI readiness checklist is a decent place to start.
- Check whether a product already does it. The UK data says most businesses buy ready-made tools, and for good reason.
- If it needs building, define "done" before you brief anyone. Volumes, accuracy, who checks the output, what happens when it fails.
- Name the internal owner. One person, with time in their week, who will run it after go-live.
- Choose the supplier type that fits the gap. Thinking gap: consultant. Building gap: agency. Ongoing core capability: start hiring, and use outside help until you've got the team.
- Contract for ownership and exit. Accounts in your name, documentation delivered, handover included, and a notice period you can live with.
- Start small and paid. A proof of concept with a clear success measure tells you more about a supplier than any pitch.
When you're ready to shortlist, use the questions to ask an AI consultancy before you sign anything.
Make the handover a deliverable
Ask the supplier to demonstrate one ordinary task and one failure recovery with your internal owner at the controls. Can that person pause the workflow, inspect the record, change a business rule and restart it? If the answer needs another paid engagement, the handover is unfinished. Record which maintenance belongs to your team and which remains under the support contract.
Frequently asked questions
Is an AI consultant cheaper than an AI agency? Neither label is inherently cheaper. A consultant advising on a decision and an agency delivering software are selling different scopes. Over several years, agency retainers can add up to more than either a consultant or a part-time internal role. Compare the total cost over two to three years, not the first invoice.
Can an AI consultant also build the solution? Many can, especially for prototypes and smaller automations. Ask who does the building, whether it's subcontracted, and who will support it after launch. A consultant who builds everything personally can become the same key-person risk as a single internal hire.
What is an AI automation agency? It's a supplier that builds automated workflows using AI models and automation tools, typically for lead handling, document processing, customer service or reporting. Most charge a project fee plus a monthly retainer. Quality varies widely, so check what they've built that is still running.
Should a small business hire an AI engineer? Usually not as a first step. Most UK businesses using AI buy ready-made tools, and experienced AI staff are hard to recruit. Bring in outside help for the first projects and hire once you have enough ongoing work to keep someone busy.
Do IR35 rules apply when hiring an independent AI consultant in the UK? They can, if the consultant works through their own limited company. Under the off-payroll rules, medium and large private sector clients and public sector bodies are usually responsible for deciding the worker's employment status; for small clients, the consultant's company decides.
Sources
- AI Adoption Research, Department for Science, Innovation and Technology (IFF Research and Technopolis), fieldwork February to May 2025
- AI Labour Market Survey 2025, Gardiner & Theobald for DSIT
- The GenAI Divide: State of AI in Business 2025, MIT NANDA (Challapally, Pease, Raskar, Chari), July 2025
- Understanding off-payroll working (IR35), HMRC
- Rates and thresholds for employers 2026 to 2027, HMRC
If you're stuck between a consultant, an agency and a hire, and want a straight answer on which gap you actually have, let's talk.
