Choosing an AI development company comes down to a few checks most buyers skip: do you own the code at the end, is the price fixed or an open-ended day rate, will they tell you when AI is the wrong tool, and can you speak to a real client who's used them? Get those right and you avoid the expensive mistakes. This checklist gives you the exact questions to ask and the red flags that mean walk away.
We build AI systems for a living, so writing a guide that helps you interrogate firms like us costs us something. That's the point. A good vendor should survive hard questions, and if this makes it easier to spot the ones that can't, we'd rather you found out before you signed than after.
How do I choose an AI development company?
Pick the firm that scopes a fixed price, hands you full ownership of the code and IP, starts with an audit rather than a build, and is willing to tell you not to build at all. Those four traits filter out most of the market before you've even looked at portfolios.
The industry has a lot of resellers dressed as engineers right now. Some wrap a ChatGPT API call in a web form and sell it as "bespoke AI." Others quote a day rate, start building, and let scope drift until your budget runs out. Both leave you holding the risk.
Work through this order:
- Start with the verdict, not the demo. A firm worth hiring will happily say "you don't need us for this, buy the £30/month tool." That honesty is the strongest buying signal there is.
- Check ownership before anything else. If you don't own the code and IP at the end, you're renting your own business logic from someone else.
- Insist on a fixed price for defined work. Day rates with no ceiling shift every delivery risk onto you.
- Ask to speak to a client. Not a testimonial on a page — an actual person who paid them.
Our companion guide on agency vs freelancer vs in-house covers which type of supplier fits your situation before you start shortlisting firms.
What questions should I ask before hiring an AI agency?
Ask the seven questions below in your first proper conversation. They're designed to surface how a firm actually works, not how it markets itself. Vague or defensive answers tell you as much as clear ones.
- Who owns the code and IP when we're done? The only acceptable answer is "you do, in full." Anything conditional is a flag.
- Is this a fixed price or a day rate — and what's the ceiling? Fixed price for a defined scope means they've done the thinking upfront.
- How will you handle our data, access control, and UK GDPR? If they can't talk about row-level security, data residency, and who can see what, they haven't built anything serious.
- Can I speak to two past clients? Real references, not logos on a homepage.
- Who is accountable, by name, if this breaks after launch? You want a person and a support arrangement, not a shrug.
- When did you last tell a client not to build something? If they can't answer, they've never turned down revenue — which means they'll never turn down yours.
- What happens after go-live? Handover, documentation, and who maintains it. Silence here means you inherit an orphan.
A firm that answers these cleanly has thought about your risk. One that dodges them is optimising for the sale.
What are the red flags to walk away from?
Some warning signs are worth a second question. Others mean end the conversation. The list below is the second kind. Any one of these, on its own, is enough to keep looking.
- They never say "don't build it." A vendor whose answer to every problem is a build is selling capacity, not judgement.
- Vague day-rate scope with no fixed price. "We'll figure it out as we go" means your budget is the only limit, and you're the one who hits it.
- You don't own the code or IP at the end. You paid for it; you should own it. Full stop.
- A thin ChatGPT wrapper sold as "custom AI." Ask what actually happens when a user submits a request. If the honest answer is "we forward it to an API," you're paying bespoke prices for a form.
- No mention of security, data handling, or access control. Anyone building on your data should raise this before you do.
- Offshore black box with no accountability. Cheap day rate, no named owner, no recourse if it's wrong. The savings evaporate the first time something breaks.
- No references or case studies you can verify. Everyone claims results. Insist on checking one.
This is "AI-washing" — dressing up commodity tooling as custom engineering. It's common enough that naming it plainly is the most useful thing this guide does. Our guide on common AI mistakes UK businesses make goes deeper on how these deals go wrong.
Here's the same signal, both directions, in one view:
| Green flag — lean in | Red flag — walk away |
|---|---|
| Fixed price for a defined scope | Open-ended day rate, no ceiling |
| You own all code and IP | Ownership is conditional or retained |
| Audit-first, build-second | Straight to build, no discovery |
| Will tell you AI is the wrong tool | Every answer is "yes, we'll build it" |
| UK-based, named accountability | Offshore black box, no owner |
| Real, checkable references | Logos and testimonials only |
| Talks about security and UK GDPR unprompted | Silent on data and access control |
| Genuine custom engineering | ChatGPT wrapper sold as bespoke |
Should I hire a UK company or go offshore?
For anything touching customer data or core operations, a UK-based firm is usually worth the premium. Offshore development can be genuinely good and cheaper per day, but you carry the extra risk: time-zone lag on decisions, UK GDPR accountability sitting entirely with you, and limited recourse if the work is wrong.
The day rate is the visible cost. The hidden costs are the ones that bite:
- Accountability under UK law. If a UK firm mishandles your data, you have a clear route to hold them responsible. Across borders, that route gets long and expensive fast.
- Data residency. Where your customer data physically lives matters under UK GDPR. A UK supplier building on UK or EU infrastructure keeps that simple. The ICO has guidance on international transfers if you want the detail.
- Communication cost. A build is a hundred small decisions. An eight-hour time gap turns a five-minute clarification into a lost day, repeatedly.
Offshore isn't wrong for every job. A well-defined, low-sensitivity project with a firm that has real references can work fine. But for the systems most SMEs commission, customer-facing and data-handling, the accountability of a UK firm earns its keep.
How much should an AI project cost?
Cost follows scope, not a day rate. A Blueprint to properly scope the work is from £4,500; a full custom build starts from £18,000. What you should resist is an open-ended day rate with no ceiling, because that structure hands all the delivery risk to you.
The healthiest arrangement, and the one we use, is a fixed-price ladder that lets you spend a little to learn a lot before committing to a build:
- A free readiness assessment to check whether AI even fits your problem.
- A fixed-price Blueprint (ours is from £4,500) that ends in an honest build-or-don't-build verdict.
- A custom build only if the Blueprint says it's worth it, quoted as a fixed price, not a running meter.
Scoping is cheap and building is not. A few thousand pounds spent finding out you shouldn't build has saved businesses the build they were about to pay for on the wrong thing. Our pricing guide breaks the ladder down properly, and choosing your first AI project helps you pick something worth scoping.
Agency, freelancer or in-house?
Before you compare firms, decide whether you even want a firm. A freelancer is cheapest and fastest for a small, contained job but carries bus-factor risk — one person, one point of failure. An agency or studio costs more but brings a team, continuity, and someone accountable after launch. Hiring in-house only pays off when AI is a permanent, central part of what you do.
| Option | Best for | Main risk |
|---|---|---|
| Freelancer | Small, well-defined jobs | One person; no cover if they vanish |
| Agency / studio | Business-critical builds needing support | Higher cost; quality varies widely |
| In-house hire | AI as a core, ongoing capability | Slow, expensive, hard to hire well |
The full breakdown is in agency vs freelancer vs in-house. The short version: match the supplier to how central and how permanent the work is. Don't hire a team for a one-off script, and don't hand a load-bearing system to a lone freelancer with no backup.
How do I compare quotes fairly?
Normalise the quotes before you compare them. The headline number is nearly useless on its own, because two firms will scope the "same" project completely differently. Force every quote onto the same footing — same deliverables, same ownership terms, same post-launch arrangement, same fixed total — and the real cost differences appear.
Build a simple comparison sheet and score each firm on the same rows:
- Total price, fixed. Convert any day-rate quote into a capped total, or treat the missing ceiling as the risk it is.
- What's actually delivered. List the concrete outputs. A cheaper quote that delivers less isn't cheaper.
- Who owns the code and IP. A quote that retains ownership is more expensive than it looks — you'll pay again to leave.
- Post-launch support. Handover, documentation, maintenance. A build with no support plan is a future bill.
- Data and security handling. Whether UK GDPR, access control, and data residency are addressed at all.
A day-rate quote with no ceiling, no IP transfer, and no support can easily cost more over a year than a fixed price that includes all three. Compare the outcome you'll actually own, not the number on the first page. If you're still unsure whether you're ready to commission any of this, our AI readiness guide is the right place to start.
Where to start without committing
If you want a de-risked first step, scope before you build. Our AI Readiness Assessment is free and tells you whether AI fits your problem at all. If it does, a Blueprint turns a vague idea into a scoped plan with an honest build-or-don't-build verdict, for a few thousand pounds, before you commit to tens of thousands. You're under no obligation to build with us or anyone.