Direct answer
Ask any AI consultancy six things before you sign: who builds it, who maintains it, where your data goes, who owns the code and accounts, what the fixed price includes, and how success will be checked. Good answers are specific and in writing. "Very few UK AI consultancies publish prices," one UK consultancy, We Call Shotgun, wrote in August 2026, so a published price list is itself a useful filter.
The six questions expand into twelve below, each with what a good answer sounds like. After that come the red flags, published prices from firms that show them, and how freelancers, agencies, consultancies and an in-house hire differ on who builds and who maintains. We're an AI consultancy ourselves, so we list our own prices on the same terms as everyone else's.
Twelve questions and what a good answer sounds like
Swipe or scroll to compare the columns.
| # | Question | A good answer | A weak answer |
|---|---|---|---|
| 1 | Which job are you taking off my desk first? | Names one process, end to end: what starts it, which tools it touches, where it waits for you | "We'll look at where AI can add value across the business" |
| 2 | Who builds it? | Names the people or roles who write the code and configure the tools, and says whether any of it is subcontracted | "Our partner network" with no names |
| 3 | Who maintains it after go-live, and at what cost? | A named support arrangement with a monthly price and hours, or a clear handover to your team | Support "as needed" with no rate |
| 4 | Where does my data go? | Which systems and AI providers see it, where they are, what is stored, and how long for | "It's all secure" |
| 5 | Who owns the code, prompts, configurations and accounts? | You do, in writing, and accounts are opened in the business's name | Accounts in the supplier's name; code licensed, not handed over |
| 6 | What exactly does the fixed price include, and what doesn't it? | A written scope listing workflows, systems, users, testing, documentation and a defect period, plus what is excluded and whether VAT is added | A day rate with an estimate |
| 7 | How will we know it's working? | A before-and-after you can count yourself, agreed before the work starts | Satisfaction surveys or "efficiency gains" |
| 8 | What does a person still sign off? | Says which steps follow fixed rules, which use AI judgement and where a person approves | "The AI handles it end to end" |
| 9 | What happens when it gets something wrong? | Describes the exception path: who is told, what stops, how it's corrected, and the audit trail | "It's very accurate" |
| 10 | Can I see something you built that runs today? | Shows a real system or a documented case, and is clear about what is a blueprint | Only slides, or results figures with no client or method behind them |
| 11 | Would you tell me if I don't need you? | Can name cases where a standard tool, a native integration or the software's own feature is the better buy | Every problem needs a custom build |
| 12 | What happens if we stop working together? | You keep everything, with documentation good enough for another team to take over | Notice periods and handover fees that aren't in the proposal |
You won't get perfect answers to all twelve from anyone. What matters is whether the answers are specific, whether they're in the proposal, and whether they still hold when you ask a follow-up.
Red flags
These are the reasons we turn work down, pointed the other way. If a consultancy shows several of them, keep looking.
- It wants to start with a business-wide AI strategy before it has looked at any one process.
- It can't say which steps a person will still approve.
- It talks about "AI employees" or agents that run without limits.
- The price is a day rate with no scope, or a fixed price with no list of what's included.
- Accounts, code or data end up in the supplier's name.
- It quotes results such as hours saved or revenue gained without naming the client, the method or the period.
- It recommends a custom build where your accounts or CRM software already has the feature.
- It can't explain what happens when the AI is wrong.
- It promises you'll be "compliant". A builder can build record-keeping; your solicitor, accountant or trade body confirms what the law requires.
One more that's easy to miss: no one at the firm has built anything. Advice is useful, but if the people advising you won't build or maintain what they recommend, you'll be paying a second firm to find out whether the advice works.
Published UK prices compared
We read these prices on each firm's own website on 24 October 2026. They're "from" prices or stated ranges, not quotes, and each firm defines its scope differently, so compare what's included as well as the number. We've included only firms whose own pages publish prices.
Swipe or scroll to compare the columns.
| Firm | Entry assessment | First build or pilot | Larger build | Ongoing | VAT |
|---|---|---|---|---|---|
| AutomateFlow | Audit £3,500 (one workflow, up to five systems) | Pilot £15,000 (one workflow in production) | Build £35,000 (one operational app) | Run £2,000 a month | Excluded |
| The AI Consultancy | Readiness Sprint from £3,500 (two weeks, no code) | Discovery and Pilot from £15,000 (four to eight weeks) | Build and Embed from £40,000 | Day rate £950–£1,500 for open-ended work | Excluded |
| Generativ | Discovery £3,500 (two weeks) | Single-process integration £3,000–£5,000 | Departmental system £10,000–£15,000+ | Maintenance £200–£500 a month | Excluded |
| OpenKit | AI Audit £10,000 | AI Implementation from £5,000 a month, worked against an agreed backlog | Bespoke builds quoted separately | Not published | Excluded |
Sources: The AI Consultancy pricing, Generativ (indicative working prices, 28 September 2026), OpenKit pricing, AutomateFlow engagement models.
Read it as a map of entry points, not a ranking. A £3,500 assessment that covers one workflow and a £10,000 audit that covers a department are different products. Before comparing numbers, check how many workflows, systems and users each step covers, whether model usage and hosting are billed separately, and whether support is included or priced on top. Prices change, so check the page before relying on a figure. Our AI implementation cost guide has a wider table and a worksheet for comparing proposals.
Freelancer, agency, consultancy or in-house hire: who builds, who maintains
The kind of supplier matters as much as the name. Each type has a place.
Swipe or scroll to compare the columns.
| Type | Who builds | Who maintains | Good for | Watch for |
|---|---|---|---|---|
| Freelancer | One person | The same person, if available | A well-defined automation on standard tools, at a lower day rate | One person is a single point of failure; ask what happens if they're ill or busy |
| Marketing or digital agency | A delivery team, often with AI added to web or marketing work | The agency, on a retainer | Website chat, content and marketing automation | Depth in your accounts, job or records systems |
| Strategy consultancy | Often a partner firm, after the strategy | Often the partner firm | Board-level strategy, larger organisations, change programmes | Paying for a plan and a separate build; ask who writes the code |
| Implementation consultancy or software studio | The firm's own engineers | The firm, or your team after handover | Integrations, automations, AI agents and bespoke software inside your existing tools | Scope creep on custom work; insist on a fixed scope per step |
| In-house hire | Your employee | Your employee | Ongoing, varied work once you know what you need | Salary, recruitment time, and the knowledge leaving with them |
Two questions settle most of this. Who will write the code, and who will you call when it breaks in eight months? If the answers are different firms, make sure both are in the contract.
Where we fit: we're an implementation firm. We build integrations, automations, bounded AI agents and bespoke software inside the systems a firm already uses, at a fixed price per step, and the business owns what we build. We work remotely, from Romania, in English and Romanian, and we don't have a UK office. We don't sell AI receptionist subscriptions or give legal, tax or compliance advice. And where a standard tool does the job, we'll say so.
Where to check a firm off-site
A firm's own website shows what it wants you to see. Check a few things elsewhere.
Companies House. Search the company name on the Companies House register. Check how long it has existed, who the directors are, and whether its accounts and confirmation statements are filed on time. For a firm based outside the UK, ask which entity you'll contract with and where it's registered.
Review directories. Clutch and similar directories collect client reviews of software and AI firms. A few detailed reviews from named clients are worth more than many short ones. A new firm may have none, which isn't a red flag on its own.
References. Ask to speak to one client whose project resembles yours, ideally one that has been running for six months or more. Ask what broke, how fast it was fixed and whether the price held.
A running system. Ask to see something that runs in production, not a demo built for the pitch. Our case studies, such as the PaxBespoke operations automation, describe what was built and label blueprints as blueprints.
LinkedIn. Check that the people named on the website exist and have built things before. It takes ten minutes and tells you who you'll actually work with.
Then go back to the twelve questions and ask for the answers in the proposal itself.
