Hiring an AI consultancy in Liverpool: a buyer's checklist

Choosing an AI consultancy is harder than it should be. The market is full of firms with strong sales decks and weak production track records, and the gap is hard to see from the outside until twelve months and a six-figure invoice into the engagement.

We are biased — we are a Liverpool-based AI consultancy and we want you to consider us. But the test below is one we would happily be measured against. If you are weighing up firms in Liverpool, the North West, or further afield, run this checklist in your next call. The answers tell you most of what you need to know.

The twelve questions

1. "Show me three systems you put into production this year."

Not pilots. Not demos. Live systems your engineers are still supporting. The right answer is specific — what it does, the metric it moves, what shipped in week one versus what shipped in week six. If you only hear about strategy decks, workshops or "transformations," you are talking to a slide-deck firm.

2. "Who would actually do the work?"

The senior people on the sales call are not always the people on your project. Ask to meet the engineer and the strategist who will be in your Slack on day one. If there is an account manager between you and the people doing the work, you are paying an account-management tax.

3. "What is the metric we are moving, and how will we know?"

Any good AI engagement is tied to a single named metric — claims processed per hour, tickets deflected, analyst hours reclaimed, accuracy of extraction at a defined threshold. If the proposed engagement does not start with a metric, it will not finish with one.

4. "What does week six look like?"

The median time from kickoff to a live system, in our experience, is six weeks. Some shops will tell you twelve months. That is sometimes the right answer for very regulated work, but it is much more often a sign that the firm does not know how to ship in short loops.

5. "What happens when the model gets it wrong?"

Every production AI system gets things wrong. The question is what the system does in that case. Does it cite a source? Does it refuse to answer? Does it escalate to a human? If the answer is "it doesn't, really, we're confident in the model," walk away. Confident hallucination is the most common project killer we get called in to rescue.

6. "Show me the eval set."

Mature AI teams run their systems against a curated evaluation set every time something changes. If your prospective consultancy cannot show you, for a previous project, the eval set, the metrics it tracks, and the regression dashboard, they are not running production AI work. They are running prototypes that survive because no one has looked closely.

7. "Who owns the code?"

You should. The right answer is "you do, in your repo, with full documentation, and we trained your team to maintain it." The wrong answer involves a vendor portal, a managed service tier, or a SaaS layer you cannot exit cleanly. AI lock-in is the new vendor lock-in.

8. "What does your monitoring look like in production?"

Token cost, latency, error rate, hallucination rate, retrieval recall at k. If the firm cannot rattle off the dashboards they would set up on day one, they have not run a system in production at any scale. Observability is the unglamorous half of AI engineering and the half that decides whether the system is still working in six months.

9. "How do you handle data we cannot leave our network?"

For regulated work — healthcare, legal, finance — the answer matters. Look for specific experience with on-premises deployment, private model hosting, or air-gapped retrieval. "We can figure it out" means they have not.

10. "What would you tell us not to build?"

A good consultancy will, in a discovery call, talk you out of at least one thing. There are always two or three ideas on the original brief that are not yet ready, not worth the cost, or better solved by a deterministic workflow. If everything you suggest is met with "yes we can do that," you are talking to a vendor, not a consultancy.

11. "What is a recent project that did not work?"

Everyone has them. The good firms can tell you exactly what went wrong, what they tried, why it failed, and what they changed for the next engagement. The firms that cannot are either inexperienced or dishonest.

12. "How do we end this engagement well?"

Ask up front. The answer should include a handover, documentation, training, a final review, and a defined point at which you are not paying them any more. If the implicit answer is "you would not really want to end it," that is the answer.

Red flags

The list above is the positive case. A shorter, harder list — these are the patterns we see in failed engagements:

• All the answers begin with "the platform we use." Means they are reselling someone else's stack and probably cannot debug it.
• The price is per-month with no defined endpoint. Means there is no incentive to make you self-sufficient.
• The case studies are about the consultancy, not the client outcome. Means they are selling themselves, not solving your problem.
• They cannot tell you the cost in tokens per request. Means they have never had to budget for it in production.
• The technical lead is from the parent agency, not the AI practice. Means the AI work is a bolt-on, not the centre of the firm.

How to scope the first engagement

If you are convinced you have found a firm worth working with, the first engagement should still be small. Two patterns work well:

• A two-week scoping engagement. They audit your operations and data, you agree a ranked list of candidate AI projects, and you decide together which one to prototype next. Costs in the low five figures, gives both sides a way out cheaply.
• A three-week proof of concept. Targeted at one named workflow, against real data, with a defined evaluation set. Either it moves the metric and you commission a production build, or it does not and you have spent a small fraction of what a full build would have cost.

What we would not do as a first engagement: an open-ended discovery, a long strategy phase with no working prototype, or a fixed-bid production build before anyone has seen the system work on your data.

Local specifics for Liverpool

A handful of things to know if you are hiring in the city specifically:

• The talent pool is real but thin. University of Liverpool turns out strong AI graduates, but the senior, production-grade engineers are scarce. Firms with named senior engineers on the website are worth more than firms with "team capacity" claims.
• LCRCA funding routes can de-risk early work. If you are an LCR-based business, the Liverpool City Region Combined Authority's AI and data programme has occasional funded routes for innovation pilots. Worth asking about even if you do not end up using it.
• A Liverpool address is not the same as Liverpool delivery. Plenty of London consultancies have opened a Liverpool office for the marketing optics. Ask where the engineer who would do the work actually lives, and how often they would be on-site if you needed them to be.
• The good firms are usually small. The strongest AI work in the city is being done by teams of three to fifteen, not by the consulting arms of larger agencies. Size correlates with overhead, not with delivery quality.

How we would handle a first call

If you take a 30-minute call with us, here is what to expect — and what to expect from any good Liverpool AI consultancy:

• Five minutes on your context, twenty on the problem, five on what a sensible next step looks like.
• We tell you, honestly, whether AI is the right tool for the problem in front of you.
• If it is, we propose the smallest possible first engagement that would prove it.
• If it is not, we say so, and tell you what would be the right tool instead.
• You leave the call with a one-page summary by the end of the next working day.

If you would like to test that against the checklist above, book a discovery call. If we fail the test, you have lost half an hour. If we pass it, you have a candidate.

LiverpoolAI is a Liverpool-based AI consultancy. We design, build and ship the AI agents, automations and infrastructure that put real intelligence to work inside North West businesses.