AI projects we ship most often for Liverpool businesses (and roughly what they cost)

Most "how much does AI cost" articles are useless because they refuse to give numbers. We are going to give numbers. The ranges below are what we have actually charged for these projects over the last twelve months — for Liverpool-based clients and for businesses across the North West.

Two caveats before we start. First, every project is a bit different, so the ranges are wide — most engagements land somewhere in the middle. Second, the numbers below are for the build. Ongoing model and infrastructure costs (mostly token spend, hosting and observability tooling) sit on top of that and typically run between £400 and £4,000 per month per system, depending on volume.

We have ordered them roughly by how often we get asked for each one.

1. Document intelligence (extraction + triage)

What it is: A pipeline that reads incoming documents — invoices, claims, contracts, forms, scanned correspondence — extracts the structured fields you care about, and routes them into your existing systems. A human reviews anything below a confidence threshold.

Why it sells: It is the single highest-ROI AI project we ship. The pattern is well understood, the failure modes are visible, and the savings are easy to measure — you can count them in hours of analyst time.

Typical Liverpool clients: Insurance, legal, finance, healthcare administration.

Time to ship: Three to seven weeks. Proof of concept in two.

Budget range: £18,000 to £55,000 for the first system. Reuse on subsequent document types is usually 40 to 60 per cent cheaper because the infrastructure is already in place.

2. Internal knowledge assistant (retrieval-grounded chat)

What it is: A chat assistant tightly scoped to your internal documentation, policies, contracts or codebase. Every answer cites its source. Out-of-scope questions get a refusal, not a hallucination.

Why it sells: Every professional services firm we talk to either has one in production or wants one within six months. Done well, it deflects internal questions and frees up senior people. Done badly — see why most AI pilots fail — it gets switched off in week three.

Typical Liverpool clients: Law firms, accountancies, professional services, mid-sized software companies.

Time to ship: Four to eight weeks. The corpus prep and eval set usually take longer than the model side.

Budget range: £25,000 to £70,000 for the first deployment. Multi-tenant or per-team rollouts add to the upper end.

3. Customer-facing support copilot

What it is: An assistant that sits in your customer support channel — usually web chat, sometimes WhatsApp or email — and either answers questions directly with citations or escalates to a human. Realistic ticket deflection sits in the 30 to 45 per cent range on a clean knowledge base.

Why it sells: Direct headcount-equivalent savings on support teams that are already struggling to keep up.

Typical Liverpool clients: Retail, healthcare administration, software-as-a-service firms.

Time to ship: Six to ten weeks. The integration with your support stack — Zendesk, Intercom, Help Scout, internal — usually accounts for half of that.

Budget range: £35,000 to £90,000 for the first deployment. The variance is mostly about integration complexity.

4. Agentic workflow (multi-step automation)

What it is: A system that runs a multi-step operational workflow end to end — onboarding a new account, reconciling a financial close, processing a compliance check, triaging a support escalation. The system can call your APIs, branch on what it finds, and surface anything that needs a human.

Why it sells: These are the projects where the math gets dramatic. We have shipped agentic workflows that have saved client teams 40 to 60 hours of analyst time per week. The catch is that they are the hardest projects to get right.

Typical Liverpool clients: Logistics, financial services, mid-sized professional services.

Time to ship: Eight to fourteen weeks. The first agent is always the slowest; subsequent ones reuse the infrastructure.

Budget range: £45,000 to £140,000 for the first system. Most of the variance is in how many integration points are involved.

5. AI strategy and roadmap

What it is: A three-to-four week engagement that audits your operations and data, ranks the AI opportunities by value and risk, and produces a costed, sequenced plan your board can actually use. No production code, just clarity.

Why it sells: Most businesses spend their first AI investment on the wrong project. A proper strategy engagement usually pays for itself by killing one bad idea early.

Typical Liverpool clients: Mid-sized businesses that have either had a failed AI pilot or are deliberately trying to avoid one.

Time to ship: Three to four weeks.

Budget range: £12,000 to £30,000. The variance is mostly about the size of the operation being audited and the depth of data review involved.

6. AI enablement and training

What it is: A bundle of hands-on training, internal tooling and a lightweight governance framework so that your team becomes capable of building and maintaining AI systems themselves. Usually delivered alongside or after one of the build projects above.

Why it sells: It is the difference between AI being a capability you own and a dependency on a consultancy. We push every client toward it.

Typical Liverpool clients: Any business that wants its first or second AI system not to be its last.

Time to ship: Two to six weeks, often split across a quarter.

Budget range: £8,000 to £25,000 depending on team size and depth.

Ongoing costs to plan for

The build is the headline number; the running costs are the surprise. Reasonable monthly ranges by system type:

• Document intelligence pipelines — £400 to £1,500 per month. Mostly OCR and LLM tokens on extracted content.
• Internal knowledge assistants — £600 to £2,000 per month. Embedding refreshes and query volume drive the variance.
• Customer-facing support copilots — £1,200 to £4,000 per month at typical SME volumes. The retrieval side is cheap; the conversational side scales with usage.
• Agentic workflows — £800 to £3,500 per month. Token usage is unpredictable; budget headroom matters.

We monitor these for you in the first three months and tune them down once usage patterns are clear. It is unusual for a system to land outside its initial estimate by more than 20 per cent.

What is not on this list

A few things we get asked about repeatedly and tend to talk clients out of:

• A general-purpose internal chatbot with access to everything. The scope is wrong. Pick a domain, build the assistant for that domain, do not let it answer outside it. See the unscoped chat window failure mode.
• An AI-powered version of an existing rule engine. If your current workflow is deterministic and works, do not put an LLM in front of it. You will pay more and get worse results.
• A custom fine-tuned model when retrieval would do. Most problems are retrieval problems, not fine-tuning problems. The default should be retrieval.

How to use these numbers

The point of the ranges above is not to box you in — it is to make budgeting honest. If you are talking to a Liverpool AI consultancy and the proposal is materially above the top of these ranges for an equivalent project, you should ask why. If it is materially below, you should ask harder why — because either someone is being optimistic about scope, or someone is going to come back for a change-order later.

We covered the question of how to interrogate any proposal in our buyer's checklist for hiring an AI consultancy in Liverpool. The numbers here pair with that.

If you want an estimate for a specific project, book a 30-minute discovery call. You will leave with a sensible range, not a sales pitch.

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.