The cost of a first AI project follows the job you need it to do, the systems it uses and the checks that keep it useful.
Start with a free 20-minute call. A focused question may suit a £250 consultation. Significant project unknowns may need a £750 paid project plan. A clear brief may be ready for a quote straight away. These are alternative starting points, rather than a sequence of fees every customer must pay.
The cheapest first move is the one that tells you whether the idea should continue.
The first cost is deciding what not to do
AI gets expensive when the first question is too broad.
"Use AI in the business" is not a project. It is a direction of travel. It can cover sales replies, website enquiries, document checks, product data, support messages, internal search, reporting, and a dozen other things.
Trying to price all of that at once gives you either a vague number or a number with too much padding inside it.
A useful first step narrows the problem. It names the repeated job, the person who checks the output, the material the AI can read, and the mistake that would worry you most.
That is why a small paid planning step often saves money. It stops the first project becoming a bundle of assumptions.
A free call should answer the fit question
The 20-minute call is not there to solve the whole project.
It should answer three practical questions:
- Is there a repeated job here?
- Is AI likely to help with it?
- What is the smallest sensible next step?
Sometimes the answer is that AI does not belong there yet. The source material may need tidying. The website form may need better questions. The team may not agree who should check the output. In those cases, jumping straight to a working tool would waste money.
Sometimes the answer is clearer. The job is repeated, examples exist, the review point is obvious, and the cost of a mistake is manageable. Then a paid next step can be useful.
A £250 consultation helps you make one decision
A prepared consultation is useful when the question is narrow enough to discuss in one sitting.
For example: should AI read website enquiries and draft first replies? Should it help the team find answers in a set of internal documents? Should it prepare a weekly report from project notes? Should it check product descriptions before they go live?
The output should be simple: written notes, the recommendation, the likely first shape, and the reasons to continue or stop.
This is not the right step if nobody can name the job yet. In that case, an audit is usually better because it gives the business space to compare several possible uses before choosing one.
A £750 project plan defines a useful first version
The paid project plan investigates one project or business problem. You receive an editable brief with the recommended first version, a diagram or rough sketches, key assumptions and practical next steps. You own it and can take it to another developer.
It is a planning brief, not a complete technical specification or working software. Where the first stage is sufficiently understood, we provide a fixed quote for that stage. A prototype or further investigation is separately agreed and paid for if needed.
Read how we plan and price projects for the starting options.
A Website and Systems Audit reviews the whole enquiry journey
The Website and Systems Audit looks at the website and the systems behind it. AI visibility is one part of the review.
It covers five areas:
- What the website is doing and where it needs attention.
- The enquiry journey, from landing on the site to receiving a human reply.
- The systems behind the website, including where people re-type information.
- How the business is found through search, its Google Business Profile, structured data and AI answers.
- Three prioritised fixes, with prices.
The output helps you choose what to improve first, whether that involves content, a form, a connection between systems or an AI feature.
An AI Audit compares possible uses
An AI Audit is wider than the site.
It looks across the way the business works: repeated admin, inbound enquiries, document-heavy jobs, internal knowledge, customer support, reporting, handovers, and places where people keep copying information between tools.
The aim is not to produce a long list of possibilities. It is to find a small number of useful candidates and rank them by value, risk, and ease of review.
A good audit should make the next decision easier. It should say which idea is worth testing first, which ideas should wait, and which ideas are not right for AI at all.
That matters because the most expensive first AI project is often the one that looked impressive but was never close enough to a real working problem.
A working AI tool is priced around its requirements
An Agent Workflow or Knowledge System is priced once its requirements are understood. We explain the proposed work, assumptions, exclusions and payment stages before you commit. If a dependency needs investigation first, we agree that as separate work.
That might be an AI assistant that sorts enquiries for review, a report that prepares the week for a manager, a document checker that flags missing information, or a search tool that answers from approved internal material.
The price is not just the model bill. It includes the shape of the work: what the AI reads, what it produces, how a person checks it, what happens when confidence is low, and how the team can pause it if it stops helping.
This is where first projects can drift if the review point is missing. If nobody owns the output, the tool can look finished and still fail in practice.
Four things change the price
Four things tend to change the price of a first AI project.
The first is the number of systems involved. Reading one source is simpler than reading a website, an inbox, a set of PDFs, and a customer database.
The second is the risk of a wrong answer. Drafting a reply for a person to check is different from changing what a customer is told automatically.
The third is the state of the source material. AI is easier to add when the business already has current, approved content. It is harder when the knowledge lives across old files, inbox threads, and people's heads.
The fourth is the review step. A tool that prepares work for a person is usually a better first project than one that tries to make the final decision alone.
These are the things a proper first step should uncover before anyone commits to the larger piece of work.
What a sensible first budget should buy
A sensible first budget should buy clarity before complexity.
It should tell you what the job is, what data is needed, who checks the result, where the risk sits, what the first version would do, what it deliberately would not do, and what would make you stop after a trial.
If the budget only buys a demo, be cautious. Demos are easy to like because they skip the awkward parts: messy source material, uncertain permissions, edge cases, customer trust, and the person who has to check the output on a busy day.
The useful question is not "how much AI can we buy?" It is "what is the smallest useful thing we can test without creating a new problem for the team?"
Keep control of the cost
We discuss relevant hosting, software and support costs alongside the proposed work. If your requirements change, we explain the effect on price and timing and agree it with you before doing the additional work.
A useful first stage should fit an agreed purpose. Adding three more systems to connect is a change to that purpose, even when the screen the customer sees looks the same.
Where to start
If you already have one repeated job in mind, start with a Free 20-minute call. We can usually tell whether it needs a consultation, a Website and Systems Audit, an AI Audit, or a more direct Agent Workflow.
If you have several possible ideas and no clear first choice, start with the AI development approach and look at the audit routes.
The aim is not to spend the largest sensible amount. It is to spend the smallest amount that gives you a real decision.
