How much does an AI consultant cost a small company?

What an AI consultant costs is usually the first question an owner asks, and the answer goes beyond the consultant's fee, because the AI itself has to be paid for, and so does the time your own people spend on the project. We do not publish prices on our site, because the same request ('we want AI to handle our supplier invoices') can turn into jobs of very different size from one company to the next. The cost depends on a handful of factors, and whoever runs the company can judge most of them without a technical background.

How many tasks, and how often do they happen?

In a small company, an AI project is in practice a set of specific tasks, such as reading supplier invoices and entering them in the ERP, or sending a first reply to enquiries that arrive through the website. Each task has its own cost, so the first useful question is how many there are and how often they happen.

Frequency helps decide whether a tool is worth building at all, since a task repeated hundreds of times a month can justify a larger investment than a report produced four times a year. When the AI is paid for per use, frequency also feeds into the monthly bill.

Variety counts as well. In a purely illustrative example, invoices from five suppliers, all in the same layout, take less work to automate and test than invoices from eighty suppliers, each with its own format and its own exceptions, because each special case adds time to the build and to testing.

We usually advise starting with one or two tasks and adding others once the first ones are in daily use, since automating everything at once makes the project more expensive and asks more of the team than it can take on.

Which systems does the tool connect to, and what state is the data in?

Almost every useful tool has to read from or write to systems the company already uses: email and documents in Microsoft 365 or Google Workspace, the ERP or invoicing software, sometimes a CRM. When the tool works inside Microsoft 365 or Google Workspace, which is where we build, the connection tends to be the most predictable part of the work.

ERP and invoicing software vary more. Some programs have an API, a documented way for other programs to read and write their data; others only export files, and some older versions installed on the company's own server allow little more than that. Each case changes the work and, with it, the cost. Before asking for a quote, it is worth asking your software supplier whether the program has an API and whether access is included in the licence you already pay for.

Then there is the data the task depends on. A tool that answers customers about delivery times is only right if the delivery table is up to date. If the information is scattered across shared folders, with several versions of the same document, or kept in a spreadsheet only one person can read, someone has to put it in order first. That work is part of the cost, whether we do it or your team does, and it is better to know about it before starting.

Do you pay for AI per user or per use?

Apart from the build, you pay whoever provides the AI model, usually in one of two ways.

With a subscription, such as ChatGPT Business or Claude's Team plan, you are charged per user (Anthropic says per seat) per month. The cost is predictable and grows with the number of people who have access. It makes sense when people use AI in their own work, to draft text or summarise long documents.

An automated tool, such as one that reads invoices, is usually paid for per use: it calls the model directly through an API, and the charge depends on the amount of text it processes, measured in tokens (small pieces of words). Because the bill follows the volume and rises in months with more invoices, the count of tasks above also helps to forecast it.

A company can perfectly well have both. The accounts stay in your company's name and we take no commissions from any vendor, so the choice between them depends on your case. Vendors change their prices, so check their own pages at the time you decide.

Your team's time is part of the cost

The time your own people give to the project is an easy cost to forget. Someone has to explain how the task is done today, including the awkward cases, and test the tool with real documents before it goes into use. The team also needs training on the new tools, ideally built around its own tasks.

Review takes time as well, because in the tools we build a person reviews what the tool produces before anything is sent. That review protects the company from the mistakes a tool will sometimes make, and since it has to be done every week, it should be in the sums from the first day.

What does it cost to keep the tool running after launch?

An AI tool needs looking after once it is in use, because the systems around it change: the ERP gets a new version, or the AI model behind the tool moves to a newer version that answers in a slightly different way. The work itself changes too, when a new supplier comes on board or the company redesigns its proposal template.

Someone has to keep the tool running, fixing what breaks and adjusting it as the work changes. That can be a person inside the company, if they have the time and the know-how, or an outside supplier. Our monthly support, which comes in three levels, covers this work and includes a check-in each month. When you compare proposals, add this monthly cost to the cost of the build.

Questions to ask any AI consultant

These questions work for any proposal, ours included, and help show what a quote covers:

• Is the estimate made per task, and what does each one include?
• Are licences and AI usage fees included or paid separately? Whose name are the accounts in?
• Does our ERP or invoicing software allow the connection you need, and who checks that?
• How much time will our team spend explaining the work and reviewing what the tool produces?
• What happens when the tool stops working six months from now, and what does fixing it cost?
• Where will our data be kept, and will you sign a GDPR data-processing agreement before work starts?

How to get an estimate before spending much

The safest way to know what it will cost is to ask for an estimate per task from someone who has seen how the work is done in your company, which is what our two-week diagnostic does. We start with short interviews with the people who do the work and finish with a written plan: the 3 to 5 tasks most worth automating, each with an estimate of time and cost. If you go ahead with a build within 60 days, the diagnostic is credited against it.

Having the plan in writing lets you decide in your own time and, if you wish, go through it with your business partner or your accountant. If you already know which task you want to solve first, the Start programme brings together, over 30 days, the workshop AI in daily work, two working tools built on the systems you already use and 30 days of support. Either way, we start with a free 30-minute call.