AI & automation

Put AI to work on your own data.

We find the repetitive jobs where AI saves you time, build it into the systems you already run, and keep a person checking the results.

We have been building systems like this for clients since 1997, and we never train models on your data.

What we build.

Most of the work our clients ask for falls into four types. If yours is different, call us and describe it.

Prediction and forecasting

What will sell, what will run out of stock, which enquiries will become orders, which customers are about to leave, and what your cash position will look like in ninety days.

Reading documents

Invoices, purchase orders, contracts, forms, spec sheets and emails turned into structured data, checked, and entered into the system that needs them.

Assistants and search

Ask your own data a question in plain English, and search across everything the business knows, including documents that are usually hard to find.

Classifying and routing

Sort, prioritise, tag, score and route incoming work. Anything the system is not sure about is passed to a person.

Where the work is

What happens around the model.

For every request we: pull the right records first, check the answer against your rules before anyone sees it, send anything uncertain to a person, write the result into your CRM or shop, and keep a full log of what was read, written and spent.

We have been building this kind of reliable system for clients since 1997. The AI model is only one part of it.

Bring us a problem

Two weeks on one job.

We take one job your business does every day and test, on your own records, how often AI gets it right. At the end of two weeks you have working software and clear numbers.

01

Pick the job

Half a day with the people who do the job now. We leave with one clear workflow, the data behind it, and a number that defines success.

02

Build it

Working software running on your data, in a copy of your system with personal details removed. Not a demo on someone else’s data, and not a mock-up.

03

Measure it

Scored against examples your team has marked by hand, so you can see how often it is right, where it goes wrong, and what it costs to run.

04

Decide

Put it live, change the approach, or stop. You own everything we built either way.

What you have at the end

  • Working software on your data that your team can try
  • Clear numbers: how often it is right, the cases it got wrong, and cost per thousand
  • A straight recommendation, including when it is not worth doing
  • What it would cost to put live and to run for a year

Start with one thing

Start with one job. The second is cheaper because the connections to your systems are already built.

Models and platforms

We are not tied to any single AI model.

We work with Anthropic Claude, OpenAI, Google Gemini, Azure OpenAI, Amazon Bedrock, and open models such as Llama and Mistral on hardware we manage. For classic problems like forecasting and scoring, a small statistical model is often better than a large language model.

Which model you use is a decision we make with you based on evidence, and you can change it later without starting again.

Talk it through

Into the systems you already run.

We put the AI inside the screen where the job already happens, in the CRM, the portal, the back office or the phones we run for you.

We are equally happy working inside a system someone else built. Most of this work connects to an accounts package, ERP or warehouse that was already there long before we arrived.

How we keep it honest.

Four things we do on every AI job, so you can see how often it is right and what it costs before you rely on it.

Measured on your records

Every job is tested against a set of real examples that your team has marked by hand. That set is how we choose the approach, and it is run again on every change before anything is released.

A person on the doubtful cases

We set confidence thresholds with you. Anything below the line goes to a person, with the draft attached. The routine nine out of ten cases are handled automatically; the tenth gets a human.

Your data stays yours

UK hosting on servers we manage, or your own cloud account with your own keys. Before anything is sent to an outside AI model, names, contact details and anything else that identifies a person are taken out, and the copies of your data we test with are anonymised in the same way. We never train models on your data, and every request is logged so you can see what was sent, what came back and who saw it.

Costs you can see

Priced per thousand requests before you commit, capped once live, and reported monthly. If a cheaper model can do the job, we use it and tell you.

Tell us the job that eats a day a week.

Describe the job in an email or on the phone. Within two working days we will tell you whether it is a sensible first AI project and what two weeks of work would cost.

Already a client?

We already know your systems and your data, which is most of the head start. Raise a ticket and we will set up a call.