Practical AI automation for UK small businesses

The work we actually automate

AI is good at a narrow set of jobs and bad at plenty of others. These are the ones that pay for themselves most reliably, and they are all things you are probably doing by hand right now.

Where we tell people not to bother

Half of this job is saying no, so here is our test before we quote anything. A task is a bad fit for automation if a wrong answer is expensive and hard to spot, if the work is different every single time, if the rules live only in one experienced person's head and nobody has written them down, or if it happens rarely enough that the automation costs more than the time it saves.

Anything involving a final decision about a person - hiring, discipline, eligibility, credit, safeguarding - stays with a person. We will build the tool that gathers and summarises the evidence; we will not build the thing that decides. That is a hard line, not a negotiating position, and it is also what the ICO expects of automated decision-making under UK GDPR.

What we do with your data, in plain terms

Sending your business data to a model provider is a data-processing decision, so we treat it as one. Before anything is built we write down what data the workflow touches, which provider processes it, where that processing happens, how long anything is retained and what the lawful basis is. You get that document whether or not you go ahead with the build.

In practice we default to sending the least data that makes the task work, stripping identifiers where the job doesn't need them, using providers whose terms exclude your data from model training, and keeping an audit log of what went in and what came back. If your sector rules mean data cannot leave your control at all, say so at the first conversation: sometimes the answer is a smaller model running on infrastructure you own, and sometimes the honest answer is that the task should not be automated.

How a pilot works, and what it costs

We start with a workflow inventory, not a tool. In one session we list the repetitive tasks across your week, and score each on four things: how often it happens, how long it takes, how much judgement it needs, and how bad a wrong answer would be. High frequency, high time, low judgement, low blast radius is where the money is, and it is usually not where people expect.

Then one pilot, on one workflow, priced from £1,500 and scoped to four to six weeks. It runs alongside your existing process instead of replacing it, so you can compare outputs before trusting it. We agree what success looks like as a number before starting - hours saved a week, or turnaround time, or error rate - and if the pilot misses it, we say so instead of proposing phase two.

Larger builds, where automation is a feature of a bespoke system and not a bolt-on, are quoted as custom software from £5,000. Ongoing running costs are the model provider's usage charges, which you pay directly and can see, plus maintenance if you want us watching it.

Proof: we run an AI product of our own

PolicyMind is our compliance platform, and it uses AI where it holds up: drafting policy documents from a structured brief, summarising long documents, and answering questions grounded in an organisation's own policy library, not in whatever a general model happens to believe.

Just as instructive is where it doesn't. Approval, publication and staff acknowledgement are deliberately manual, because a policy nobody signed off is worse than no policy. That balance - automate the drafting, keep the deciding - is the same one we bring to your workflows, and it comes from running the thing ourselves, not from a whitepaper.