03 / AI

AI agents that do one job properly

There is a great deal of noise about AI and very little of it survives contact with an actual small business. The version that works is narrow: one repetitive job, clearly defined, handed to software that does it consistently at three in the morning.

From £1,500 — one agent doing one job, in production

An agent is not a chatbot bolted to your homepage. It is a process that reads something, decides something and does something — replies to an enquiry, pulls the details out of an invoice, writes the first draft of a quote, flags the email that actually needs you.

The failure mode is always the same: someone buys a general-purpose assistant, nobody owns it, and three months later it is switched off. So every build here starts with one measurable job and a number that tells us whether it worked.

What this covers

AI agent development in practice

01

Enquiry handling & lead qualification

Answers the routine questions instantly, asks the qualifying ones, and passes you a summary with a recommendation. You stop losing the enquiries that arrive at 9pm on a Sunday and stop ringing back the ones that were never going to buy.

02

Document & admin automation

Reads invoices, purchase orders, certificates or forms, pulls out the fields that matter and puts them where they belong. The work experienced people should not be doing, done in seconds.

03

Internal knowledge assistants

An agent that has actually read your procedures, pricing and past quotes, so your team gets a correct answer instead of asking the one person who knows. Scoped to your documents, with the sources shown.

04

Custom workflow agents

Something specific to how you operate — triage, drafting, monitoring, reconciliation. If you can describe the job precisely and it happens often, it is probably a candidate.

How a build runs

No surprises in the scope.

Everything below is part of the quoted price. If something you need is missing, it goes in the proposal with its own number rather than appearing on the final invoice.

  • A discovery session that picks the one job worth automating first
  • A defined success measure agreed up front — time saved, or enquiries answered
  • A working prototype you can try before the full build is committed
  • Guardrails: what it must never do, and when it hands over to a human
  • Integration with the tools you already use — email, CRM, spreadsheets
  • Hosted on Azure, with your data staying inside your own environment
  • A written handover, plus training for whoever will own it

Where AI still is not good enough

Anything where a confident wrong answer is expensive — final legal or financial advice, irreversible actions, regulated decisions — should keep a person in the loop, and I build them that way. Jobs that happen a handful of times a month rarely repay even a modest build cost. And if your process is not written down anywhere, automating it will just make the confusion faster; that is a process job first, and it is cheaper.

AI agent development — common questions

What can an AI agent actually do for a small business?

Reliably: answer common enquiries out of hours, qualify and summarise leads, extract data from documents, draft replies and quotes for a human to approve, monitor an inbox and flag what matters, and answer staff questions from your own procedures. Those are jobs with a clear input, a clear output and a way to check the result.

How much does it cost to run?

The build starts at £1,500. Running costs are usage-based and for a typical small business agent land somewhere between £15 and £80 a month. You get an honest estimate for your volume before you commit, and a cap so it cannot surprise you.

Is my data safe?

The build runs in your own Azure environment, your data is not used to train anyone else's model, and I will tell you exactly which third-party services see what before you sign anything. If you handle personal data, we cover the GDPR position in the scoping session rather than afterwards.

What if it gets something wrong?

It will, occasionally — so it is built assuming that. Agents work inside defined limits, hand over to a human when they are unsure, and log everything so you can see what happened. For anything consequential the agent drafts and a person approves.

Thinking about ai agent development?

Fifteen minutes on the phone, no obligation. You'll get a straight answer on whether it's worth doing and roughly what it costs.