Seven jobs an AI agent can do for a Yorkshire business

Yorkshire and the Humber has around 191,000 VAT or PAYE registered businesses, about 7% of the UK total (UK business: activity, size and location, ONS, 24 September 2025). Very few of them are tech companies. They make things, move things, care for people, build, sell, advise and repair. And the AI conversation, as it reaches them, is mostly about chatbots.

That’s a shame, because the jobs AI is genuinely good at in a business like that are not chatbot-shaped. They’re paperwork-shaped. Here are seven we see across the county, with a note on each about where the line is.

1. Reading the inbox and sorting it (every sector)

Most small businesses run on a shared inbox that one or two people spend the first ninety minutes of every day triaging. Enquiries, invoices, supplier updates, complaints, spam, the odd thing that’s genuinely urgent.

An AI agent can read every incoming email, work out what it is, pull out the facts (who, what, how much, by when), log it where it belongs and draft a reply for a person to check. The person’s job becomes approving instead of typing. This is the single most common thing we build, because it’s cheap, it’s quick, and it’s wrong in ways that are easy to see.

The line: draft, don’t send. Keep a human on the send button until you’ve watched it be right for a month.

2. Turning an enquiry into a quote (engineering, construction, trades)

Sheffield still makes things. So does the whole M1 corridor down through Rotherham and Doncaster, and the Calder Valley, and plenty of Bradford. A lot of those firms quote the same way they did in 1995: an estimator reads the drawing or the email, looks up past jobs, works through the price book, and types it up. Days, sometimes.

An agent can do the reading and the first pass. It pulls the specification out of the enquiry, matches it against jobs you’ve done before, applies your pricing rules and produces a draft estimate with its working shown. The estimator corrects it rather than starting from nothing.

The line: the estimator signs it. Always. A quote is a commitment, and the agent doesn’t know that the client in question is always late paying.

3. Chasing the money (every sector)

Credit control is the job everyone hates, puts off, and loses money on. Automation does it without embarrassment: watches the ledger, notices what’s overdue, sends the polite nudge at seven days, the firmer one at fourteen, and flags the thirty-day cases to a person with a summary of the account history. The AI part is writing the messages so they sound like you rather than a robot, and reading the replies (“cheque’s in the post”, “we disputed this invoice in March”) to route them properly.

The line: no automated threats, no automated referrals to collection. Those are human decisions with legal consequences.

4. Compliance paperwork (care, food, construction, transport)

If you run a care home, a food business, a haulage firm or a building site, you generate a staggering amount of paper that exists to prove to a regulator that you did the thing. Audits, checks, incident reports, statutory notifications, training records.

This is where AI earns its keep twice over. First by making the form faster to fill (dictate it, let the system structure it). Second, and more valuably, by reading across all of it to tell a manager what’s going wrong before the inspector does. We built Guardian Care for exactly this, for a care group with homes across the country, and the reading-across-everything part is what the regulator noticed.

The line: the system supports the registered manager’s judgement. It does not make it. And every shred of personal data in it needs handling under the ICO’s AI guidance.

5. Onboarding a new client or job (professional services, agencies, Leeds generally)

Leeds has the biggest legal, accounting and financial services cluster outside London, and every firm in it has an onboarding process that involves the same information being typed into four systems by three people. Engagement letter, client record, risk check, matter opened, welcome email, first meeting booked.

An agent takes the signed engagement and does the lot, pausing only where a person genuinely needs to decide. The same pattern works for a marketing agency taking on a client, a recruiter placing a candidate, or a letting agent signing a tenant.

The line: anti-money-laundering and identity checks get a human look. The agent can gather and summarise; it shouldn’t be the one who’s satisfied.

6. Research and first drafts (marketing, sales, anyone who writes)

The least glamorous and most reliable use. Finding what competitors are ranking for, answering the questions your customers keep asking, turning a subject expert’s twenty-minute ramble into a usable article. We built a set of automations that does precisely this for a marketing agency, and you can read how in the AI Content Lab case study.

The line: first draft, never final. AI prose reads like AI prose, and your customers can tell.

7. Answering the questions you’ve already answered (every sector)

Not a chatbot that pretends to be a person. A tool that knows your opening hours, your delivery areas, your returns policy, which of your products fits which model, and can say so accurately, with a link to the page, and hand over to a human the moment it’s out of its depth.

This is the one people ask for first and should usually do last, because it’s customer-facing and the failure is public. A German court held a company liable in May 2026 for things its website chatbot invented, which we wrote about in who’s liable when your chatbot gets it wrong.

The line: it answers from your documents only. It never guesses. It escalates early. And someone reads the transcripts every week.

What ties them together

Every one of these is the same shape: information arrives, someone reads it, decides something routine, and types it somewhere else. The ONS found in July 2026 that adoption is 58% in information and communication businesses and 13% in construction (ONS, 20 July 2026). It’s tempting to read that as construction not needing it. We read it as construction having the most of this shape of work still done by hand.

And every one of them keeps a person where the consequences are. That’s not us being cautious. It’s what the evidence says works, and it’s what we’d want if it were our business.

If one of the seven sounds like a job in your firm, tell us about it. We work with businesses across Leeds, Sheffield, Bradford, Huddersfield, Wakefield, York and the rest of Yorkshire, and we’ll say honestly whether it’s worth doing.