Let an AI agent do the lookups, and keep your best people for the conversations

Think about the last enquiry your sales team answered. The reply took a minute to write. Getting ready to write it took twenty.

In businesses that sell to trade customers, the pattern is remarkably consistent. An email arrives asking for a price or availability. Before anyone can answer, they must:

1. work out what’s actually being asked;
2. find the product in the catalogue;
3. check the price list (and which price this customer gets);
4. look the customer up in the accounts system;
5. check stock;
6. write the reply.

Steps two to five are lookups. They need care, but not judgement. And they’re spread across tools that don’t talk to each other, so a person does the walking between them.

What an enquiry agent does

We built exactly this for a timber merchant, as described in the case study. An AI agent reads the incoming email, works out what’s being asked, checks the product catalogue, the price list and Xero, and drafts the response. Anything unusual or unclear is flagged for a person.

The important design decision is in that last sentence. The agent handles the routine; the human handles the exceptions. That’s what makes it safe to use and pleasant to work alongside.

Why this is a good first AI project

Plenty of small businesses are curious about AI and unsure where to begin. Recent UK surveys paint a consistent picture: usage is climbing fast (the British Chambers of Commerce found 54% of firms actively adopting AI in 2026, up from 35% the year before), but confidence lags. In Simply Business’s research, 39% of small business owners said they weren’t sure how AI could be useful to them.

An enquiry agent is a good answer to “where do we start?” because:

• The task is already defined. Your team follows a routine today, so there’s a clear “before” to compare against.
• The sources are your own. The agent works from your catalogue, your prices and your accounts, not from guesswork.
• A person stays in the loop. Drafts can be reviewed before sending, which builds trust gradually.
• The payback is easy to see. Replies go out faster, and your experienced people stop spending their day on lookups.

What makes the difference between a demo and something you can trust

It’s easy to make an AI write a plausible email. It’s harder to make one that’s right. In our experience, four things matter most.

Ground it in real data. The agent should quote prices and stock that came from your systems, not from what it thinks sounds right. If it can’t find the answer, it should say so.

Define what “unusual” means. A customer asking for a product you don’t list, a price that doesn’t match, a tone that sounds upset: these go to a person. Agreeing those rules with your team is half the project.

Keep a human approval step at first. Start with drafts. Once the team trusts the output on routine enquiries, you can decide how much to let through automatically.

Be careful where data goes. Security and privacy are the top concern for small businesses that haven’t adopted AI. A sensible build keeps customer data in your own accounts, limits what’s sent to the model and logs what happened.

Where it doesn’t help

We’d rather tell you now. An enquiry agent is a poor fit if:

• your enquiries are mostly complex, bespoke quotes that need real judgement every time;
• your product and price data is a mess (the agent will faithfully reproduce the mess; tidy that first);
• you get a handful of enquiries a week (a template might be all you need).

Is it worth it?

Count the enquiries you get per week, multiply by the minutes of lookup work each one needs, then by the hourly cost of the person doing it. The sum is rarely small, and it lands on your most experienced and expensive people. That’s the case for handing the lookups over.

Where to start

If your sales inbox looks like this, tell us about it. We’ll begin with a short discovery sprint: follow an enquiry from inbox to reply, write down every lookup and judgement call, and show you which parts are worth automating. For more on where AI earns its place, see our Automate service.