Ask ten small business owners whether chatbots are worth it and you'll get ten answers shaped by whatever they last read. Half have been told chat is now essential. The other half tried one years ago, hated it, and never looked again.

Very few of either group have actually run the numbers.

We build conversational software for a living, so consider our bias declared. What follows is the calculation we'd want a sceptical owner to run before spending a penny with us or with anybody else. It takes about ten minutes and a rough sense of your own margins.

The short answer

For most local service businesses, yes. The reason is unglamorous: one additional customer per year usually covers the entire cost.

For a low-ticket online shop, a business where a customer spends £20 once and never comes back, the answer flips. You'd need a steady stream of incremental sales to justify the line item, and chat on its own rarely delivers that.

The deciding factor is arithmetic you already have the inputs for.

Start with one number: what a single new customer is worth

Almost every owner we speak to knows their average job value. Far fewer can tell us what a customer is worth over the whole relationship, and that gap is exactly where the worth-it debate gets stuck.

Average job value understates the truth badly for service businesses. A boiler service becomes a boiler replacement. A first filling becomes twenty years of check-ups. A one-off conveyancing job becomes the client's second property, and their sister's.

Four inputs get you close enough.

Multiply the first three, then add the referral effect. Here's what that looks like for a mid-sized trade.

Those figures are illustrative, not measured. Swap in your own and the shape of the answer rarely changes much. For anyone doing skilled work at a few hundred pounds a visit, a customer is worth thousands over the relationship.

Two of the four inputs cause arguments, so here's how we'd settle them.

On gross profit versus revenue, use profit if your margin is thin and revenue if you genuinely don't know. Getting the number roughly right beats spending an afternoon getting it precisely right, because the decision you're making is binary and the gap between the two answers is usually enormous.

On referrals, be conservative. If you reckon a happy customer sends you one more every four years, that's 0.25, not 1. Understating it keeps you honest and the conclusion generally survives anyway.

This is also why retention economics get so much attention. The classic Bain work on customer loyalty found that small increases in retention produce disproportionate profit gains, because acquisition is paid for once while the margin keeps arriving year after year. A new customer isn't a transaction on your P&L. They're an annuity with a shelf life.

The break-even question that actually matters

Once you have that figure, stop asking whether chatbots work in general. Ask the version you can actually answer.

How many extra customers a year does this need to win before it has paid for itself?

For most local service businesses the answer is less than one. That's a genuinely low bar, and it reframes the decision from "is this technology impressive" to "do I believe it will produce a single conversation I'd otherwise have missed".

The third column is the honest one. When a customer is worth under a couple of hundred pounds, break-even stops being one conversation and becomes a sustained volume of them. That's a much harder promise for any tool to keep, and you should be suspicious of anyone who makes it casually.

There's a second reason we prefer the break-even framing to a percentage return. Percentages invite fantasy. A supplier can quote "300% ROI" without ever touching your numbers, whereas "this needs to win you one extra bathroom fit-out a year" is a claim you can weigh against your own experience in about four seconds.

What a chatbot actually costs, including the parts nobody quotes

The subscription is the visible cost and usually the smaller one. Budget for three others before you decide.

Setup time. Somebody has to tell the system what you charge, where you travel to, what you refuse to quote for over the phone, and how you want leads formatted. Expect a few hours if the tool is well built and considerably more if it isn't. This is time from the person who knows the business best, which is rarely the cheapest hour in the company.

The first month of tuning. Real visitors ask questions you never anticipated. Reading the first fifty transcripts and correcting the four or five things the bot fumbles is what turns a mediocre deployment into a good one. Most of the disappointed owners we meet skipped this entirely.

Follow-up capacity. More captured leads means more callbacks. If your day is already full, the honest cost includes whatever it takes to answer them promptly, whether that's an hour of admin time or a phone diverted differently.

None of these are large for a small business. They are, however, the difference between a tool that returns nothing and one that returns many times its price, and they're never on the pricing page.

Where chatbots genuinely earn their keep

Four situations account for most of the return we see, and every one of them concerns leads that were already on your site.

Enquiries outside working hours. Evenings and weekends are when people research tradespeople, clinics and solicitors, and they're precisely when nobody is answering. We've written separately about the cost of missed after-hours leads, and it's usually the single largest leak in a small business funnel.

Speed of first response. Enquiries decay fast, and the decay curve is steeper than most owners assume. Harvard Business Review's audit of thousands of lead responses found firms that replied within an hour were vastly more likely to have a meaningful qualifying conversation than those that waited even a little longer.

That research predates modern chat tools, and it's about outbound response rather than on-site conversation. The underlying human behaviour hasn't changed though. Somebody comparing three local firms on a Tuesday night gives their job to whoever engages first, not whoever is best.

Qualification before you pick up the phone. A chatbot that asks for postcode, job type, timeframe and rough budget hands you a lead you can triage in seconds. That saves the callback you shouldn't have made, which is worth real money when the person answering the phone is also the person on the tools.

The saved time compounds in a way that's easy to miss. A tradesperson who stops returning six out-of-area enquiries a week gets an afternoon back every month, and that afternoon is either billable or it's the evening they were going to lose to paperwork.

Traffic that would never have filled in a form. Static contact forms ask for commitment before they give anything back, and most visitors decline. The mechanics of that difference are covered in our piece on chatbots versus contact forms, along with the reasons a conversation converts a browser that a form loses.

If you want the underlying behaviour explained properly, our breakdown of how lead-generation chatbots convert goes through it stage by stage.

Where a chatbot is not worth the money

We'd rather you didn't buy one than buy one that sits there irritating people.

The traffic point deserves emphasis. A chatbot converts visitors, it doesn't create them. Twenty sessions a month means a handful of chats at best, and no amount of clever configuration rescues that. Fix acquisition first and come back to this question in six months.

The follow-up point matters just as much. Capturing a lead at 11pm achieves nothing if nobody reads it until Thursday. The tool creates an obligation, and if you can't meet it you've built a faster way to disappoint people.

There's a subtler failure case too. If your enquiries genuinely require professional judgement in the first sentence, say a solicitor handling contested probate, a bot that tries to triage the matter can do quiet harm to your credibility. In those businesses, keep the scope narrow: capture the name, the number and one line about the matter, then get out of the way.

The objections we hear most, answered straight

"I tried one and it was useless." Most likely you tried a scripted decision tree, which could only answer questions somebody had anticipated in advance. Modern systems read your own material and answer in your own terms. That's a real difference, but it only shows up if somebody actually loads your pricing, service area and exclusions during setup.

"It'll invent prices." It can, if you let it. The fix is boundaries rather than optimism. Tell it what it may quote, what it must never quote, and what to say when it's out of scope. A well-configured bot saying "I can't price that accurately without seeing it, but I can get Dave to call you in the morning" is a perfectly good answer.

"My customers are older and won't use it." Sometimes true, often assumed. The people typing at 9pm are frequently the same people who'd have phoned at 9am, and chat volume tends to surprise owners in both directions. Give it 90 days of data before you decide on their behalf.

"What about data protection?" A chat widget collects personal data, so the usual duties apply. Say what you collect and why at the point of capture, keep it only as long as you need it, and check where your supplier stores and processes it. The ICO's guidance on artificial intelligence and data protection is the plain-English starting point, and it's linked at the foot of this article.

What separates a chatbot that pays back from one that doesn't

The variance between a chatbot returning ten times its cost and one returning nothing is almost entirely down to setup rather than brand.

Get those right and the tool behaves like a member of staff who never sleeps and never forgets to ask for a postcode. Get them wrong and you've bought a widget. The wider thinking behind this sits in our guide to conversational marketing for service businesses.

How to prove it in 90 days

Don't take anyone's word for the return, including ours. Set the test up so the answer is unambiguous before you switch anything on.

• Write down your baseline firstEnquiries per month, by source, for the three months before you start
• Record your lifetime value estimateThe four-input calculation, saved somewhere you'll find it again
• Tag every chat-sourced leadSo you can tell chat leads from phone and form leads at the end
• Track quoted work, not just conversationsChat volume is a vanity metric; quotes issued and jobs won are not
• Note how many arrive outside working hoursThis is usually the clearest evidence of genuinely incremental leads
• Review at 90 days against break-evenWon work versus twelve months of cost, not versus your hopes

Ninety days is roughly the shortest honest window for a business with a sales cycle measured in weeks. Judge it sooner and you're reading noise rather than signal.

One caution on interpretation. If chat leads climb while phone calls fall by a similar amount, you've shifted channel rather than grown. That's still worth something, since chat leads cost less to handle and arrive pre-qualified, but it isn't the outcome the break-even maths assumed.

The cleanest evidence is usually the out-of-hours count. A lead captured at 10:40pm on a Sunday, from a postcode you serve, for a job you do, is about as close to incremental as attribution gets in a small business.

So, are chatbots worth it?

For a local service business with real website traffic and a customer worth four figures across the relationship, we think the answer is clearly yes, and the reason is the low bar rather than the technology. One extra job covers the year. Everything after that is margin.

For a low-ticket, one-off, low-traffic business, we'd spend the money on getting more people to the site first and revisit chat once there's something worth converting.

Run your own four numbers before you decide. If a single customer covers twelve months, the risk is small enough that the only real question left is whether you'll answer the leads it finds.