Live chat, chatbots, and whether they help small businesses

Chat gets sold as a conversion tool. For a business with staff assigned to it, that is fair. For a five person company where the owner is on a roof, it is a promise the business cannot keep.

The question is not whether chat works. It is whether you can staff it. Those are completely different decisions and only one of them is about software.

When does live chat actually help?

Three conditions, and you need all three.

Someone is genuinely available to answer during the hours it appears. Not "we check it a few times a day." A visitor who types a question and waits four minutes has already left, and now they have a bad impression instead of no impression.

Your buyers have a specific question that stops them from acting. Availability, coverage area, whether you handle their exact situation, whether you take their insurance. If people are hesitating over one recurring unknown, chat can resolve it in the moment.

And the answer is faster in chat than on the phone. If the honest answer is "call us," chat is an extra step between the customer and a call.

When those hold, chat catches the buyer who will not call and will not fill in a form. That is a real segment and it is not small.

Chat is a staffing commitment with a widget attached. If you cannot staff it, do not install it.

The reason most small businesses fail the first condition is structural rather than lazy. ServiceTitan's platform data shows home services shops with fewer than five technicians booking 24% of their calls against 59% for shops with 25 or more. That gap is not discipline. It is capacity, because the small operator is under a sink when the phone rings, and the same person is the one the chat widget is waiting on. Restaurants have a sharper version of the same problem: Maple's analysis of 1.2 million calls across more than 1,000 locations found 68% of calls arriving during the lunch and dinner rushes. The channel peaks exactly when the floor is least able to answer it, and a chat widget does not change who is standing there.

What does chat actually cost to staff?

Run the arithmetic before you run the trial. This is the calculation the vendor will not put in the proposal.

Volume. Say the widget produces 45 conversations a month during business hours, at an average of six minutes each. That is 4.5 hours of actual work. It sounds like nothing.

Availability. To answer within two minutes you need someone reachable for every open hour, not for 4.5 of them. Fifty open hours a week is about 215 hours a month of readiness. You are buying 215 hours of readiness to deliver four and a half hours of work.

Yield. Of those 45 conversations, suppose 12 are genuine new inquiries and the rest are existing customers, directions, and salespeople. Close one in three and that is four customers. At $960 of gross profit each, $3,840 a month, which pays for any widget on the market.

Incrementality, which is the number that decides it. How many of those 12 would have called or filled in the form anyway? If eight would have, your real gain is four inquiries, about one and a third customers, roughly $1,280 a month. You bought that with 215 hours of readiness and 198 minutes of interruption during service.

Then the tiebreaker. $1,280 a month is still real money. The question is whether that same person, given the 215 hours back, would produce more by answering the phone faster or by writing the answers onto the page.

What does a chatbot really do for a local business?

Mostly it collects contact details in a conversational wrapper. Sometimes that converts better than a form, because a single question at a time feels lighter than a page of fields.

Where bots earn their place is answering the same handful of questions perfectly, at two in the morning. Hours, service area, whether you do a specific service, how to book. Those are known answers and a bot never gets them wrong.

Where they fail is anything outside the script. The visitor asks something specific, the bot loops, and they leave more annoyed than if there had been no bot. Every failure of this kind costs you more goodwill than the successes earn.

Be careful with tools that promise to answer freely using your website content. They can, and they can also state something you never said, in writing, to a customer, in a way you will be held to. If you run one, restrict what it can answer and route everything else to a human. The same principle applies to every automation decision, and it is the same reasoning behind judging which Google Ads scripts and automations are worth using rather than switching everything on.

Set expectations plainly. Label a bot as a bot. Buyers are not offended by a bot. They are offended by discovering they were talking to one.

Do I need chat after hours?

Probably less than the pitch says, and the numbers used to sell you the after-hours case do not survive checking.

Four figures circulate in this category and none of them has a source. That 43% of calls happen after hours. That 64% happen outside business hours. That 62% of HVAC calls come after 5pm. That 52% of contractor calls arrive after hours. Every one of them appears only on vendor blogs selling after-hours coverage, with no sample size, no method, and no link to a primary document. Ask the salesperson for the study. There is not one.

What is actually measured is narrower and more useful. Slang AI, working from millions of full-service restaurant calls, reports that 66% of restaurant calls occur during business hours, that 26% are either simultaneous or after hours, and that roughly 71% of all calls are directly tied to revenue. Maple, from 1.2 million calls, says the average paying restaurant receives roughly one in five of its calls outside the hours it is open to take them. Patient Prism, from 11,552,668 dental calls across 8,280 locations, attributes 9.3% of all lost opportunities to voicemail, disconnects and hold hangups. All three are vendors reporting on their own platforms, and all three disclose their samples, which is the difference.

A defensible planning assumption, and label it as an assumption: somewhere between a fifth and a third of inbound contacts to a consumer-facing local business arrive outside staffed hours. For emergency trades the value skew matters more than the volume skew, because after-hours calls are disproportionately emergencies and emergencies are disproportionately high ticket.

One more figure to handle carefully. You will see a claim that roughly a quarter of people switch to online chat when a business does not answer the phone, attributed to a 2025 consumer survey by a call tracking company. We could not locate a primary page publishing it, and that company's website returns an identical page shell for every URL requested, including nonsense ones, so a page loading there proves nothing. Until someone produces the survey, treat it as unverified rather than as a reason to buy a widget.

What is the cost nobody mentions?

Attention, in two directions.

The first is yours. Chat generates a stream of interruptions, most of which are not leads. Existing customers asking about invoices, people who want directions, salespeople. During a busy week it is one more inbox to fail at.

The second is the visitor's. A chat bubble that slides open on its own covers content on a phone, sometimes covering the phone number or the button you actually wanted tapped. Widgets also load extra scripts, which slows the page you just optimized.

And there is the question of what the widget replaced. Owners often install chat instead of answering a question on the page, which is backwards. If forty people a month ask the same thing in chat, the page should have answered it before they asked.

If a visitor needs a lot of information before they will act, the answer is usually a fuller page rather than a chat window, which is one of the reasons long pages sometimes convert better than short ones.

What should I do with the questions people ask in chat?

Treat the transcript as the best conversion research you will ever get for free.

Every question a visitor typed is something your website failed to answer. Export a month of chats, group the questions, and count them. The top five belong on your service pages, in the visitor's own words.

You can predict most of what you will find, which is an argument for writing the answers now. Maple's taxonomy of 1.2 million calls puts hours, location and directions at 22% of all restaurant calls and menu questions at another 15%. That is more than a third of contacts spent on facts a web page can state permanently. Your trade will have a different top three. It will not have a longer one.

Do that and two things happen. Chat volume drops, which is a good outcome and not a failure. And your pages start converting people who would never have opened a chat window at all.

This is also the honest alternative to the testing programs you will be pitched. Real questions from real buyers tell you what to change, whereas split testing cannot resolve at local traffic levels and will not tell you anything for a year.

Judge the whole experiment on a quarter, not a week, and on booked jobs rather than chat volume, since measuring conversion honestly at small traffic means using windows wide enough to mean something.

What to do this week

Answer one question before you install anything. Who is answering chats, during which hours, and within how many minutes. If you cannot name a person and a time, stop here and put a tappable phone number where the widget would have gone.

Run the staffing arithmetic above with your own numbers. Readiness hours, expected conversations, and how many of those would have contacted you anyway. That last figure is the one that decides it.

If it clears, run a limited trial. Business hours only, with the widget hidden outside them. A widget that says "we are offline" is a worse first impression than no widget.

Check on a phone that the bubble does not cover your call to action or your form, and that it does not open by itself.

At the end of the month, read every transcript and write down every question. Put the top three answers on the relevant pages, in plain language, near the top.

While you are editing those pages, look at what else is missing, because photography of your actual work answers more silent questions than most chat conversations ever will.

Questions about whether chat is right for you?

Email me at eric@seod.com with the chat or chatbot quote you were given, plus who would be answering it and during what hours. I will tell you whether the staffing works and which of the questions it would field could simply be answered on the page instead.

If chat is a good fit for your business, I will say so plainly. For a lot of local businesses it is not, and knowing that before you sign a subscription is worth the email.

There is more on conversion in the library.