What an AI chatbot for a small business actually does (it's not what most popups suggest)

Someone fills the contact form on your website at 10:47pm. You see the email Tuesday morning at 6:15am. You reply at 8:00am. They've already booked with the place that replied first. This is what fast lead response actually means now, and it's why people keep telling you to put an AI chatbot for a small business on your site. But the real case isn't the chatbot. It's the math behind the missed reply window, and that math has nothing to do with you being lazy.

From minute five to minute ten, the odds of qualifying a web lead drop fourfold

This is the number nobody quotes you when they pitch chatbots, and it's the only one that matters. The MIT Lead Response Management Study found that from 5 to 10 minutes after a lead is created, the odds of qualifying that lead drop 4x. Not 4%. Four times. The same research found that calling a web lead at 4-6pm is 114% more effective than calling at 11am-12pm. That's when the prospect is back at their desk, not because anything about your business changed.

Translation for a local business: by the time you read the email, the lead is already cold. The same thing happens with text — 87% of consumers check a new text within 15 minutes, and 32% check immediately. The window where the lead is still warm is small, and it doesn't care whether you were closing up shop or driving home.

Almost half of inquiries come in after the lights go off

Housecall Pro's first-party data is the cleanest read here. 42% of HVAC calls and 47% of general home-service inquiries arrive outside standard business hours. 41% of jobs booked online come in after hours. Invoca's call-tracking data shows 27% of inbound calls to home-service businesses go unanswered, and Housecall Pro estimates an average of $1,200 in lost revenue per missed call. The pattern repeats in the data we wrote about in your best blog topics are already in your phone calls: owners know they're missing inquiries, but the loss is invisible until you count it.

If you do the math on your own business, it's three numbers: average ticket size, close rate on inbound inquiries, and inquiries missed last month after 6pm. That's the number this conversation is about. Most owners never calculate it because the loss is invisible. The prospect just goes somewhere else and you never know they tried you first.

The thing on the website isn't a chatbot in the sense you're imagining

You've seen the bad version. A popup says "Hi! How can I help?" The visitor types a real question. The bot responds with three menu options that don't match. The visitor closes the tab. Two-thirds of consumers report having had a bad chatbot experience, and the top two reasons are exactly what you'd guess: the bot couldn't answer their question, and it didn't understand what they needed.

The version that's worth $100-200/mo for a local business is different in three concrete ways. First, it reads from a knowledge base built from your actual services, hours, pricing, FAQ, neighborhood served — not from a script someone wrote three years ago. Second, when it can't answer, it captures the visitor's name, number, and what they asked, and emails it to you in the same minute. Third, it tells the visitor it's an AI assistant up front — which is what Quebec Law 25 requires, and what keeps the legal trail clean if anyone asks later.

The math reduces to three anchors: cost-substitution, response-time, and missed-call recovery

Three anchors, all from clean primary sources. Skip the conversion-lift percentages other people quote — there's no public clean A/B comparison of SMB conversion with vs. without a chatbot, so those numbers are all selection-biased and not worth your trust.

The honest math:

• Cost-substitution. A receptionist in the U.S. earned a median of $17.90 an hour in May 2024, per the Bureau of Labor Statistics. A full-time receptionist for a Quebec salon or clinic runs $40-50K a year. An AI chatbot subscription is in the hundreds of dollars per month, all-in. You aren't replacing a receptionist; you're filling the hours they're not at the desk.
• Response-time. Use the 4x MIT number against your own close rate. If you close 30% of inbound inquiries when you reply inside 5 minutes and 7-8% when you reply the next morning, the difference times your monthly volume is the bot's job.
• Missed-call math. $1,200 per missed call in home services, per Invoca. One recovered job a quarter covers a year of the bot.

If your business has steady after-hours inquiry flow and an average ticket above $200, the math usually says yes. If your inquiries all come Tuesday-Thursday between 10am and 4pm and your average ticket is $40, the math usually says no — keep the contact form, hire a part-time answering service, and spend the budget somewhere else.

What an AI chatbot for a small business doesn't do

It doesn't replace you. It doesn't replace your receptionist. It doesn't make decisions on price, refunds, or anything where one mistake costs you a customer. It doesn't pretend to be human. Under Quebec Law 25, identifying as AI up front isn't optional, and the bots that try to sound like a real person are the ones generating the bad experiences in that two-thirds number.

What it does: it answers the same forty questions you've answered a thousand times, in the moments you can't be at the desk, in the language the visitor wrote in. When the question goes beyond what it knows, it gets you the lead before the prospect closes the tab. That's the whole job. If you want the full delivery spec — knowledge base, escalation rules, Law 25 disclosure — we publish it at /services/ai-chatbot.

How to test a chatbot pitch before you sign

If someone's pitching you a chatbot and the headline number is a conversion-rate lift percentage, ask them which clean A/B study it came from. There isn't one. The honest case is the response-time math, the after-hours volume in your specific business, and the cost-substitution against what a human in that chair would cost. Anything else is the same generic chatbot pitch you've already rejected, dressed up in newer language.