Straight answer: An AI chatbot can answer questions faster than a human ever could, but for most small businesses it also stops the phone from ringing, and the phone call is usually where the real money was. If you have installed a chatbot in the last year and your average order value or job value has quietly dropped, this is why, and there is a fix that does not involve ripping the thing out.
The bot that made a heating engineer poorer
Last year I worked with a heating engineer in Kent, one of those small trades businesses with three vans and a booking system held together by a spreadsheet and a lot of goodwill. He was proud of himself. He had installed an AI chat widget on his site, the kind that answers "do you cover boiler servicing" and "what areas do you cover" at 11pm without him lifting a finger. He told me it had cut his admin time in half.
It had. It had also cut his average job value from around 280 pounds to 90 pounds in about six weeks, and his overall bookings were down 18 percent. Nobody had told him this was happening because on paper the numbers looked fine, the chatbot had "handled" over 300 conversations that month. He was busier than ever, answering fewer calls, and making less money.
Here is what was happening. Before the bot, someone with a dead boiler on a cold Tuesday would ring him directly, panicked, and he would talk them into the full service plus a callout the same day, sometimes an upsell to a new thermostat. That is a 280 pound conversation. After the bot, that same person typed "boiler not working" into the chat widget, got a polite, accurate, entirely reasonable answer about how to book a standard callout, filled in a form, and booked the cheapest slot on the price list. The bot was doing exactly what it was built to do. It was just doing the wrong job. It was answering questions when it should have been starting a sales conversation.
The uncomfortable part most people skip past
Nobody selling you an AI chatbot tool is going to tell you this, so I will. Most small business chatbots are not installed because customers asked for them. They are installed because the owner wanted to feel modern, or a marketing agency sold it as "24/7 availability," or someone saw a competitor had one. The customer research almost never happens. Nobody sits down and asks whether their highest value customers want to type to a bot at all, or whether those customers were the ones who used to ring up, chat for four minutes, and get talked into the bigger package.
The bot is measured on the wrong thing. Response time, resolution rate, conversations handled, all of those look brilliant on a vendor's dashboard. None of them measure average transaction value, and almost nobody checks that number before and after installing a bot, so the drop goes unnoticed for months. I have seen this on service businesses, on ecommerce stores, and on a coaching business that switched all its inbound enquiries to a chatbot and watched its high ticket bookings fall by a third while low ticket downloads went up. The dashboard said success. The bank account said something else.
Where AI chat earns its place
This is not an argument for switching everything off. I use AI-assisted tools myself, and there are places where a bot is clearly the right tool for the job:
- Answering the same ten factual questions over and over, opening hours, delivery times, returns policy, sizing charts
- Qualifying out people who were never going to buy, wrong location, wrong budget, wrong product entirely
- Catching enquiries at 2am that would otherwise sit unanswered until 9am and go cold
- Ecommerce order tracking, which is tedious for a human and fine for a bot
What a bot should almost never do on its own is handle the first contact for anything with a wide price range, anything emotional, or anything where trust is the product. A boiler breakdown is emotional. A wedding photography enquiry is emotional. A first consultation for a business coach is emotional. Those are exactly the conversations that build the case, in a human voice, for spending more, and they are exactly the conversations most owners hand straight to the bot because it is the busiest, most stressful part of the day.
The five minute audit I now run for every client
Before I let a client near a chatbot, I make them pull three numbers from before the bot went live and three from the most recent full month. It takes about five minutes if the booking or sales data is in one place.
- Average transaction value, before and after
- Number of enquiries that turned into the top price tier or top three products, before and after
- Number of enquiries that went to voicemail or were never answered, before and after (this is the one the bot is supposed to fix)
If average transaction value has dropped by more than about 10 percent, the bot is filtering people into the cheap, obvious option before a human ever gets the chance to sell them the bigger one. That is fixable. It is not a reason to delete the bot, it is a reason to redesign what it is allowed to do.
What the fix looks like
For the heating engineer, we did three things. First, we changed the bot so that any conversation containing words like "not working," "broken," "leak," or "emergency" triggers an immediate callback request with a real phone number, not a booking form. Second, we removed the price list from the chat entirely, because once a number is visible people anchor to the cheapest one and never ask what else is available. Third, we set the bot to only quote prices after it had asked at least two qualifying questions, which mimics what a good phone receptionist does naturally without thinking about it.
Within two months his average job value was back above 220 pounds and his bookings had recovered most of the drop. The bot was still handling the boring 40 percent of enquiries it was always good at. It just stopped being the first and only voice a high value customer heard.
The channel problem people miss entirely
There is a second layer to this that almost nobody talks about, and it is about where the conversation happens, not just how it is handled. A lot of small businesses now route everything, chat, email, social messages, into one AI tool and treat it as a single funnel. It is not. Someone messaging you on Instagram is behaving completely differently to someone filling in a contact form on your website, in the same way that the audience GoPro built by putting real customer footage front and centre in their marketing behaves nothing like someone reading a cold email. If your bot answers an Instagram DM with the exact same script it uses on the website, you are throwing away the informal, trust-building tone that made Instagram DMs convert so well in the first place. Different channel, different intent, different script. Most bot setups I audit use one script for everything because it was faster to build that way.
Email is its own case entirely. If your chatbot is quietly hoovering up the email addresses that used to go into your newsletter list and just answering the question instead of ever capturing the contact, you are losing the long game. A well-run list, the kind Mailchimp's own case studies are built on, compounds over months. A chatbot answer that closes the loop in ninety seconds compounds nothing. Make sure every bot conversation that does not end in a sale still ends in an email capture, or you are trading long term relationship building for short term convenience.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Why "efficient" is the wrong word to chase
Joe Pulizzi built an entire career on the idea that the businesses which win are the ones that build an actual audience and actual trust over time rather than chasing the fastest possible transaction, and it is worth remembering that when you are tempted to automate the exact moment where trust gets built. I wrote about some of the lessons from his approach here, and the short version is that the slow, human bit of the sales process is usually the bit doing the actual work, even when it looks inefficient on a spreadsheet.
The same logic applies to product-led businesses. Figma grew a huge amount of its early traction through community and word of mouth rather than automating support away, and if you look at how they built that reputation, a lot of it comes down to making people feel heard by an actual person at the exact moment they were deciding whether to trust the product. That is the bit a chatbot cannot fake, no matter how well it is trained.
A simple rule I now give every client
If the question is factual, let the bot answer it instantly, that is what it is for. If the question has an emotional word in it, urgent, broken, worried, disappointed, or if the price range for what they might buy varies by more than about 50 percent depending on what they need, get a human into that conversation within minutes, not hours. That single rule alone fixed more revenue leaks for my clients last year than any prompt engineering or fancy integration ever did.
None of this means AI chat is a bad investment. It means it is a tool that needs a job description, the same way you would not hand a brand new junior staff member your highest value sales calls on day one without training them first. If you are not sure whether your current setup is helping or quietly costing you, that audit is one of the first things I do with clients who bring me in through my AI consulting work for small businesses, and it usually takes less than a week to find the leak once you know where to look.
What to check this month
- Pull your average transaction value for the three months before your chatbot went live, and the last three months. Compare them honestly.
- Read the last twenty chatbot conversations that did not end in a sale. Count how many mention an emotional word or an urgent situation.
- Check whether your bot hands over to a human before or after it quotes a price. After is almost always too late.
- Confirm every conversation that does not convert still captures an email address for your list, not just a closed chat window.
Frequently asked questions
Do AI chatbots reduce sales for small businesses?
Not always, but they commonly reduce average transaction value because they tend to hand out the cheapest, simplest option before a human gets the chance to sell a bigger package. The fix is to route emotional or urgent conversations to a human quickly rather than letting the bot handle them start to finish.
How do I know if my chatbot is hurting my business?
Compare your average transaction value and your top tier bookings from the three months before the chatbot went live against the most recent three months. A drop of more than around 10 percent usually means the bot is filtering customers toward the cheapest visible option.
Should small businesses avoid AI chatbots entirely?
No. They work well for repetitive factual questions like opening hours, delivery times, and order tracking. The mistake is letting them handle the first conversation for anything emotional, urgent, or with a wide price range, which is exactly where a human converts better.
What should a chatbot never be allowed to do on its own?
It should never quote a full price list before asking qualifying questions, and it should never be the only response to words like broken, urgent, emergency, or disappointed. Those conversations need a real person within minutes, not a form and a wait.
Related reading: Is ChatGPT Recommending Your Small Business? Here's How to Check (and Fix It) and AI Overviews Are Quietly Killing Your Website Traffic. Here's What To Build Instead.
This goes deeper into a conversational AI consultant in Israel.
