The answer depends on three conditions, not on the product. Here is the scoring sheet, the honest no, and the cheaper alternatives that win more often than anyone admits.
A two-chair barbershop paid for setup and a monthly subscription on a chatbot that handled nine conversations in its first month — six of which were some version of "are you open today." The owner could have answered all nine from his chair between clients. Across town that same quarter, a remodeling company with roughly 340 inquiries a month had two people doing nothing but answering the same twelve questions, and inquiries were still sitting unanswered overnight.
Same product, opposite verdicts. Whether a chatbot is worth it is not a property of the chatbot. It is a property of your question volume, your response-time economics, and whether a human owns the thing after launch. Those are checkable before you spend anything.
The three conditions that make it clearly worth it
When all three are true, the decision usually makes itself.
- High repeated-question volume. The same small set of questions — hours, pricing, availability, service area, booking — accounts for most of your inbound. Volume matters because it is the denominator under your maintenance cost: the same weekly upkeep that is absurd at nine questions a month is trivial at three hundred.
- Speed actually wins the business. In appointment businesses, rentals, and emergency service work, the first responder frequently books the job. If your leads contact three competitors in one sitting, response time is not a nicety — it is the product.
- Someone internal owns it. A named person who updates the facts file when hours or prices change and reads transcripts weekly. Not "we'll figure out who after launch." That sentence is where chatbots go to die.
Missing one condition, the math gets shaky and you should fix that condition first. Missing two, you are buying an expensive way to disappoint people.
The three conditions that make it a bad idea
- Low volume. Under roughly twenty to thirty real questions a month, saved replies and a disciplined phone habit beat any bot on cost and on quality. That threshold is a field heuristic, not a law — but the order of magnitude is right.
- No maintainer. A facts file nobody updates is stale within weeks, and an agent with stale facts states wrong hours and wrong prices with total confidence. An unmaintained chatbot is worse than no chatbot, because silence does not mislead anyone.
- The business sells on personal relationships. High-ticket, consultative, trust-based purchases — the kind where the client is buying you — lose something real the moment a script fronts the first conversation. Automation here does not save labor; it spends trust.
Score your own situation
Fill in the last column honestly. Scores of 1 mean "sort of."
| Question | 0 points | 1 point | 2 points | Your score |
|---|---|---|---|---|
| How many real questions per month? | under 20 | 20–100 | over 100 | |
| Are they repetitive? | all unique | some repeat | mostly the same ten | |
| Does a fast reply win the job? | slow is fine | sometimes | first responder usually books | |
| Who maintains the facts file? | no one | "we'll figure it out" | a named person, hours blocked | |
| Do customers buy you personally? | no, they buy speed or price | mixed | yes, relationships sell |
Read the total this way: seven or more points, a pilot is worth building. Four to six, exhaust the cheaper alternatives below first — they may solve it outright. Three or fewer, or a zero on the maintainer row, the answer is not yet. A zero on the maintainer row overrides everything else, because it is the row that determines whether the bot is still telling the truth in month three.
[!NOTE]
Before you score anything, run the manual version for two weeks: log every question that comes in, mark which ones repeat, and time how long each answer takes you. That log produces the volume number the scoring sheet needs, and it costs two weeks instead of a subscription. Deciding without it is how the barbershop above happened.
The cheaper alternatives that often win
Most of what people buy a chatbot for can be bought for almost nothing. Try these first, in order.
- Saved replies and text shortcuts. Your five most common answers, stored as phone text shortcuts or inbox saved replies. Sending one takes a tap, and the answer is always right because a human sends it.
- A plain auto-responder with a route. "Got your message — for booking, call or text [number]. We reply within [window]." No AI, no improvisation, and it converts after-hours contacts into first-in-line morning callbacks.
- A FAQ page where customers already look. Your Google profile, your booking page, your link-in-bio. Half of "chatbot questions" are questions your public pages should already answer.
- A fixed answering schedule. Two daily alarms — clear the inbox at opening, clear it again an hour before close — with the missed-lead recovery process behind them. Answering in ninety minutes manually beats a bot answering in thirty seconds badly.
Notice that none of these require judgment calls in your absence, which is precisely the thing a chatbot cannot reliably do.
When the honest answer is "not yet"
"Not yet" is a real answer, not a failure. It is the correct answer when volume is growing but not there yet, when the owner is wearing every hat and maintenance would quietly never happen, or when hours and prices are changing weekly and the facts file would be fiction by Friday. A chatbot is easy to add later and hard to unwind publicly once it has embarrassed you. Write a revisit date — one quarter out, with the volume number you need to see — and move on without guilt.
Four mistakes people make when deciding
- Buying for the volume they want, not the volume they have. Projected volume is not in the denominator. Actual, counted volume is.
- Counting questions your pages already answer. If your FAQ and booking page resolve it today, a chatbot answering it again is not a saving — it is the same saving counted twice.
- Deciding by demo. The vendor chose the questions in the demo. Your customers did not.
- Treating it as all-or-nothing. The best deployments are usually narrow: a bot for after-hours and the five repeated questions, humans for everything else. Buying "a chatbot" instead of "a bot for the ten questions that repeat" is how scope creep starts.
What to do today
- [ ] Count last month's real questions and mark which repeat. That number is row one of the scoring sheet.
- [ ] Score your situation in the table above, out loud, with whoever will maintain the thing.
- [ ] Set up saved replies for your five most common answers today — before any purchase decision.
- [ ] If the verdict is "not yet," write the revisit date and the volume number that changes it.
A chatbot is not a strategy. It is a response to a measured volume of repeated questions, owned by a named human. When that measurement exists, the decision is obvious. Until it does, the cheapest correct answer is usually already sitting in your phone.
This guide lives at AgentFast, where you'll also find a free audit tool that checks your own agent's instructions against these failure modes. Posted here with a canonical URL pointing at the original.
