Are AI chatbots worth it for a small business?

Starly is one of these tools, and this page still starts with the shops that should not buy one yet. For a recognizable group the honest answer is not yet, and it is named here before the winners are.

By · · 11 min read

Pages asking whether an AI chatbot is worth it for a small business almost always arrive at yes, and almost always because whoever wrote one arrived with something to sell. Starly is one of these tools, and the answer is still no for a recognizable group of shops. That group is worth naming before the winners are, because the decision turns out to be mostly not about vendors at all. Before you compare a single feature, three facts about your own shop decide the outcome: whether your policies exist in writing anywhere, whether your product data and stock are current, and how many questions you get in a week. A shop with all three gets value out of a plain tool. A shop with none of them gets a wrong answer said confidently, in public, whichever tool it bought.

The case for, kept concrete

It answers while you are shut, which for most shops is most of the week. What it should say during those hours is a configuration job with a page of its own; what belongs in a buy-or-wait decision is only the count. Open your analytics and total the sessions that land when your inbox is effectively closed. A question that sits from Friday evening until Tuesday morning is usually not a sale by Tuesday morning. It also answers in the customer's own language without you hiring for it, which starts to matter the first week a tenth of your inbound arrives in a language nobody on your team reads. And it does not tire of the fifteenth shipping question of the day. The answer at 5pm is word for word the answer at 9am, which is more than most of us manage.

A question can also end in a cart rather than in a sentence. Someone asks which size fits a 40cm shelf and the reply names the variant and puts it in front of them, ready to buy, while they are still on the page. How that works, and how it degrades off the storefront, is a Shopify question with its own page. What matters to the decision you are making here is only the gap between two outcomes that look equally correct in a transcript: one leaves the customer to go and find the product themselves, and the other does not. Correct-and-finished beats correct, and it is the only thing on the list of benefits that a well-written FAQ page cannot also do.

And it writes things down. Every conversation is a transcript, and you can open the trace behind an answer to see which card or product it used. Inside a month you have a list of the questions your product pages fail to answer, the sizes people keep asking for that you do not stock, and the objections you have been losing to without ever hearing about them. Some owners get more out of that list than out of the answers. Demand you never hear about is the expensive kind, because you cannot decide not to act on it.

The case against, given the same weight

A confidently wrong answer costs more than no answer, and the two failures are not on the same scale. Silence makes somebody email you, which is a cost you can see and price. A wrong delivery date makes somebody act, and the bill for that arrives later, through a channel that never mentions chat at all; the page on why customers hate chatbots follows that bill to where it lands. What the asymmetry means for a buying decision is this: the upside of chat is bounded by your traffic, and the downside is bounded by nothing you control. That is the whole argument for keeping the assistant inside your own product data and your own written rules rather than letting it draw on a model's general sense of how shops like yours usually work, and it is why no tool rescues a shop that has neither.

Which leads to the work. An assistant needs your policies in writing: what shipping costs, how long it takes, the order cutoff time, what happens when something arrives damaged, how long the return window runs, who pays for the return label. If you have carried all of that in your head and decided case by case, you now have to settle questions you have been deferring for two years. Nobody can do that for you, because you are the only person who knows the answers. Set aside an afternoon and a document before you set aside a budget.

A stale catalog produces stale answers in public, at volume, faster than you would catch them by hand. If the feed says twelve in stock and the shelf is empty, twelve is what customers hear, twenty times before anyone checks. If a variant was discontinued in January and is still in the export, it will be offered in August. Discount codes are the one place with a backstop, because the assistant reads the store's live discounts and quotes only codes that actually exist. Nothing you typed by hand has that protection.

Some customers want a person and resent the detour, and no amount of good writing changes that. The answer is not a better argument, it is a one-step exit: a handoff that alerts your team by email or phone the moment someone asks for a human, with the conversation attached so they are not made to repeat themselves. A shop that hides that exit earns every complaint it gets.

And it is one more thing to check. Unchecked, it drifts, because the shop keeps moving and the written answers do not. You added forty products, you shortened the return window to fourteen days, your supplier's lead time went from two days to five. The assistant keeps saying what the cards say until somebody changes the cards, and nothing about that failure is loud.

A readiness checklist you can answer today

Four questions, answered honestly, before you look at a single vendor.

Three yeses and a maybe is fine, and normal. Two noes means the money is better spent somewhere else this quarter, and the two noes are usually the same two: nothing written down, nobody watching.

The three shops that get their money back fastest

First, shops losing the traffic that arrives while they are closed. If a real share of your sessions land when nobody is reading anything, you are already paying to bring those visitors in and then not answering them, which is the most expensive way to run a shop. The assistant is not adding a channel here. It is covering hours you have already bought and are currently wasting. It works in the other direction too: if you would rather it went quiet when you are shut, hours and an away message will do that, and doing it properly is a short project in its own right.

Second, shops answering the same five questions all day. Count them for one week rather than guessing, and note how few distinct ones there are; most shops find the same handful, and it is the same handful the handoff page says should never reach a person at all. If you can recite your top five without checking, those are also the five you can write down in an afternoon, and after that you stop typing them. This is the fastest payback of the three, because it converts a recurring cost into a one-time one, and the arithmetic is the same whether you are answering ten of those a week or two hundred.

Third, shops selling into a language they do not staff. Hiring for a language to serve eight conversations a week is not a real option; answering in it is. This is the case where owners are most often surprised by the volume they had been quietly losing, because unanswered messages in a language you cannot read look like noise in the inbox rather than demand.

The wait list, said without hedging

Four situations where the answer is no, or at least not this quarter.

That last one is worth being blunt about, because it is the most common mismatch and the easiest to buy your way into by accident. Nothing in the chat window touches it: the assistant will hand those messages to you, which is exactly where they already are. The remedy is in the shop rather than the chat tool, and on Shopify most of it is machinery you are already paying for and probably have half switched off — the Shopify page names the specific settings. Buy a chat window when the questions are about products and policies. Do not buy one hoping it will answer 'where is my parcel'.

What to do instead if you are on the wait list

Write the shipping and returns page you keep meaning to write. Real numbers: cost, days, cutoff time, the return window, who pays return postage, what happens with a damaged item, and what you do about a package marked delivered that never arrived. It takes an afternoon and it will answer more questions than any chat window, because a good number of people read it before they ask.

Then fix the product data. Materials, dimensions, weight, what is in the box, and whether the stock count is true. Every ambiguous product page generates questions, and right now you are paying for those in your own evenings. Then put a real email address somewhere findable and answer it, because the shops that complain loudest about chatbots are usually the ones whose only alternative was a form that went nowhere.

None of that is a consolation prize. All three are exactly what an assistant reads from, so each one makes it answer better later anyway. If you do them and your question volume grows, the groundwork is already done. If you do them and volume stays at five a week, you have kept $300 a year and improved the shop. There is no version of this where that afternoon is wasted.

The attention cost, named honestly

The shops that get nothing out of this are almost never the ones that picked the wrong vendor. They are the ones that stopped looking at it in week two. Plan on a short standing slot every week, and on the loop it exists to run: find the two answers that were vague or wrong, fix the card behind each one. What to read in that slot, how many conversations, and how to tell a good month from a bad one afterwards is a discipline with its own page and its own time budget. All that belongs in a buy-or-wait decision is that the slot is not optional. Skip it for a quarter and the cards are describing a shop that no longer exists, in public, to people who were about to buy.

Fifteen minutes a week is thirteen hours a year. If you cannot commit that, the answer to the question in the title is no for you, whatever your volume and your catalog look like. This is the wrong tool when nobody owns it. It is the wrong tool when your questions are about individual orders rather than products. It is the wrong tool when there is nothing written down for it to answer from, and when five people a week ask you anything at all. None of those are fixed by picking a different vendor, and a vendor who tells you otherwise is selling you the version of this that ends in a refund.

Worth it compared with what

The question in the title hides a comparison, and the comparison decides the answer more than any feature does. If the alternative is nobody replying at all, almost any working assistant wins, because the baseline is a lost conversation. If the alternative is you, in the evening, then the tool has to beat you, and it will not beat you on accuracy or judgement. It wins on hours and on volume, and only if you actually stop doing the typing.

That last condition is where most of the value leaks out. Owners install the assistant, keep answering everything themselves out of nerves, and conclude two months later that it saved them nothing. It did not, because they never allowed it to. And if the alternative is a part-time person you already employ, run the sum with their hours in it rather than against zero: removing an hour a week from somebody paid for forty has saved you nothing unless that hour goes somewhere. Decide in advance where it goes. The shops that get the most out of this are usually the ones where the answer was obvious before they bought anything — packing orders, shooting product photos, or leaving at six.

So, is it worth it?

With the first week free, the test is cheap enough to run rather than argue about, but run it after the checklist and not instead of it. If it removes an hour of repeat typing a month, it has paid for itself; whether it does is a question about your volume, not about the tool. For the three profiles above, that arithmetic is not close. For everyone else the honest answer is not yet, which is a different answer from no. The wait list is short, every item on it is fixable by you, and none of the fixes require buying anything.

Common questions

Are AI chatbots worth it for a small business?

It depends on three things you can check today, not on the vendor. If you get more than about 10 customer questions a week, your shipping and returns policies exist in writing somewhere, and your product and stock data are current, an AI chat assistant is worth testing. If you get around five questions a week, have not launched, sell a single product, or your policies only exist in your head, wait. Write the policies and fix the product data first; both make the assistant answer better later anyway.

How many customer questions a week do I need before a chatbot makes sense?

Roughly 10 a week is the point where it starts to be worth the setup and the ongoing attention. Under 10, answer them yourself and spend the money on stock. Between 10 and 50 the return is time saved plus the evening questions you would otherwise answer on Tuesday. Over 50 it is usually the cheapest line on your list. Volume is not the only test, though: you also need policies written down and someone willing to read transcripts for fifteen minutes a week.

Can an AI chatbot tell my customers where their order is?

Starly does not do customer-facing order-status lookup, so if most of your inbound is 'where is my order', a chat assistant will not touch the bulk of your volume. Those messages will be handed straight back to you. The fix for that queue sits in your shop's own notification settings rather than in any chat tool, and on Shopify it is largely already built and switched half on. An AI assistant is for questions about products, sizing, materials, shipping rules, returns and stock, not for tracking individual parcels.

What do I need ready before setting up an AI chat assistant?

Three things. Your shipping and returns policy in writing with real numbers: cost, days, cutoff time, return window, who pays return postage, and what happens with damaged or missing items. A current product catalog with correct materials, dimensions, variants and stock, because the assistant reads it out to customers. And one named person who will spend fifteen minutes a week reading conversation transcripts and correcting what was wrong. Without the third, the first two go stale and the answers drift with them.

Will customers be annoyed by an AI chatbot on my shop?

Some will, and that group does not shrink because your writing is good. What matters is that they can reach a person in one step instead of being made to argue with a chat window first. Configure the handoff so your team is alerted by email or phone, keep the assistant honest about what it does not know, and answer from your own catalog and policies rather than guesses. Most of the resentment people report comes from bots that hide the human or answer confidently and wrongly, and both of those are settings, not inevitabilities.