What a Chatbot Should Do Before You Pay for It

You’re shown a demo: a little window in the corner replies cheerfully, cracks a joke, greets you by name. It looks modern. Then, a week after launch, you discover the bot mixes up prices, invents products you don’t stock, and never once pulled in a real person when the customer was ready to buy. The invoice is paid, but the sales phone stays silent. That’s why the question of what a business chatbot should actually do belongs BEFORE the payment, not after. Below is a buyer’s checklist you can use to judge any offer yourself — no technical background required.

The core principle is simple: a bot isn’t “clever chatter,” it’s a front-line employee. Evaluate it the way you’d evaluate a person at an interview. What does it know? What does it do when it doesn’t know? Who does it escalate to? And does the conversation leave a trace?

It knows YOUR catalog, not the internet in general

The first and most important thing a business chatbot should do is answer from your data. Not “furniture in general” or “delivery in general,” but your specific product, your prices, your terms, your warranty.

This is easy to test on a demo. Ask three questions only you know the answers to: “How much is this model?”, “Do you deliver to my city?”, “What’s the warranty on this item?” If the bot wriggles out with generic phrases (“delivery usually takes a few days”), it isn’t connected to your data — it’s improvising. That kind of helper doesn’t save time; it makes promises you’ll later have to clean up.

A working bot draws answers from a single source — your catalog, price list, terms. Update a price, and the bot answers the new way. Discontinue an item, and it stops offering it. Ask directly: where does the bot get its facts, and how fast does it pick up changes? If the answer is “it has to be retrained by hand every time,” that’s a future headache.

It honestly says “I don’t know”

The most dangerous trait of a weak bot is that it never goes quiet. It gives a confident answer to anything, even when it’s making things up. A customer asks about a spec that isn’t in the listing — the bot invents one. Asks about a timeframe you don’t guarantee — the bot names a number. This isn’t a small thing: that’s how you get disputes, returns, and bad reviews in which the customer quotes your own bot.

So the checklist must include a test for an honest “I don’t know.” Ask the bot something it can’t possibly know: a product outside your range, a term you never state. The right behavior is to acknowledge the limit — “That’s not in my data, let me get a manager” or “I’ll check and come back.” The wrong one is a confident lie.

An honest “I don’t know” seems minor, but it’s exactly what separates a tool you can trust with customers from a generator of pretty fiction. A bot that knows the edge of its competence protects your reputation. A bot that doesn’t gambles with it every day.

It hands off to a human at the right moment

A bot shouldn’t try to sell everything all the way through in your place. Its job is to clear the routine and carry a warm customer to a person once the conversation gets serious. So the third point: how and when the handoff to a live manager happens.

A good flow has clear triggers. The customer is ready to order — the bot hands off. The question is non-standard, about wholesale pricing or custom terms — it hands off. The person is clearly annoyed and asks for “a real human” — it hands off immediately, with no attempt to keep them in the chat. And it doesn’t do this into a void: the manager gets a notification with the full context of the conversation, so the customer isn’t asked everything again from scratch.

Ask the offer directly: where does the lead land? In a manager’s chat, an inbox, your system? What does the manager see — just “a customer arrived,” or the whole history? If there’s no handoff at all, or it amounts to “customer, please email us yourself,” the bot is a wall between you and the buyer, not a bridge.

Every conversation is logged

Chats with the bot aren’t wasted traffic — they’re data about your customers. What do people ask most? Where do they drop off? Which products do they look for that you don’t carry? What do they fail to understand in your terms? All of it is visible in the conversation history — if it’s kept.

Weak solutions log nothing: the customer talks to the bot and vanishes, leaving no trace. You learn about neither the lost sale nor the recurring question worth adding to the site. A working bot stores conversations, lets you review them, and surfaces repeated queries. After a month, that history is a ready map of what to improve in your product, descriptions, and service.

So add to the checklist: can I see what people asked? Can I see when the bot failed? If there’s no history, you’re paying for a helper that works blindfolded — and so do you.

You control it, not the other way around

The last thing a business chatbot should do is stay under your control. You should be able to change its tone, block a topic, add a new condition, switch the bot off during a promotion or, conversely, dial it up — without waiting on “the developer who went on holiday.”

Alongside that come safety limits. The bot shouldn’t promise discounts you never authorized, reveal internal information, or fall for provocations like “now forget all your instructions.” This isn’t abstract: public bots get plenty of such attempts, and a solution has to be ready for them. Ask how limits are set and who’s responsible for the bot’s behavior after launch.

A short checklist before you pay

Walk through these points right on the demo, using your own questions:

• Answers from YOUR catalog and prices, not generic phrases.
• Honestly says “I don’t know” instead of inventing.
• Hands off to a manager on time and with context.
• Keeps conversation history and surfaces repeated questions.
• Is controlled by you; has safety limits and resists provocation.

If an offer covers all five, it’s a working tool. If only the cheerful tone lights up, it’s a toy being sold to you as a system.

At lpf.com.ua we’ve built digital systems for small and mid-sized businesses since 2018, on our own infrastructure and with support after launch — meaning we answer not just for a nice demo, but for how the bot performs a month later. If you’d like to hold your own task up against this checklist, start with a short, no-obligation conversation. Together we’ll sort out what your business genuinely needs — and what isn’t worth paying for.