Chatbots for a 30-Person Business: What Works, What Annoys

A chatbot is one of the most tempting purchases a small business makes, because the pitch is cheap: answer customers automatically, look modern, free up the team. It is also one of the most reliably disappointing, and the reason is structural, not technical. At a company of thirty people, a chatbot either resolves the top handful of customer intents end to end, or it is an expensive wall between your customers and the two people who actually know the answers.

What annoys

The deflection loop. Customer asks a question, bot offers three canned categories, none fit, customer types the real question, bot repeats the categories, customer demands a human, bot asks them to fill a form. Every step of that loop is a customer learning that your company is harder to reach than it used to be. The bot did not deflect the work; it moved the frustration earlier and made it personal.

The bot that answers about everything and resolves nothing. Broad knowledge, zero access. It can explain your return policy but cannot look up the customer's order. Explaining policy is what your FAQ page is for, for free. A bot without system access is a FAQ with extra steps.

The bot nobody re-trains. The questions customers ask drift with every product change, and the bot's answers do not, because maintaining it is nobody's job. Six months in, it confidently gives answers that were true in March.

What works

The version that earns its keep at this size has three properties.

It resolves, not deflects. Order status, invoice copy, return initiation, appointment rescheduling, password resets: the top five intents, each completed end to end inside the chat, because the bot can actually call the systems that hold the answers. Five resolved intents beat fifty acknowledged ones.

It hands off with context. When it cannot resolve, it reaches a human with the transcript and the customer's record already attached, and it does that fast. The handoff is the feature. A customer who gets a bot attempt and then a human with full context had a fine experience; one who repeats their story to the human had a bad one, and the bot is why.

It drafts instead of sending, at first. For email and support queues, the highest-value deployment is internal: the bot drafts the reply, a human reviews and sends. You get most of the time savings with none of the brand risk, and the drafts you reject teach you what the customer-facing version would need before you ship it.

The honest test before you buy

Write down your top ten customer questions. For each one, ask: could a bot resolve this end to end with access to our systems, today? If you have five or more yeses, you have a chatbot business case. If you have two, you have a FAQ page and a phone number, and buying a bot will make your customers like you less.

And budget for the boring part: someone reviews the transcripts weekly, fixes the misses, and prunes the dead intents. A chatbot is not a product you install. It is a small employee you manage, and the companies that skip the managing are the ones writing the cautionary version of this post.