Chatbot: what it is and when to use it (without frustrating the user)

A chatbot is a system that converses with users through text or voice to answer questions, guide processes, or automate tasks. It can be very simple (rule-based, using decision trees) or advanced (based on AI language models).

Implemented well, a chatbot reduces repetitive support load and helps the user get to what they need faster. Implemented poorly, it's the screen that stands between the user and the answer—and the number-one reason people abandon a brand with the phrase "I wish I could talk to a person."

Types of chatbot

Rule-based chatbot

Follows a predefined decision tree. The user chooses options, the bot responds based on the path. Predictable, cheap to implement, limited: it only knows what you programmed.

AI / NLP chatbot

Understands natural language, can answer questions you didn't anticipate, maintains context. More capable, more expensive, requires training or a connection to a model. The vast majority of useful chatbots today combine AI with rules for critical tasks.

Hybrid chatbot with human escalation

The most sensible option for real commercial cases. The bot resolves what it can and routes to a human when it gets stuck or when the user asks. The quality of the handoff defines the experience.

Use cases where they work

• First-level support FAQs. Repetitive questions with a clear answer: order status, password, hours, return policy.
• Capturing qualified leads. Conversational forms with less friction than a traditional form.
• Appointment scheduling. Available calendars, proposals, confirmations.
• Product recommendation. "You're looking for X with budget Y for Z" → filtered recommendation.
• Inquiry triage. Classify the inquiry and route it to the right team or channel before involving a human.
• Onboarding of complex products: guide the new user step by step.

When NOT to use a chatbot

• When the inquiry is emotional (a serious complaint, a sensitive problem). An upset user gets more angry with a bot; with a person they calm down.
• When the answer requires judgment. Atypical cases, decisions with nuance, custom proposals.
• When the cost of an error is high (legal, financial, medical). The bot can make a mistake in a way a human never would.
• When you don't have human capacity behind it. A chatbot without a functional human handoff is just an FAQ with a poor interface.

The most expensive mistake: a bot that blocks with no way out

The most damaging pattern for a brand: a chatbot that insists on resolving what it can't resolve, with no visible option to talk to a person. The user loses 5–10 minutes clicking useless options, leaves, and forms a negative association.

Practical rule: there should always be a visible "talk to a person" button within fewer than 3 interactions with the bot. If the bot can resolve it, it tries. If not, it routes cleanly.

Best practices

• Start with real inquiries. Audit 200–500 support conversations. What repeats is what the bot should resolve.
• A tone consistent with the brand. A bot that's too formal for a casual brand breaks the experience; one that's too jokey for a serious brand does too.
• Don't pretend to be human. Saying openly "I'm an assistant" reduces frustration when it makes a mistake.
• Measure resolution rate, not usage. If 1,000 users talk to the bot but only 200 resolve their issue, the bot is hurting, not helping.
• Constant iteration. What the bot didn't resolve this week is input for improving it next week.
• Handoff with context. When it routes to a human, the agent receives what was already discussed. Without this, the user has to repeat everything.

Common implementation mistakes

• Launching without measuring real conversations. You assume what people ask and get it right 30% of the time.
• No periodic review. What worked six months ago doesn't necessarily work today.
• Too many menu options. If the first screen has 12 buttons, you've failed at the design.
• A bot that "sells" too soon. The user who comes for support and gets an upsell leaves.
• No satisfaction metrics. Resolution rate isn't satisfaction. Ask directly.

Chatbots in creative operations

For an agency or creative team, a well-designed chatbot is a project of conversational content production: each flow is a script, each response a piece of copy, each tree decision an editorial decision. Producing it without a system that connects brief, writing, approval, and maintenance creates inconsistent flows that age badly.

Treated like any other creative piece, with a calendar, briefs, and approvals, the chatbot maintains quality and evolves. Treated as a technical "implement and forget" project, it deteriorates fast.

At Polimake, conversational flows are designed in Studio (briefs and approvals), copy versions are managed in Studio, and the associated visual assets (avatars, bot illustrations) live in Media.

Related concepts

• SaaS / software as a service
• Workflow
• Lead nurturing
• Conversion funnel
• Content production at scale

---

This piece is part of the Polimake glossary and the cluster on creative operations. If you manage customer service or automation at an agency or in-house team, also read creative approval workflows.