Designing a Chatbot Flow That Does Not Trap the User

How to design a chatbot flow: step count, option design, human handover, drop-off points and improving with flow analytics.

Building a chatbot flow takes an afternoon. Building a good one starts after you build it: by seeing where people give up and fixing that step.

Design rules

  1. At most four options per step. By the fifth option the user has forgotten the first; the same limit as an IVR menu applies here.
  2. No more than three levels deep. Drop-off rises sharply at the fourth. The aim is to reach either an answer or a human within three taps.
  3. Write options in the user’s language. Not “order status enquiry” but “Where is my order?” — people search for their problem, not your internal names.
  4. Put an exit in every step. “I want to speak to a person” must be reachable everywhere. This is the single most important item in the flow.
  5. Say it is a bot. Users work it out anyway; hiding it only costs trust.
  6. Handle the unexpected answer. When someone types free text where an option was expected, hand over to the assistant or an agent instead of saying “I did not understand”.

The three most common mistakes

Dead ends

A step where the user does not know what to type and has no way out. The result: they abandon the channel and never write again.

Unnecessary verification

Order number, ID number, date of birth… Before asking for information, ask whether the step is genuinely needed or merely habit.

Ignoring opening hours

The same flow running in and out of hours. During hours the goal is a fast handover; outside them, to resolve as much as possible.

Flow analytics: where the real work is

Four numbers to watch once the flow is live:

  • Starts: how many people entered the flow? A low number means a visibility problem rather than a flow problem.
  • Drop-off per step: how many were lost at each step? The step with the biggest fall is the first to fix.
  • Completion rate: the percentage who reached the end.
  • Handover rate: how many went to an agent? This should be neither very high nor near zero — high means the flow is not working, near zero means users cannot find the exit.

Read all four together. A high completion rate alone is not good news: the user may have finished the flow and still not solved their problem.

The improvement loop

  1. Find the step with the highest drop-off.
  2. Look at what users typed there — free-text answers reveal the missing option.
  3. Make one change (add an option, simplify the wording, or remove the step).
  4. Measure for a week. Two changes at once and you cannot tell which worked.
  5. Move to the next drop-off point.

Flow or assistant?

One question decides it: can the steps be written down in advance? Booking an appointment, checking an order and filling a form can — use a flow. “Does your product work in this scenario?” cannot — hand it to the assistant. For most businesses the right answer is hybrid: structured tasks in flows, open questions with the assistant, complex issues with an agent.

Frequently asked questions

How many flows should I build?

A few well-measured flows beat many forgotten ones. Building flows for your three most frequent requests and measuring them is the right start.

Should the flow be identical on every channel?

The logic can be, the design is not. WhatsApp brings the 24-hour rule, web chat brings session length, and button support differs by channel.

What if the user changes topic mid-flow?

A good flow handles it: detect the new intent and switch flows, or hand over to the assistant. Trying to keep the user in the old flow is the fastest route to abandonment.

Next step: to see the flow designer and analytics screens, request a demo.