It is noticeable that there are very few great executions of chatbots out in the wild. So, we decided to dissect the art (and science!) of building a chatbot and in doing so, we put together this brief guide on building one that is effective.

  1. Personality — A chatbot with a personality can be 50% more convincing to users than one without a personality (according to Stanford University)
  2. However, always make it clear that it is a bot and not a human — this avoids customer frustration and negativity around the brand that may arise from the chatbot’s own limitations.
  3. Create a comprehensive conversational flow diagram — adopt both a traditional visual mapping approach, as well as bot prototyping tool like Bot-Society to visualise it before finalising it.
  4. Create the visual assets that correspond to the outcomes — e.g. images, video, animated gifs, emojis.
  5. Err on the side of simplicity for early versions and avoid trying to do natural language processing — i.e. include buttons for path selection
  6. Allow flow loops to fix incorrect email/data input.
  7. Hand-off to a real person or a site rather than allow for dead-ends.
  8. Use a tool like, HubSpot, TARS, FlowXO etc to build the bot.
  9. Decide where is the bot running — this will have an impact on the design of the bot, as different platforms will allow for varying levels of customisation. Options include FB Messenger, WhatsApp, WeChat, Telegram + iMessage and Web e.g. a chat panel or a full site experience.


Top 4 Most Popular Bot Design Articles:

  1. Five Tips to Create a Chatbot Brand Voice.
  2. “Alexa, are you ok?”: Test Automation for Alexa Skills.
  3. How to introduce Upsell workflow in the florist Chatbot.
  4. Three Steps to Designing Chatbot Conversations like a Professional.


Making a decent bot will require a lot of testing, asking for a second opinion and conducting testing with real end-users (unprompted).

Beware of the common mistakes:

  1. Too long messages: It’s very important to keep in mind that most of the conversations held with a chatbot happen on mobile. Keep your messages short and make sure people don’t need to scroll to read the message.
  2. Tone variation: Unless you’re trying to replicate some multiple personality disorder, try using the same tone throughout all your speech so users can feel they’re talking to the same character and engage more.
  3. GIF/image/emoji abuse: Emojis and funny GIFs can be effective but it’s important to ensure that the user isn’t visually attacked by overly colourful/distracting images.
  4. Question duplicity: One of the worst experiences a user can have is being asked for the same piece of information twice. Make sure this never happens in your flow and, if you can’t avoid it, confirm fields instead of asking for them again (e.g. “Your email is [user’s email], is this correct?”).
  5. Unnatural loops: Rich chatbots with back-and-forth freedom are great, but be careful with the loops you create and make sure they don’t subtract freedom instead of providing it.


Like any successful digital product, a successful bot will require an ongoing test and learn approach:

  1. Implement a measurement tool like Chatbase to allow for monitoring/ iteration.
  2. Try A/B testing of different paths to see what converts best.
  3. Remove less-trafficked paths e.g. if less than 10% of the audience venture down a specific conversation path — kill it.

Thinking about the above in some detail during the design and development phase will make for an infinitely better product and experience for users and will also go a long way to unearthing problems before users find them.

If you need some further advice or would like to explore a challenge with us, feel free to drop Johanna, our Commercial Partnerships Lead, a message at