5 tips for researching chatbots for your business

by Tom Stewart on 9 November 2017

Chatbots enable brands to help, inform and engage with customers. They present businesses with the unique opportunity to create a personality for a digital, initiate dialogue and help customers to meet our everyday goals. For more info about the fundamentals of conversational UIs, check out our previous article.

Thinking back to 50 years ago, stuff just got done. How? By talking to people! In a conversational interface, we have the power to harness the emotions of having a real-life conversation, but in a brilliantly easy text-based fashion.

A chatbot is just an interface designed to replicate that experience of talking to a person.

As in any good User Centred Design process, the best designs are developed through research. So that's what we did. And based on our experience, here are our 5 tips for researching chatbots:

1. Focus relentlessly on what problem your bot solves

A bot needs to serve a purpose unless you're in the market for gimmicks. Remember, being novel doesn't mean it doesn't serve a purpose - our favourite chatbot in this space was Robbot helping fans engage with Robbie Williams, the music artist. Using Robbot as a marketing channel, they served up new music content, tour dates, and jokes - all based on direct research with Robbie Williams himself.


Measuring the success of bots can be a challenge, especially for more complex conversations. Setting out KPIs and metrics is as important as any other interface. Set out the markers for your research, things to look out for, and you can use the data feeds to review every interaction if you need to.

Whether it's generating tour ticket sales, providing support with mental health issues, or helping people navigate around London, it's easy to get distracted and end up designing things that don't solve a real user problem...

Focus every interaction on driving your chatbot conversation towards your user's goal.

2. Expose your bot to stakeholders early

Your job as the designer is to help remind stakeholders exactly what you've created, which user problems you're trying to solve, how you're planning to solve them, and what that means for their business.

Managing expectations of what your bot can and can't do are very important, especially if they come to observe user testing - which we'd highly recommend.

Not all stakeholders will understand the ins and outs of chatbots like you might, so here's a coaching exercise for you. Messaging interfaces tend to make people smile and have a chuckle as they can play with personality, emojis and GIFs, but do not lose sight of solving the user's issues, or humour can quickly be replaced with anger.

3. Let people test your bot on their own devices

The joy of chatbots, conversational interfaces, and artificial intelligence is that we can use this technology to engage with people where they are already hanging out. If participants are used to using WhatsApp, Facebook Messenger, or iMessage, then let's test the bot there.

It minimises barriers for users & helps them get straight into a conversation without having to learn a new interface.

4. Make sure you nail your error messages

Broken record chatbotParticularly when you're in the proof-of-concept stage, there are going to be a lot of holes in your conversation design. To make sure there's no dead ends in your conversation during research, invest some time in curating your error messages for when things don't go to plan. You'll then be able to learn and iterate on your error wording as well as the main content.

Broadly speaking, there are two occasions when you'll need this;

  1. We didn't understand what the user said
  2. We know what they said, but can't answer their query

Your error message should be honest, open, and friendly. Provide ways someone can get their answer - give us a call, try online, try typing something else or provide more structured question options.

5. Remember chat isn't always right

There are times when using a conversational or messaging interface simply isn't the best way to do the job that the user needs. And that's ok.

Key areas we found that you should be ready for:

  • When your bot can't answer their query
  • Topics that are too serious to be dealt with in messaging apps (e.g. dealing with complex topics where you really need human expertise & emotional support)
  • Complex data entry (e.g. financial planning with lots of numbers & data points)
  • Non-linear content (e.g. I'm lonely, doesn't have a distinctive transactional flow)
  • Bots aren't always delivered on mobile, so remember the wider context

TFL Travel Bot

Some ways to get around issues like this:

  • Always provide a help route to speak to a human
  • Build in feedback loops & direct learning from users
  • Use menus where you can to provide consistent access to common tasks.

In summary

Researching our audience's needs and problems is key to ensuring your chatbot can keep customers happy. Don't get bogged down in the details too early and ensure these challenges are shared amongst stakeholders. Your role in guiding the audience and company through this process will be easier the more transparent you are about the build. Utilising humour, emoji's and gifs will be great tools to develop your brand's personality but customers need to know that they can trust the bot and the brand first and foremost.

Do the research first and the chat will flow.

Like what you see?

How exciting, let's get started

Thank you for your comment. It has been submitted for approval.

Comment Again?

Leave a comment

Please enter your name.
Sorry, this email address is not valid.
Please enter your comment.

Course basket x


Price per place

Add another courseCheckout

Pay now by credit card or later by invoice (invoice payments only possible if all courses are 3+ weeks in advance)