Our Blog



Making your AI business case too big to fail (Part 2)

In part one we talked about your bot preparation; where it will live, what its capabilities will be and choosing a platform to build it on – now let’s design it. The main thing you’re designing is the conversation. These fundamentals will give your bot a good start in life but remember, to grow it needs to constantly learn and evolve.

I visualise a time when we will be to robots what dogs are to humans, and I’m rooting for the machines.”
Claude Shannon

bots, bots everywherePhoto by 

In part one we talked about your bot preparation; where it will live, what its capabilities will be and choosing a platform to build it on – now let’s design it. The main thing you’re designing is the conversation, these fundamentals will give your bot a good start in life but remember, to grow it needs to constantly learn and evolve.

Designing your bot


Although Google is trying pretty hard to make their AI completely vanilla, most people tend to be anthropomorphic in nature so it’s important to give your bot a personality. If you don’t your users will do it for you and you’ll lose control over the bot’s brand perception.

Given bots are largely text and voice based the words used ARE its personality so this is an important part of the design. The way a bot talks should be a direct reflection of the brand. For instance, even though they might perform similar tasks do you think users would have the same conversation with an Apple vs a Microsoft bot?

Do you want your bot to be perceived as:

  • Male or female
  • Young or old
  • Casual and friendly vs formal and polite
  • Openly a bot or masquerading as a person
  • Having multiple personalities

The last one, creating a bot with multiple personalities, is a great way to tailor a bot experience for different people but, be careful to give the bot a core set of on-brand values to base its interactions. The ability of AI, like concierge bots, to switch personas to match culture and language of the audience is powerful, but if not given clear brand guidelines the bot can turn into a schizophrenic mess.

As a rule of thumb if this is your first bot then stick to a single personality. When you’ve mastered the basics gradually enhance the bot’s personality based on user feedback and data.

Style, cadence, speed and frequency

In WWII this was called a German radio operator’s ‘Fist’, it identified them to their counterpart in England. Your bot’s Fist is another way to enhance its personality and engage in a more meaningful way with its human counterpart.

When we talk to each other natural pauses and our individual style make conversations flow. Unfortunately, AI has not developed to a level where a bot will instinctively pause as if pondering a question. Bots still need to be programmed to pause, provide multiple answers to a single question without being asked, or capitalise particular words for emphasis.

The bot’s Fist needs to blend into your existing brand personality. In your design consider the following:

  • Short, curt responses or lengthier sentences
  • Immediate responses or timely pauses
  • Breaking up responses into multiple messages or multi-line mega-replies

When deciding on the above think about what task your bot is helping with and the people it’s talking to. For instance, if your bot tells commuters what time their train leaves then any unnecessary pause will worsen the user experience (or worse cause them to miss their train). If the bot is helping someone find a Mexican restaurant in the city that does great Tacos, then multiple suggestions and graceful pauses might better mimic the user’s decision making thought processes. 

‘Context is king’

Bots can seamlessly and simultaneously engage users across multiple channels. Essentially, you’re creating a single intelligence and allowing it to interact with your audience through multiple touch points (that old omni-channel experience thing again).

This presents a design challenge. Take our example of a commuter trying to find out their train departure time. If they are on their mobile and running late the bot can tell them departure time, platform number, and use the phone’s native location services to find the user and direct them to the platform.

What if the same bot was being used at a kiosk display in the ticket hall by a commuter who wasn’t running late? The bot’s capabilities remain the same but text can now be replaced with a digital human-like face. What if the commuter is from another country? Again, the bot’s capabilities remain the same but it might now change how it responds to allow for cultural differences.

To answer these design questions use your journey map and your design personas to make decisions and set the bot’s priorities. On day one your bot just needs to help your commuter catch their train, that’s it. Real user feedback and behavioural data will steer where the bot needs to be improved and what additional capabilities it needs.

Maintain engagement

People become disengaged and bored when repeatedly exposed to the same stimulus. Likewise we are stimulated by unpredictability – one of the reasons gambling is so popular.

When people can’t predict what is going to happen they pay closer attention. This is hardwired into all of us and it is this aspect of unpredictability that needs to be programmed into your bot.

For instance, the number of potential questions your bot could face is extensive and there is only so far the AI can go. Rather than the endless “I’m sorry, I didn’t quite catch that” provided by the likes of Siri, think about changing the response after the third request and directing the user back to an interaction the bot is capable of helping with. You could even have some fun with this and include some smart responses to all those hecklers out there.

Booking.com have a great solution to this. Since there are a lot of standard questions typically asked about a property – how many rooms, availability, location and day rates – the bot is given the ability to answer these. For all other questions the bot openly directs the user to the property owner. This has worked so well it has reduced the number of enquiries property owners have to field by 80% whilst maintaining booking rates.

Remember Remember Remember

The more history your bot has about previous interactions with each person, the more value it will be able to provide with each engagement. This is the natural way we build relationships between ourselves and bots are no different. Strong relationships create trust and trust leads to loyalty. If your customers trust your bot they will return time and time again.

Remembering key things about a previous engagement may also shorten the time it takes the bot to help achieve the user’s goal which is good for business. If you ordered a coffee your bot should remember your order next time, something good coffee shops do as a sign of good customer service. However, the coffee bot also knows what time you take your coffee, can send you a reminder, lets the café know when your 5 minutes away and adds the loyalty points post purchase. Here the bot far outperforms the barista in customer service.

Keywords are the new buttons

If the message thread is the bot’s interaction design then words are its buttons. Buttons should be:

  • Easy to find
  • Clearly labelled
  • Memorable

Wherever possible use standardised labelling (don’t use ‘store’ if ‘save’ is the norm) and always give the user an option to go ‘Back’. Your bot should use labels to direct conversation, present multiple options at once, and provide a way out if the user is at a dead end. eCommerce is an area where keywords can smooth the purchase experience to completion.

Understand intent

If you don’t know what someone wants how can you help them achieve their goal? Everyone (and I mean everyone) has a goal when they engage online. Make sure the bot is asking the right questions to elicit the appropriate response from users.

If necessary, prioritise and summarise information based on known user preferences prior to delivering an answer. This can work especially well where multiple options are available such as eating out or B2C retail. Take an online clothing retailer. The bot could ask if the user was looking for something suitable for winter or summer, or which colour they are trying to coordinate with. This narrows down options and allows the bot to provide more relevant answers.

Never release a final anything

One last thing, and possibly the most important takeaway.

Once launched your bot is now tapping directly into the needs of your customers and gathering vast amounts of invaluable data. Your bot has the ability to continuously learn from the data it gathers and improve with each engagement. But you need to help it on its way.

Continuously review conversations and improve them and add capabilities based on insights that emerge after people use it. To be a real bot master, feed insights into Marketing, Product Development and Operations.

Now you’re ready to release the hounds - sorry I mean developers. Giving them the background into why the bot exists, what purpose it serves and how it will help humanity will allow them to apply their technical know-how and enable the experience you want. Don’t stop talking to them throughout build and have fun.

This blog is part of the #datareimagine series. For more experts' insights, clients' experiences and to download our datasheets, click the banner.

For more experts' insights, clients' experience and to download our datasheets, click the banner #datareimagine

Posted by: Adam Ford, Practice Lead, Digital & Experience Design | 14 August 2018

Tags: AI, bot, customer experience, Artificial Intelligence, #CXreimagine, #datareimagine, #AIReimagine

Top Rated Posts

Blog archive

Stay up to date with all insights from the Empired blog