Can bots sell?

Duncan Anderson
January 13, 2021


The prevailing opinion seems to be that the purpose of automated Conversational Systems (chatbots) is in customer support. But that conventional wisdom might be wrong.

Most (every?) business is driven by sales. Sales is what keeps the lights on. It’s what ensures there’s money to pay employees. Without sales, few businesses (except maybe those funded by Softbank) can survive. Not surprisingly, many businesses have a high degree of focus on sales.

So when our client said to us “our boss wants us to focus on building something for sales”, it really shouldn’t have been a surprise. Nevertheless, our emotional reaction was: “hmm, if this works it could be pretty awesome. Emphasis on the if!”


We built a sales-focussed chatbot, because that’s what the boss wanted. It worked.

We sold over £2m of stuff (not vegetables, but room reservations) through it, in the space of just one month.

Everyone celebrated 🥳

🤔 How did we do that?

The answer is most definitely not about technology. No matter how good our technology, if our thinking isn’t dominated by the User Experience and the people we want to serve, we’d likely fail.

To force us to think hard about the User Experience, we asked ourselves two tough questions: How can we foster an impetus to act in our target audience?

What is the fastest, most efficient way possible to achieve the customer’s goal? Here’s what we discovered when we asked those questions.

👀 Getting eyeballs requires marketing

The amount of stuff our chatbot can sell is directly proportional to the number of people who use it. If people don’t know our chatbot exists, we have no chance of them buying anything we’re trying to sell. The marketing of our chatbot was therefore going to be just as important as the chatbot itself.

In our situation we were trying to encourage people staying in an accommodation to rebook that accommodation. So the target customers weren’t new customers, but existing ones. As such and more accurately, we needed a re-marketing campaign.

To attract customers we sent emails, or SMS’s, to our client’s customers. Sometimes we send to everyone, sometimes to select subsets — for example, we may send only to those who’ve not used our chatbot yet. Or we might email those who’ve used the chatbot, indicated an interest in making a purchase, but not actually completed that purchase.

Every email/SMS we send includes a button with a personalised “magic-link”. That magic-link launches our chatbot and tells us who pressed the link — meaning we can personalise the resulting conversation. The button in the emails was entitled “Hold my room!”, so those who pressed it arrived at our chatbot with a clear purpose.

This combination of knowing who the customer is and why they are using the chatbot meant that instead of starting the dialog with “Hi, how can I help you?”, our chatbot starts with “Hi Duncan, let’s check out your rebooking options.”

This is a big difference.

The average “Hi, how can I help you?” chatbot startup question is a big turn-off. I’ve watched many people freeze and not know what to do when presented with that question. And of course I’ve also seen people ask things the chatbot can’t do. Given that attention spans are limited, if our first chatbot interaction is being told “I can’t do that”, most of us will give up at that point and leave. So it’s safe to assume that at Humanise.AI we’re not big fans of “Hi, how can I help you?” chatbots.

Knowing who the user is and why they’re using our chatbot means that we can drive the conversation. The user reacts to things the chatbot asks, rather than the other way around. We’re ‘flipping the script’, as it were.

We don’t need to ask people what they want to do — we already know the answer. And we know who they are, so we can get straight down to helping them.

🗑 In the quest for simplification, we can jettison quite a lot of stuff

So we’ve safely delivered an expectant user to our system, primed and ready to go. We’ve nurtured an impetus to act.

Our job now is to make the subsequent journey as efficient and engaging as possible.

Our client is an accommodation provider and they wanted to encourage those already in an accommodation to rebook. The way this is done on their website is very sophisticated — their customers can view a map of the building and choose which room they’d like. It’s great! Only we theorised that it might be a bit too good — there’s a lot of steps and therefore a lot of places where indecision might strike, with people abandoning the process.

However, it turns out that most of the time people just rebook their current room. So we asked “what if our chatbot only does that?”. By eliminating all of the room choice steps, we could dramatically simplify the whole process.

For anyone who wants a different room, we send them to the usual website. Rather than try to replicate the full functionality of the website in our chatbot, we complement it with a dramatically simplified solution that works for most people.

🔑 Remove the need to login = remove a lot of friction

One other point of friction in the existing website process was the need to login. Too many people abandon at that point, unable to recall their userid and/or password. It’s a huge point of friction.

With our magic-links, we’re automatically aware of who the user is — no login necessary.

Had we needed a higher level of authentication, we would have SMS’d the customer a verification code, which they would have entered into our chatbot. Whatever the level of verification needed, there is no necessity for the chatbot to require people to go hunting around for a password. We’re huge fans of passwordless login — removing that friction is a big gain!

🙏 3 Button presses is all it takes

The final design of our solution books you a room within three button presses — which is a massive contrast to the norm. We lean heavily on button-driven input for two reasons:

  1. We have 100% certainty that we understand a button press, whereas no matter how good the AI, there’s always a risk we’ll misunderstand free form text. Guaranteed accuracy wins over uncertain accuracy every time.
  2. Buttons are quicker for the user and our focus is on making their lives easier. Pressing a “Yes” button is one button press, whereas typing “y-e-s-[return]” is four button presses. One button wins over four buttons every time.

👩‍💼 Human help is only a button press away

What I’ve described is a hugely simplified process that works for most.

But it won’t work for all — there’s always going to be exceptions. So we offer people the option to request a human agent to join the chat and help them out, if they need it — so help is only a button press away! This is the great benefit of Conversational Systems —we can’t get human agents to join on a non-conversational website! It turns out that only about 8% of customers need that option, but it’s important that it’s there to provide a fallback option.

Offering human help means that we can keep our chatbot focussed on the sales process, rather than having to try and make it address all sorts of obscure edge cases. Human agents are there for edge cases — and they are better at it than chatbots (edge cases being tough to predict and tougher still to train an AI to handle).

💰 An approach that’s driven £2m of sales

What I’ve described here is an approach to designing the User Experience for a Conversational System (chatbot) that’s so far closed over £2m of sales. The emphasis on marketing was critical, because that’s how we can ensure that people actually use it.

At Humanise.AI we believe we’re in the game of “Conversational Marketing”. Rather than building a Conversational System (chatbot), we build a marketing campaign that links to a Conversational System (chatbot). This difference makes all the difference.

We also placed an emphasis on stripping away complexity, removing login requirements and getting to an answer in just three button presses. Too many people abandon complex processes, so by simplifying we can drive up the proportion of people getting to the stage where they might make a purchase.

I’m writing about how Conversational Systems can be an impetus for User Experience thinking, because I feel it’s an often ignored art. At a time when the emphasis is too often on cool technology, a simple emphasis on making peoples lives easier is sometimes lost. We named our company Humanise.AI because we believe in the art of humanising technology — the right answers are so often dependent on thinking about people and how they behave. Suffice to say: the success or failure of a Conversational system is down to how much we design the user experience and challenge the status-quo.

ℹ️ What to know more?

If you'd like to explore how a combination of a chatbot, an email marketing campaign and our PlatformOne communications system can bring you these kind of results, we'd love to talk!

Contact us at

Previous article
When AI writes poetry
Next article
Habitat Case Study — Student Accommodation