Why we killed the Slack chatbot we built for small businesses

Rexbot.io was aimed at small businesses suffering from the effects of assumed knowledge. We built it in 5 days, then killed it. Now it’s available as an open source project.

Rexbot.io was designed to tackle the problem of organisational ‘assumed knowledge’

Every year, Mint skidaddles out of London for a week and splits into teams to work on a challenging brief. We call it the Web App Weekender (it has grown in size since it was first named). Last year the location was beautiful Devon, and the challenge was to build something that would make money by the end of the week.

Here, we’re going to look at what one of the two teams built — a Slack bot called Rex. We’re writing about Rex after all this time because, for a little while, it felt like it might be a viable opportunity for a Mint venture. The reasons it didn’t — which will become apparent in this article — are hopefully illustrative of the rigorous approach Mint has to validating products.

The efforts that went into creating the prototype, meanwhile, are an example of what our CEO Tim has referred to as ‘startup training’.

We’ve open-sourced Rex, so if you like the look of the product feel free to take him for a walk and see where he leads you.

The idea

I (content) had the pleasure of being on a crack team of multi-disciplinary masters: namely Colin (finance), Shoshi (front-end development), Simon, (back-end development) and Vala (design).

Early meetings (conducted in advance of the Weekender) suggested we were going to struggle to find and build something that would allow us all to exercise our respective skillsets. Then we made the decision to effectively ditch the brief and build something else instead, which is when we hit on something we were all really excited by.

We had been mulling over the idea of a chatbot for a little while when the notion of a Slack bot to help small businesses took hold. We breathed a sigh of relief and focused our energies on making this product happen.

The name

A perhaps disproportionate amount of time was spent trying to come up with a name for this bot that didn’t exist yet. We knew the character would need to be friendly and helpful, and ideally compensate for the unavoidable fact that he/she would be a bot.

We started thinking about using an animal as the character, then a dog (human’s best friend and all that), and, eventually, we landed on the name Rex. That decision meant Vala could get started on creating the visuals for the character who would be the centrepiece of the product.

The vision

Our vision for Rex was that he was the small company’s tool to combat the dangers of ‘assumed knowledge’. Startups tend to leave it pretty late to invest in the HR function — they need to concentrate on building a product/company first. As a result, a company’s operational information tends to be shared informally by word of mouth — here we mean things that help an employee feel like they’re working at a ‘real’ company: holiday entitlement, pay dates, invoicing procedures, expense claim procedures, etc.

That’s fine until the company grows further and silos develop. Suddenly employees don’t know who knows what, and with a lack of a central information resource knowledge can fall between the cracks. Culture and company progress are both jeapordised.

We felt that employees could be introduced to Rex as part of an employer’s onboarding process, and trained to ask him operational questions. Before long he would become part of the company culture, and the default way people would find answers to repetitive queries. Tracking what people asked him, meanwhile, would provide directors with useful data on employees’ problems.

The deliverables

We agreed there were three things we needed to deliver to be in a good spot to communicate the product’s value come the end of the week:

  • A functioning Slack bot. Something we could actually interact with to demonstrate the experience a user would have.
  • A website that would demonstrate what the product does and familiarise people with the brand.
  • A web admin interface that would allow for the entry of the content that would comprise Rex’s responses. This would need to be user-friendly enough to appeal to staff whose job it would be to add information.

The content

We had a vital secret weapon in our team when it came to content: his name was Colin, Mint’s Head of Finance. With several years’ experience at Mint, Colin was able to bring a comprehensive understanding of the kinds of questions staff have about workplaces.

This meant we were able to populate Rex’s memory banks with plenty of questions and answers, covering a wide range of topics. Things like:

  • How can I find out my holiday entitlement?
  • Where can I get a copy of my payslip?
  • How are client invoices issued?
  • How do I claim expenses?
  • How do I book a meeting room?
  • Is there a cashpoint near the office?

When we were devising the questions for the development team we highlighted the main key phrases that Rex should focus on to pull up relevant answers. So if Rex detected the phrase ‘meeting room’, he would work out that the user was asking about booking a meeting room, and he would return the appropriate response.

Personalising answers was too complex for the time period we were operating in. So we couldn’t expect Rex to answer, say, “What is my holiday entitlement?” But we could tell users what steps to follow to find out this information.

So if ‘holiday entitlement’ was the phrase Rex was to focus on, he would answer “What is my holiday entitlement?” like this:

  • Q: ‘What is my holiday entitlement?’
  • A: ‘You can find out your holiday entitlement by doing X, Y & Z.’

By returning the relevant question and the answer, it would at least help the user see that while they weren’t getting an answer to the question they asked, they were getting a response that would help them find that answer themselves.

The call-and-response nature of Rex meant we were also able to have a bit of fun with Easter Egg-type hidden responses. So, for example, typing “Woof” would return “Woof back at ya”, and “Lie down” would return “You want me to take the rest of the day off? Fine by me!”

Making the decision that Rex had a slightly mischievous sense of humour meant we were able to apply that to other standard responses, such as those that would be returned if he couldn’t find an answer, etc. Touches like these helped develop Rex’s personality and how we thought about him — during the week of development, it didn’t take long for the team to stop referring to ‘the bot’ and start referring to Rex by name.

When it came to the website, we decided we wanted to have animations of Rex in action — one of them showing the bot itself, and the other demonstrating the admin interface. In the end Vala created a designed version of the bot interaction, while a Quicktime screen recording did the job for the admin interface.

The demo

When preparing to demo Rex at the end of the week, we felt we had an ace up our sleeve in that, rather than showing people how it would be used, we could ask them to use it themselves. We would kick off with a presentation talking about the problem we were trying to solve, before putting Rex directly into our colleagues’ hands.

To do this, we secretly added Rex to Mint’s Slack team. Then we added questions and answers relevant to each member of the team that would be trying Rex out. Then, during the demo, we prompted each team member to ask Rex that question. We knew that as long as a key word/phrase was used, Rex would be able to return the desired response.

So for example when Brad had his phone in his hand, we told him to ask Rex about the cost of flights to Dubai (one of his regular holiday destinations). Rex returned the answer: “You can Google that yourself, Brad.”

This had a great impact, in that we were illustrating how the product worked while also personalising responses to the people using it. We could see how an approach like this might work as a cheeky preamble to a pitch meeting with a client considering purchasing Rex.

An iteration of the Rexbot homepage.

The outcome

The Enemy, aka The Other Team, had created a product with more potential to fill the brief (i.e. to make some money by the end of the week). It was a transactional Facebook app called Friends11 which allowed you to create a football team from your friends, share it and tag them in the process, and thereby power viral growth.

In the event, neither team technically filled the brief by the end of week, in that neither made money (although Friends11 would proceed to generate revenue after further development a few weeks later).

That meant that deciding on ‘a winner’ came down to a vote amongst those present, which Rex won. But it’s safe to say that ‘winning’ in the Weekender is less important than building something useful, learning new things, forging bonds with teammates and feeling proud at the end of the sprint.

In that sense it was a hard-fought, good-spirited draw that left everyone with a sense of accomplishment and excitement at what the products might yet become.

What happened next? (aka Why isn’t Rexbot.io a thing?)

We envisioned Rex as a potential SAAS business, with paid tiers based on the number of users. Bolstering our confidence was the announcement that Slack was seeking to invest in startups building bots for the platform.

At this point we hadn’t done much analysis of what else was out there offering the same proposition. So the next step was to do that. We found several startups operating in a similar area to Rex, many of whom had taken investment and so had considerable resources at their disposal.

Now we needed to answer the following questions:

  • Given the competitive landscape, what would it take for Rex to succeed — and is this possible/realistic?
  • What do we know or have that our competitors do not?

Implicit in the questions was the consideration that Mint is not a big company, and that it has existing clients and commitments. While devoting the majority of the company’s resources to one startup has happened before, it’s a big decision to make and one we’d need to be very sure about — i.e. we’d need very strong answers to the above questions.

Additionally, with the resources available to the competition, competing on a level playing field would probably require outside investment.

Suddenly, the prospect of turning Rex into a product worthy of Mint’s sustained investment were diminished. So we took the decision to put Rex down.

There’s no shame in this: as a successful startup studio Mint has killed dozens of nascent startups that weren’t right to proceed with, and we take the view that each one is a stepping stone towards the next Big One.

The Rex team thinks back fondly on our helpful little dogbot, the fun we had dreaming him up and the week we spent in his company. And we do so safe in the knowledge that somewhere up there in heaven, Rex is helping an overworked angel find out its holiday entitlement.

Want to work with Rex? We’ve made him available as an open source project.


Originally published at mintdigital.com.