Responding to the growing complexity of navigating the digital workplace, enterprise technology has evolved to offer new routes to the information employees need.
Just as instant messaging / chat has been upgraded for colleague-to-colleague conversation (and validated by enterprise apps like Slack, MS Teams, and Workplace) chatbot tech itself is now a more recognised interface and channel for internal usage. No longer a novelty, chatbots bring a new and useful way for employees to get the answers they need.
Some communicators and ICT people may still wonder why they need to look at chatbots for employee communications. Chris McGrath, founder and CEO at Tangowork, believes that it is time for a full discussion about the value and usefulness of chatbots for workplace communications. His arguments are persuasive and worth considering.
MARGINALIA met McGrath at Intranet Now 2017 – for which Tangowork built the conference bot. In this interview, McGrath walks MARGINALIA through the best practices for companies to ensure a bot’s success, and describes the benefits to both the organisation and employee. He also shares his views on what to expect from chatbots in the years to come.
Gloria Lombardi: What’s behind the proliferation of chatbots?
Chris McGrath: Messaging is the number one use of mobile phones and also something we do heavily on computers. The ease in which we message people can be brought to the way we talk to computers and retrieve information. So the big driver of chatbot proliferation is that chatbots are accessed from where people already are – in their messaging app. It’s not a separate app or new interface. So, people can chat with their colleagues and then jump over and start messaging the bot.
This ease of access is complemented by ease of use – people can just use everyday language; there’s no need to learn special command words or navigate around a unique interface. People already know how to chat. They can just ask normal questions and the chatbot gives them answers. Chatbots are exceptionally easy to use, which supports adoption.
GL: How will chatbots change workplace communications in five to ten years time?
CM: In five years time, 90% of companies that have an intranet or employee app will have a chatbot of some sort. In 10 years time, any question we can think to ask in the workplace will be answered by a chatbot, either by voice or text.
Think of the day-to-day questions we have at work: ‘Can I take tomorrow off? Are there any IT alerts that I should be aware of? What were sales last month? What’s the date for our site launch?’ Right now, people search the intranet for answers, or email their manager, or HR rep, but a chatbot can give them the answer instantly. I see chatbots completely transforming the way we access information at work.
Any question we can think to ask will be answered instantly by a chatbot.
We don’t have to wait five or ten years — the technology exists today. But it’ll take some time before chatbots have realised their full potential within every organisation.
GL: What are the benefits for workers?
CM: Most companies have many different internal applications. Although everybody would like to have just a single source of truth, that’s not the reality. After a merger, an organisation may have several payroll systems, different benefits platforms, various intranets for their disparate divisions, and all sorts of file storage services. The digital workplace can be a confusing jumble, and people may not even know where to start to get the information they need.
A well-built chatbot can gather information from all those different systems so the employee has just one, simple interface to use. This single interface is the ultimate goal, allowing employees to access enterprise information in a natural manner on their messaging app or via a voice-controlled device.
So the benefit is that the employee can get the information they need right away, without having to think about where to start, who to ask, or how to search.
GL: Are people ready for chatbots?
CM: It feels like magic when a chatbot understands you and answers correctly straight away. So adoption can be quick, but it hinges on the quality of the information put in, and properly setting the expectations of the end-users.
Chatbot adoption depends on how well the chatbot answers employees’ questions. Bots in the consumer space get asked a couple of questions and if they don’t respond well, the user loses interest. Internally, at work, people are somewhat more tolerant because they don’t have many alternatives. But still, if the bot is repeatedly unsuccessful at answering, employees will abandon it.
The chatbot needs accurate information and relevant responses in order to serve people’s needs. When it’s built to be useful, it’s likely to be used.
GL: How can companies ensure successful deployment of chatbots?
CM: There are two best practices. The first one is to focus on a single domain of information. The second one is to grow the pilot.
Siri is backed by a team of hundreds of Apple engineers. But it’s likely that your first chatbot within the enterprise will be developed by only a handful of engineers and subject matter experts. So your chatbot cannot be ‘Siri for the enterprise’ – at least not right now. The scope of incoming queries would be too broad. Instead, the internal team needs to focus on a particular domain; a single theme or set of topics.
For example, a company that is relocating their headquarters has adopted a ‘move bot’, which is all about the move to the new office. Employees can ask ‘Why are we moving? When are we moving? What’s the nearest tube station? Where is my desk?’ The bot is purely focused on the practicalities of the move. This singular focus provides the foundation of success as there are only so many specific questions employees are likely to ask.
Another example comes from a financial services company, which has a complicated technology ecosystem. They’re implementing a chatbot that helps employees to understand the digital workspace and their app and service options. For instance, an employee asks, ‘How do I share a file?’ The bot might say, ‘If it’s not confidential you can use email or Dropbox. But if it’s confidential, you should use IntraLinks because it watermarks documents and it’s absolutely secure’. The employee can then reply, ‘OK, how do I get Dropbox?’ And the answer will help them get a business Dropbox account or tell them who to contact.
There’s another organisation with a large sales team. They were interested in streamlining the on-boarding process, especially to the sales team. They’re introducing a chatbot to answer all the obvious questions – the sort of things that sales team managers are asked again and again by new joiners: ‘What’s the pricing? Can we give discounts with this product? Can I get a coupon code?’ – this all saves time and helps people work better.
Each of these cases has a focused domain – the bot does not try to answer every question that an employee can potentially have.
GL: So that’s your ‘single domain’ advice; what’s the best practice for growing the pilot?
CM: To ensure success with their chatbot, companies have to grow their pilot gradually. There is no way for a small team to anticipate all the questions that a person could potentially ask a bot – and all the different ways people can express themselves.
So initially, a small team of stakeholders and subject-matter experts establishes an initial body of questions and answers. Then they roll out the chatbot to a friendly pilot group of 10 or 20 people, and they watch the transcripts coming in from the chatbot conversations. They’ll see dozens of queries that weren’t anticipated or that the bot is misunderstanding. They’ll teach the bot to properly handle those queries, and then grow the pilot to another 10 to 20 people. They’ll watch again as the transcripts come in, teach the bot more, and keep on growing and growing.
As the pilot grows, the percentage of successful responses climbs higher and higher. Once the success rate is in the 90-95% range, the company is ready to launch the chatbot to the entire audience. But they need to reach that point by growing the pilot gradually. Training the chatbot can continue after launch, of course; maintaining the accuracy of the information is vital.
GL: How long does it take to build an enterprise chatbot?
CM: In terms of complexity, building a chatbot is similar to building a website or web application. There are tools and methods to build a website in a day, but useful web applications can take months or years. It’s the same with a chatbot; it all depends on the scope. And again, like a website, you can build a skeleton quite quickly, but getting the navigation and content right can take a lot longer.
We can launch a chatbot pilot in a day or so, but we can also develop the functionality, content, and use cases over months. In many ways, it’s just a new channel; instead of a browser, the access and interface is provided by the messaging app. But the backend is as complicated as any web application.
While there’s almost no visual design to do, you still need to develop the personality of your chatbot to match your brand and deliver an appropriate experience for the audience. Should the chatbot present itself like a person, or like a sci-fi robot? What about gender? How might it use humour?
Most organisations I work with create their chatbots in a two to four months project, which takes around one to three hundred hours of the company’s time.
GL: Are chatbots based on AI?
CM: With bots, we talk about Natural Language Processing (NLP), which is a form of AI. Interestingly, Microsoft uses the word ‘Cognitive Services’, which for some people is more accurate than saying AI. Every chatbot uses cognitive services.
At Tangowork, we use Microsoft’s NLP cognitive services. It is called LUIS, which stands for Language Understanding Intelligence Service. When someone says something to one of our chatbots, we pass it to LUIS to extract the meaning, allowing the chatbot to then find a relevant answer from its info stores. We have to constantly train LUIS, to teach it different expressions and vocabulary that people might use.
But usually when people ask me about AI, they’re wondering if the chatbot can come up with its own answers to questions. And practically speaking, the answer is no. It would require massive sets of training data — perhaps millions of questions and associated answers — and we’re unlikely to have such huge data sets in an internal communications department. That’s why AI has been most successful in areas where those data sets exist, such as in language translation, facial recognition or medical imaging.
Chatbots at this stage in their evolution are more about ‘query and response’, and the “AI” is in understanding the meaning of the query and deciding how best to respond. It’s efficient and effective to anticipate questions (during the pilot) and write specific answers – relying on the artificial intelligence to extract the meaning and figure out the appropriate response.
The evolution towards enterprise-wide AI services accessed through chat and voice is very exciting, and the journey has started with chatbots for internal communications.