Chatbots for behaviour change in L&D, with Vincent Han

This interview with Vincent Han is the start of a section of this series that focuses on chatbots and L&D. When I started to record this podcast series I didn’t expect so many of the interviews would be focused on chatbots. Chatbots are the easiest AI technology for L&D to implement and have a high impact. Vincent is one of the leading thinkers in this area of chatbots and L&D. He is the founder of the Mobile Coach platform, which is an excellent way to get started building chatbots.

While this interview with Vincent is at the start of this section in the series, it wasn’t recorded in this sequence. What happened is, as I was doing some of the other interviews in this section, with Jamie Good and Emma Webber, I discovered that they were both using Mobile Coach to build their own chatbots. As a result, I thought it would be great to talk with Vincent for this series.

Vincent doesn’t come from a learning background; his background is technology, so his focus on learning as behavioural change is refreshing. In this interview, Vincent gives a great overview of how chatbots can be used to personalise learning over time and how the process of a conversation can be emotional, even if it’s with a bot.

Often I think about a chatbot as being on a website. Mobile Coach can be used for building web-based chatbots, but it can also be used with other other chat platforms and SMS. SMS messages are simple, personal and can a be powerful tool for behavioural change.

If you want to begin using chatbots, the Mobile Coach platform is a great entry point.

Download the how artificial intelligence is changing the way L&D is working eBook

To go along with the podcast series on AI and L&D, we have released an eBook with transcripts of all the interviews. The eBook also gives a brief explanation of what AI is and an overview of how it is being used in L&D.

In the eBook you will learn:

  • Some of the jargon behind the technologies e.g. what data scientists mean when they talk about ‘training a model’. 
  • How AI is being used in L&D today to gain insights and automate learning. 
  • Why you should be starting to look at using chatbots in your learning programs.
  • How you can get started with recommendation engines 


Useful links from the podcast

Transcript - Chatbots for behaviour change in L&D, with Vincent Han

Robin: Vince, how did you come to be working with chatbots for behavioural change?

Vince: Well, my background is in technology. I've always been fascinated with behavioural change. I'm a serial entrepreneur, and I've founded several different technology companies. About five years ago, I was very fascinated with the whole behavioural change space, which includes fitness apps, weight loss apps, and whatnot. I had a personal story as well, where I was quite overweight and was looking for technology solutions for myself as well.

When learning apps don’t include behaviour change, something is lacking

Vince: As I was researching the different sorts of behavioural change and user engagement applications out there – in other words, what apps help people to be accountable, and help them know what to do – I wasn't very impressed with what was out there. Although there were some pretty nice looking apps, they were really catered towards people that were going to be active anyway. I sort of joke that MyFitnessPal and Runkeeper are apps that help people who are already fit easily brag about their exploits on Facebook.

That's the backstory, and as I researched about what types of technology solutions could really help people, I realised that the key nut to crack, so to speak, is how you get someone's attention on a consistent basis over enough period of time, to where you can really influence their thinking, their behaviour, and whatnot.

How to get the most from a chat interaction: analysing which messages get responses

Vince: That brought me to chat. I asked myself the question: what messages do I not ignore? The answer was: messages from my friends, family, peers and colleagues, that are coming on my phone. This was five or six years ago, before chatbot was really a thing. I thought, ‘Well, could we automate a personality that would get that same type of attention from a user?’ Through some early experiments we found that we could do that, and thus the ambition towards doing chatbot work was born.

Robin: There are lots of interesting things in there: the fact that mobiles have become this intimate platform for communication for us with our family and friends, and the simplicity of the nature of it. Someone said, one day in a workshop session, that you couldn't teach project management on a mobile phone. One of the Sprout team piped up and said, ‘But Robin spends his whole day on the phone doing project management.’

It’s a communications device and yes, it's actually returning it back to the sense that a phone is for communicating with someone, or something.

It's also interesting you came to it from that behavioural change point of view, rather than a content point of view. What did you think was really lacking in some of those other behavioural change apps, where you might have to do things like check-ins to try to be accountable?

Vince: Well, I think that they all had a very short lifespan with users. I think that the challenge I had with some of those apps was that changing your behaviour is very hard. That's why I think the statistics around people who give up on a New Year's resolution, for example, are so high. It's astoundingly high. Even when a New Year's resolution is life or death, I think there's plenty of data that shows that really important behavioural change is extremely difficult.

So I didn't find the hooks and key features – check-in features, things like that in those apps – had the persistence level that was required for people who really needed behavioural change. That was my big takeaway.

Robin: Yes, it's that accountability bit in behavioural change that seems to be the powerful thing that a chatbot can enable. It's what a personal coach can sometimes do.

Emotional pathways to sustained behaviour change in learning

Vince: I think that there's an emotional level to it. A lot of times, technology is very pragmatic. Like, if we take a look at those fitness apps: here's how many calories you need to eat, and here's how many steps you need to walk a day, and maybe there's even some gamification to help ‘motivate’ you to action. But when it comes down to making important choices, those choices can be very fleeting. A lot of it depends on how we feel. When it's emotionally-based, it's not pragmatic, it's not logical.

I think a chatbot is interesting because it sort of diffuses this idea of user interface. Conversation, I think, can be a better position from which to help someone's mood, or help influence them in a way that a more structured app cannot.

Robin: Because talking and chatting with a bot isn't something that people all of a sudden relate to as an emotional experience. I think people could relate to the fact that the barrier to behavioural change is emotion, but you don't actually think that a script or an AI would be able to help you deal with an emotional ‘if, then, what?’ situation.

Vince: Yes. I don't think a chatbot can be very successful in being a psychotherapist, and so that's not really what I mean.

What I mean is that when my emotions are getting in the way of healthy messages getting to me, like having a good reminder, then a ‘come and get it’ in, for example, a text message – I'm going to be more prone to look at it and read it, than a push notification, or an email, or something like that. When I say emotional, I mean making decisions based on how I'm feeling. I could be feeling good for all that matters. We're just more likely to see a message that's in a conversational style, going to a more structured application.

Robin: I've just come out of a coaching session myself, and there was an interesting moment to think through. By verbalising some thoughts, I actually sorted the thoughts, and the coach was sometimes reflecting back and helping me sort through those things.

That process of articulating what's going through your mind actually helps to sort it, which is interesting in terms of the chatbot prompting those types of possibilities.

Vince: Yes, that's a great comment. I would agree with that.

Robin: What do you think are some of the possibilities for chatbots in workplace learning?

Matching the profile of a chatbot to the profile of the end user/learner

Vince: Gosh, I think that there's really a depth of possibility. What we have found in our work is that if you can design the personality of a chatbot to be appropriate to the demographic of the learner, and be appropriate to the level of content that the learner is trying to be successful in engaging, then you're going to have that person's attention. That gives you a platform then to teach, to remind, to influence. So some of the possibilities can be in having sustained learning which is measurable, for once.

I think a chatbot also – since it's technology-based – can meet learners where they are. I think those of us that have been in learning and development, and building programs, are often handcuffed by creating a single curriculum that everyone has to engage with, regardless of their learning style. All of a sudden, the chatbot can now personalise learning style as well as pace of learning.

Imagine a chatbot asking you, ‘Robin, I can send you these exercises for you to work on to improve this particular skill. Would you like to do that once a week or five times a week? Would you like to do that in the afternoons or in the mornings? Would you like me to remind you or just challenge you?’ Imagine having all these sort of preferences that the chatbot can learn from a learner, and then have the user experience match those preferences.

I think that's a depth of capability that can really transform how effective training can be in the workplace.

Robin: That's a really nice example of personalisation. Mark from Filtered, who have a recommendation engine, talked about how he was using chatbots for profiling people to be able to build those personalisations. It's just a different, more flexible way compared to filling out a form with that layer of personalisation as well.

There are two things that I see, two big areas in chatbots. There's that whole thing around support and just-in-time performance support, which is where quite often they're being used in software to help people find answers to things. Then there's this other thing that you're really interested in, which is more of the ‘you need to learn something to change your behaviour, move forward on something’.

This might even be during an onboarding program: you've got a series of common questions that are asked, and the chatbot can be programmed with those questions and answers.

Behavioural change in learning is still an emerging use scenario for chatbots

Vince: Yes. You know, I'm actually in both of these cases, but I think we find that the approach of using a chatbot to proactively influence and teach and try to change behaviour is a little bit more rare, for sure.

An easy way to think about it is: can you have a chatbot that pushes messages proactively for the purpose of behaviour change or influencing? As well as having a chatbot that can pull, that will wait for an inquiry from the user and a performance support use case? You can blend them as well. I think that if you can create enough value in the messages, where the learner can trust the chatbot, then really the best of both worlds can apply. In terms of having the chatbot both be responsive to questions, and situations that a learner might find themselves in and needing a little bit of support, as well as having the chatbot to be a coach.

I think that some organisations might be wary of, ‘Will a learner want to have an AI bot proactively messaging them at work?’ I think that there are some valid concerns around, ‘Will that feel annoying or intrusive? Can it miss the mark a little bit?’ But we found that not to be true.

Can chatbot interactions improve on human interactions?

Vince: I think that if as long as the messages feel valuable and there's enough context, people are actually quite welcoming of messages that are catered to helping them improve. One big lesson I've learned is that you can be a bit more liberal and aggressive than one might initially think in terms of having a chatbot really try to push a conversation along, and push a particular learner along a development path.

Robin: That's a lovely bit of insight, to realise that you can really push. Do you think that's also because the technology feels less invasive? That people know it's not actually a person doing the pushing?

Vince: We have some pretty interesting anecdotes about that. In some instances, I think, learners would prefer to interact with a bot than a real person. Because if you're interacting with a real person, particularly someone that you know, there are all these other expectations around the relationship, that you have to be cognizant of. But if it's a bot you feel less guilty about delaying in response, or being curt in your responses, because you're not going to offend the bot.

If you take some of those dynamics as advantages in designing your chatbot, then absolutely, you can have learners really engage with this bot.

Using real-world human profiles and scenarios to design chatbots

Robin: That lets me go into one of my next questions, which is around designing a chatbot. You talked a bit about personality. It's that sense of: if someone was thinking about designing a chatbot, where do you think they should start, Vince?

Vince: Let's take a use case where there's some frontline supervisor training that we want to support with a chatbot. Maybe I have a thousand frontline supervisors in my organisation across the world. The design sensibility I would start with is that if I had the time to individualise a follow-up campaign with each of these frontline supervisors on my own, what would I say? When would I say it? How often would I say things? Would I say things differently to my frontline supervisors in the United States, versus New Zealand, versus Australia, versus the United Kingdom? Start asking those questions.

Instead of working backwards from ‘what would a computer say?’ I think a good design sensibility is ‘what would I say?’ and trust that intuition. That, I think, is a great starting point in terms of designing a personality and tone of a chatbot.

Robin: It's interesting that we're back to that personality word because you're the first person who's talked about personalities in chatbots. It's possibly because you come to it from this sensibility of thinking through what would a person do, how would that person engage in an intimate way with people, which is a very non-tech way of going about it.

That's the support scenario, but how about if the whole learning experience was a chat experience?

Creating chatbot interactions: push versus pull

Vince: There's an instructional design bot in the chatbot. So first of all, I think there's a couple of components. For the push component, if you're wanting to have this sort of push curriculum, then that's a calendar-based type of instructional design, to say, ‘Okay, how long do we want this chatbot to go over this curriculum?’ Maybe it's one month, maybe it's three months, maybe it's longer. Then you ask yourself the question: how many times during a week would it be appropriate for the chatbot to push some sort of exercise, or reminder, or some sort of video, or a piece of learning content? The answer might be once a week, twice a week, every day. It depends on the nature of the content.

What I found with most people in the learning world, they have really great sensibility about answering those questions: how long of the experience should it be, and how many times in a week does it feel right? Typically, we're finding that for a chatbot experience in a push scenario of 60 days, or two months, sending messages once or twice a week seems to be a good sweet spot. I think that makes common sense for most people thinking about that problem.

Then, once you have that calendar, you're simply creating this micro learning campaign. You say, ‘Okay, on week one, what do we want to emphasise? On week two, what do we want to emphasise?’ After you have that outline you drill down to the actual messages. Then, ‘Okay, in week one we know we want to emphasise this and this. How do we want to write those messages?’ That's a pretty straightforward process for the push piece.

For the pull piece, you're really just anticipating the different types of questions and queries that a learner might ask. The most important design element there is giving them instructions on how to use the chatbot.

I always point to well-known AI bots like Alexa or Siri or Google Assistant. Sometimes those companies' marketing messages around those products are: ask it anything, it will be able to answer anything.

In reality, I'm an iPhone user and so I use Siri, and Siri is actually very poor in answering a lot of the questions I would like Siri to be able to answer. In a workplace, you want to avoid communicating to learners, ‘Hey, ask our AI bot anything.’ Instead, you want to say, ‘Hey, if you would like scenarios, or if you'd like examples, or if you'd like a job aid, ask the chatbot. It's going to be the fastest way to get to those resources.’ Then I think it can be really successful.

From a design perspective, giving people clear directions about what the bot can do and what the bot cannot do is paramount in the design.

Robin: In two of the other podcasts in this series, both the people have ended up using your platform, Mobile Coach, which is interesting for many reasons. How is your platform different to other chatbot platforms that are out there? Also, why is it a good platform for learning workplace activities?

Vince: Our sensibility around chatbot building is that we believe there needs to be a robust authoring tool for chatbots. Fortunately, there is more and more that are emerging, which is welcome news for those that want to use chatbots.

For a long time, really, if you wanted to build a chatbot, you had to be quite technical. Let's say you wanted to build a chatbot on Facebook, on Facebook Messenger. While the Facebook Messenger has SDK (software development kit), if you were intimidated by writing code, then you would never end up writing a chatbot. The other downside with SDK chatbot authoring is that it’s cumbersome to make changes.

Our platform is five years old now, so we've been working at it for some time, and our goal is to make authoring as simple as possible. If someone is brainstorming about what they want a chatbot to do and thinking, ‘Boy, that sounds really impossible, how do we ever get a chatbot to do that?’ our goal is to have feature sets in our authoring platform to allow that to be possible, and hopefully quite easy to configure.

I think that that's a unique point of view, which makes our platform stand out a little bit. It's just providing an author with a robust authoring set, in an easy-to-use interface.

I'd say maybe the one last thing that really distinguishes us is that we're enterprise-friendly. From day one, we were thinking that in large enterprises, there are going to be a number of stakeholders who are keenly invested in the outcomes of the learning and/or behaviour range of a constituency. Whether it's employees or new managers, new employees, or even customers, enterprises have unique requirements. There are requirements around privacy, about data integrity, different stakeholders. So building workflow controls that are friendly for the enterprise into our platform, including robust reporting, have set us apart.

Robin: Yes, and you're right, that essentially there are a lot of products, especially say, in probably the chatbot area, that are designed more for software support for a smaller company, but don't really meet those needs of an enterprise.

Vince: Yes.

Robin: If someone wanted to get started with designing and building a chatbot, what would be your suggestions?

Vince: My first suggestion would be get familiar with a chatbot as an end user. So go find some chatbots on Facebook Messenger, on WeChat, on Slack. Interact with them and be really cognizant about what works and what doesn't work as a consumer. I think being a user is the first step. Then if you're ready to embark on creating a chatbot, there are a number of different authoring solutions out there including ours at Mobile Coach.

Choose what channel you'd want your chatbot to live on, whether it be on SMS, or Facebook, or some other messaging app. If it's a messaging app, most likely that messaging app will have the chatbot SDK. Or anyone is welcome to reach out to us and try out our authoring tools as well, at mobilecoach.com.