Performance support and chatbots, with Jamie Good

Jamie Good is one of the people who have pioneered using chatbots in L&D. In this interview we explore how chat can be used in L&D. He talks about a great example of a conference chatbot that focused on helping conference attendees to collect their takeaway from sessions and making those takeaways more actionable.

For Jamie, one of the most powerful applications of chatbots in L&D is in performance supports, which is something that Paul Healy from Learning Pool also talks about in a later podcast. This interview is also packed full of great advice from Jamie about getting started with chatbots in L&D.

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

Subscribe using your favourite podcast player or RSS

Subscribe: Apple Podcasts| Spotify | Amazon Music | Android | RSS

Useful links from the podcast

• Find out more about Jamie
  
• Find out more about Woebot

Transcript - Performance support and chatbots, with Jamie Good

Robin: Jamie, you've done lots of work and thinking around the potential of chatbots in L&D. What are some of the possibilities for the use of chatbots in workplace learning?

Jamie: One of the reasons I got into chatbots in the first place was because I saw the potential for them to be used in almost any situation. It just depends on what you want your goal to be. But with learning and development specifically, I think one of the great things a chatbot can provide is performance support.

I've seen, just in my experience at least, that we’re not really all that great at performance support. We’re good at building the learning experience or the scenario for someone to learn. But then we forget about the follow-up. I think that a chatbot can do a really great job in performance support in terms of automation, making it personalised for the user, and making it easy and frictionless for example, by just having it on someone's mobile phone.

Robin: Chatbots provide a very easy way to build something that's actually in the flow of work – which is what performance support should be – rather than something that only lives in an LMS.

Chatbots and mobile experiences

Jamie: We live pretty much on our phones now and they're in our pockets. So why not build the learning there too? Texting is the number one activity on a phone. We're already doing that. A chatbot works the same way. You don’t have to teach someone how to use it.

In an LMS, sometimes the first module is how to use the LMS. With a chatbot, you just start texting. It’s barrier-free, frictionless learning. And the fact that you don’t need another login password to be able to use it is really good. The fewer barriers we put in front of the learner, the more likely they are to learn.

Simple text interfaces

Robin: Yes, text is a really simple technology. The first computer interfaces were essentially command lines, and it's really interesting that as we move forward we're returning to something just as simple. Why do you think we’re moving towards a simple text-based way of working?

Jamie: Because there's too much around us right now, I think. There's so much complexity. I have this term that I've been using for a few years now: ‘infobesity’. It's almost like a disease. There’s so much information being thrown at us from so many places. There's always a new app or a new software tool. And we have to learn how to use that one and then learn how to use this one and then the tools are upgraded and changed and so we've got to learn a new feature. Texting is just simple. Use your thumb or even speak to text and send a few words and get some back. I think that if we're mixing learning with something complicated, the cognitive load becomes such that it's more difficult to learn and it's not as enjoyable.

People just want to get straight to what they want as quickly as possible. And you can do that just by texting. The other thing about chatbots that I like is that you can pull or push content. You can design it so that people can grab what they need, 24/7 – a just-in-time approach – but you can also automate nudges and reminders.

Examples of chatbots in learning – a conference takeaway chatbot

Robin: Have you got some examples of ways in which chatbots work nicely with performance support?

Jamie: One performance support chatbot that I built that was pretty successful arose because I was tired of going to conferences and learning all these great things from keynotes and sessions, and then getting back into work and forgetting everything. I was like, ‘Wow, I flew across the continent. I spent money on the flight and the hotel. And then I forgot what I learned.’

So I built a chatbot – I guess you’d call it performance support for conferences. You’re sitting in a session and you hear something you want to remember, so you just text the bot the word ‘takeaway’.

It would say ‘Okay, I'm ready, what's your takeaway?’

You text a few words, then the chatbot would ask you when you want to be reminded of that takeaway. The chatbot reminds you. It would then go a bit further and ask who might help you make the takeaway a reality, what are the barriers you need to get out of the way first, who else might be accountable, etc. So that’s a simple way of giving people performance support for their own personal development when they go to a conference.

Robin: That's a really cool example Jamie. It's nice to be able capture the takeaway but also to think about the barriers, which is helping people make the idea a reality.

Jamie: We're learning all that great stuff in that conference, but if we're not doing anything with what we are learning, then what's the point, right?

Robin: Technology can be used for different types of note-taking, logging, and journaling in lots of different situations at work.

Jamie: I've seen many people have their laptop open during a conference, or they have their notebook with their pen. They're writing feverishly, and I would bet that not many people go back to those notes once they get back to work.

Robin: Yes, we tried to do a joint collaborative Google doc for the virtual learning conference and it was really interesting. I found I couldn't personally split my attention between working on notes and actually facilitating. We're working on a screen trying to do two things, which is not great. Whereas maybe an actual mobile device that is separate might be the right way capture notes.

Chatbots are often used for support and just-in-time help in software.

Jamie: Yes.

Robin: You’re doing a lot of work around soft skills at the moment. Do you think there's an application for chatbots in those other sorts of less technical areas?

Jamie: I think if you're creative and smart about how you approach it, you can help people learn quite a bit through something simple, like texts. I mean, of course it has its limit. You can also include links to videos and PDFs, or blog posts. You can include images, GIFs. So there's actually a lot that can be included apart from just texts.

You did have to consider, though, the data and whether people are willing to use up their mobile data. You don't have to be limited to just texts. I think that in soft skills you could create conversations that help people realise, or at least reflect on certain things that they're doing or saying, or how they’re behaving.

You could have them reflect on how they're applying or not applying these things outside of that texting conversation, like in their workplace. Sometimes I think just getting people to think about something, and reflecting and pausing, is even valuable in a situation like that.

Robin: Yes, and I do too. When the podcasts are all published, Jamie, I encourage you to go and listen to what Emma Weber at Lever Learning has been doing. She's using chatbots to trigger reflective thinking, to activate learning. It's interesting that that's what you're talking about as well.

Jamie: Yes, I think that's a good idea.

The limitations of chatbots for learning

Robin: Did you want to go over something that's a bit negative? You said that there are some limitations. What do you think the limitations are?

Jamie: I think it'd be pretty cool in a chatbot if you use kind of a branching scenario like Cathy Moore talks about. Where you take people through a variety of choices, and one choice leads them down the wrong path but another choice leads them down the right path and they learn through discovery. But there's a limit to how much people want in chat. I mean, it depends on the person, but I sense that people don't really like to be texting on and on and on and on with someone for an hour. After five, ten minutes of back and forth with someone, sometimes we just make the phone call. So I think some of the limitations are just the amount of texting back and forth that it takes for someone to learn something.

For example, with the Domino's Pizza chatbot you can order your pizza, but it took me so many texts back and forth to talk about the toppings, the thickness of the crust, my address etc. that I could've just done it online or on the phone faster. One thing I think we need to consider is just how much are you asking people to text back and forth, before they actually get some value out of it.

Robin: Yes, so it's possibly more suitable for smaller chunks of learning. Larger pieces of learning are possible over time.

Jamie: Over time is another advantage of a chat, but we can create a space-to-learning experience by having that automated over time. I think that's another benefit.

How to develop a learning chatbot

Robin: You've developed a few of these now, Jamie. How do you go about planning them? How does the the technology work? I’m trying to demystify the process a little bit for listeners.

Jamie: I’m not a coder or the programmer. For me, I need to find a DIY approach. What I use is Mobile Coach out of Utah. And the reason I use their platform is because I didn't need to code or program. I kind of know how programming works, but I didn't actually need to program the backend. I just needed to think about the scripts and what kind of language and punctuation, responses etc., I was going to put in there. But the action of that being automated and going to someone's phone, they had already built in the background of their platform. So you don't have to be a programmer or a coder to do this.

It seems to me almost every week there's a new platform up there to help people build chatbots, who don't know how to program. These platforms are really just plug and play: you put in the pieces you want and then the backend is already built for you. People think artificial intelligence is amazing, and it is, but it’s not where it needs to be for something like natural language processing. For example, if I want you to learn something, I'd ask you a question that would be something like, ‘What did you do today?’ because I want you to reflect on your activity for the day. You could really respond with anything. There's no way that the chatbot would be built to be able to handle any possible response to a question like that.

Chatbots as decision trees

Jamie: So what I think is a smarter approach at this time, until natural language processing catches up, is that you basically guide the conversation by giving the user certain responses that they can choose from. For example if I ask you, ‘What did you do today?’ I might give you three responses: I worked really hard, I slept all day, or I visited with my friends, or something like that. I give you the choices and you just tap on one of those and then the next question follows. Otherwise, the bot will be what we call broken. People will break the bot. Because if the bot doesn't recognise the response it gets, it doesn't go any further and then the conversation ends for the user.

That's something to really watch out for. When you're planning and preparing to build, you have to think about making sure it’s not going to break. You need to think about a decision tree and how it’s going to guide people, while also making sure they feel like they're in control.

Robin: So it’s almost the decision tree process, where at moments there are choices that are like a series of multiple choice questions. Then there are moments that are more open. And some of the moments that are more open-ended questions, you're not actually getting the chatbot to respond. Those are more about the reflective thinking process.

Getting started – check out Woebot

Jamie: If listeners haven’t used a chatbot before, there's one on Facebook Messenger that I think is quite good at what it does, and it’s been built by Stanford. It's called Woebot. It’s a mental health, cognitive, behavioural chatbot. It will ask you, for example, ‘What are you doing right now?’ And then you will just write that in and the bot will respond with the next thing that it wants you to do. It won't respond to what you just typed because it can't. So basically it just gives you the opportunity to sit and reflect and go, ‘Oh okay, what am I doing right now? To be present?’ But it's not built to respond to that necessarily.

It can capture responses and feeds them back the user, but assuming that the chatbot can respond to anything people throw at it, is something else. That's just not possible, really.

Robin: Yes, I think technology is at a very interesting turning point, especially around natural language processing. It's still fairly primitive as technology, but it’s fascinating how essentially we're still able to do a lot and get different types of results from using the fairly primitive tools. When it comes to scripting, what's your other advice about writing the scripts?

Writing for a mobile experience – keeping it short

Jamie: So I think scripting, it's interesting. I did an in-house learning summit for a company earlier this year, and I had them practise writing scripts for a chatbot. Of course issues came up like what I just spoke about, where they were writing really open-ended questions and I told them it would be impossible for the bot to respond to that. They had questions that also, when written on the whiteboard, looked fine but then they said, what's that going to look like on someone's small screen?

So I think practising scriptwriting, even if you're not going to be able to chatbot, but just thinking about that skill, is really good because it can be transferred to so many other things. Like if you want to work on microlearning, or you want to be more succinct in your eLearning scripts. If you're thinking about that, it may be right for texts, and thinking, ‘What does this look like on someone’s phone? How many times will they have to scroll with their thumb to read the whole thing?’

Those types of things you really have to think about. Because – I don't know if you feel the same way – but if something doesn't fit in my screen I sometimes go, ‘Oh that's too much to read,’ right? Or if I'm constantly scrolling and switching it with my thumb, I might give up after three or four scrolls. What makes sense on a small screen?

What I have recommended to people a lot is get yourself what we call in Canada a cue card, or a recipe card. And just keep that next to you while you're working on your chatbot to remind yourself of the size of screen you're designing for.

Robin: That's a lovely idea, Jamie. Just to have that piece of paper next to you to sit there in the corner of your screen to see the real size, not this big huge screen.

Jamie: Because we work at our desktops, so we assume that everything that looks good on our desktops is going to look good on our phone too. It often doesn’t.

Writing for mobile experiences – being informal

Jamie: The other thing I recommend about scriptwriting that I forgot to mention too, is to think about being informal. When we text our friends and family we are not super formal. We don't put periods at the ends of our sentences. We often forget commas. We sometimes don't even put in apostrophes for words like don't and can't.

Think about that when you're writing your script too, because if you go at it thinking, ‘I’ve got to have the punctuation and the grammar and everything perfect,’ it's not going to come across like a real texting conversation. Even though people know they're interacting with a bot, you want them to feel like they're having a conversation with a person, to stay engaged.

Robin: Another thing in research shows that learning using informal, casual language actually is perceived a lot faster than more formal language. We have been trained that written language should be formal but in learning experiences language needs to be more informal. I imagine that's even more important with chatbots.

Jamie: The interesting thing is when I started building my chatbots a few years ago, it was still a thing where you had to consider the character count. I believe it was 160 characters I had to play with. So I purposefully removed punctuation at that point just because it took up too much real estate in terms of character limit.

Getting started with chatbots for L&D

Robin: I just realised we might have to make it clear to listeners that what you're talking about is using SMS for chatbots, not just web chatbots. Which is where some of the character restrictions come in. For the L&D person who is getting started in chatbots, what do you think would be a really good first thing to do?

Jamie: I think first, try one out. There are tons of them out there now. There are some that work well and some that don't. Woebot, like I said, is a really great example of a well-built chatbot. They incorporate emojis for part of the questions or responses. They give you opportunities for reflection. There's spaced out learning, there are retrieval opportunities. I think it's really, really well built. Another one that's super popular is called Poncho. It's basically just about the weather. But it's very fun. It uses GIFs. It uses kind of crazy humour. It’s irreverent sometimes. So I recommend that people really just try one.

Facebook Messenger is probably the largest platform for bots because there's just so many users. But you can also find them on Telegram. There are a few for Skype. There are some for Slack – there are lots on Slack. But I recommend people try one first so they have an idea of what it looks like and how it works. Especially with Woebot, see what it looks like to give people response options, as opposed to just letting people respond in any old way that they want. That's the first thing I'd recommend.

Robin: I think that's a pretty great bit of advice. How about trying to figure out when it's appropriate to use a chatbot? Have you got any thoughts on that? Is it just the times where you might be thinking that performance support might be the right solution, that a chatbot might be an option?

Getting started with chatbots and L&D – focus on performance supports

Jamie: I suggest getting started with performance support. Just because I think it's something important, and as I said before, I don't think we do it enough of it. But if you're doing performance support then you might have content that you can tweak and fit into a text-based chatbot solution. L&D is really good at building too much content. So I think it would be nice if someone took existing content and then started removing all of the unnecessary stuff and putting it down into a succinct text-based format. This would probably be a smart way to start.

Because the other thing about performance support and the chatbot is that you're getting people the support whenever they need it. A just-in-time, on demand, 24/7 kind of approach with a tool that's just in their pocket.

I think that would be a great way to start, to give people an experience of a chatbot and see how they feel about it. If you are telling people who have never really used a chatbot before that they're going to learn something completely new, in a completely new way, that may be a difficult sell. As opposed to saying, ‘You've already learned this thing, now I'm going to give you this really cool, convenient, mobile tool to be able to remember what you learned and put it into practice when you need it at the time of need.’