Adventures in Conversational Commerce

One type of tool that Shopify has developed for entrepreneurs is messaging bots that behave like “virtual employees” – they autonomously handle customer service interactions, online marketing campaigns, and even make sales.

These very simple bots help overwhelmed entrepreneurs stay on top of their businesses without spending the time and money required to hire additional human staff.

In this talk, Elizabeth discusses the UX research and design work that goes into developing these “game-changing” employees for busy entrepreneurs, and shares lessons learned along the way about facilitating communication between businesses and their customers.

It’s only recently that brands started to engage with their customers on social media. It’s still a little weird but it’s how things are. What a time to be alive! 😉

Talking to brands is becoming known as Conversational Commerce.

Things have changed in a lot of ways, there was a time not very long ago that you generally had to physically go into a store to buy things. Physical shopping was conversational, you’d deal with sales staff who represented the business or brand. So it’s not truly a new thing to talk to brands online, it is in some ways natural.

As a UX researcher for Shopify Elizabeth does a lot of user testing in stores, talking to merchants. One of their channels is instant messaging, eg. facebook messenger.

Shopify use quite a few bots at the moment. Elizabeth will be talking about two of them:

  1. Kit: helps merchants market their products. Texts you suggestions and runs small campaigns for you. Entirely automated, no human intervention.
  2. Messenger Bot: helps merchants communicate and sell. There’s a “message us” button on stores; which starts in an automated browsing system. Real people from the merchant can jump in if they are available. Sales can be done entirely in Messenger.

In reality these are “baby bots” rather than proper bots with natural language processing (NLP).

Tips if you want to make a bot:

  1. Know your users
  2. Understand the medium the bot lives in
  3. Simplify the interactions
  4. Be thoughtful about the bot’s personality
  5. Solve problems, don’t create new ones

What does Shopify know about their users?

  • Merchants are busy, lonely, often not tech-savvy, #1 stuggle is marketing.
  • Customers: shopping online, don’t know shopify, may be skeptical of bots (don’t think they are any smarter or faster than humans).
  • Understanding users and their context helps make good choices with bots.

The medium

  • “The medium is the message” – a conversational medium like messenger should be conversational
  • Customers may open up more in a social medium; you can even learn from the emoji they use
  • People find Messenger feels very personal
  • People tend to say thanks more in Messenger than they do for automated emails
  • …so it’s nice if your bot knows how to acknowledge that! “Thanks” “Happy to help”

Simplicity

  • Simple purpose
  • Simple process

Kit helps with marketing, which isn’t actually simple; but it sticks with things that are relatively impersonal like running an ad campaign. Shopify push back on suggestions of trying to do full customer service because NLP isn’t really good enough for an authentic-feeling experience. Yet.

Balance being proactive with giving the user control. Give people options, but not so many options they get into decision paralysis.

Messenger – they focused on making sure it worked well out of the box for busy merchants. Use data to guide the decisions; keep things simple and set smart defaults.

Personality

Bots have no personality or culture unless you create it for them. Users will tend to project their ideas onto the bot regardless, eg. despite the gender-neutral name, most women think Kit’s male and most men think Kit’s female.

Keep it consistent – eg. don’t randomly change the tone from serious to informal. With Kit they have prototype conversations where they work through the way an exchange should go.

Because Kit is an assistant, it should behave like one. The principle became “collective success”, Kit’s a member of the merchant’s team. So Kit says things about “we” and “our” about sales and products.

Solo merchants really like the impression that there’s “someone there working with them”, and yes sometimes it gets a little weird. They also have had people send messages that really sound like they think Kit is a real person, although that is rare (most people are clear it’s a bot).

Watch out for lack of understanding from the user; negative reactions to the bot; context mismatch between the bot-world and the real-world. They had trouble with Kit occasionally talking about “our first sale” as the first interaction, ie. before Kit had done anything to help.

Solve problems, don’t create new ones

Bots are tools, they should solve problems.

Problems that can occur:

  • Not respecting the user’s time
  • Not respecting the user’s context

Merchants are busy! That includes configuring the bot, you can customise the messages sent by the Messenger bot but most merchants haven’t done it.

Don’t make the bot feel like the aggressive salesperson!

Sometimes the messenger bot replies too fast – it makes people feel pressured to keep going

Merchants often feel it would be too creepy to jump into a session and take over from the bot, because it reveals they’ve been watching the user.

If the user says “no” to Kit a few times, it will ask why and their feedback goes to the team.

(upbeat music) – This is my brother.

A couple months ago he had one of the happiest days of his life.

It wasn’t his wedding, or his bachelor party, which is what’s pictured here.

It wasn’t when he bought a house, it wasn’t the birth of his child.

One of the happiest days of his life was when he got this message in his Facebook Messenger inbox, inviting him to a private taco tasting with Taco Bell where he’d get to sample a whole bunch of new menu items and vote on which ones he liked the best, and he’d also get paid for this as well.

So, this idea of brands inserting ourselves into conversational spaces, even just like a year or two ago, would feel really really weird.

Honestly for me, if Taco Bell were in my messenger box, it would feel a bit weird for me, but my brother is really into it.

And by the way, when I was in Sydney, I tried to find Taco Bells and there weren’t any, I hope this reference lends it’s just a taco, should be a taco place.

But yeah, not okay, there aren’t any in Sydney. I also looked in Melbourne and there’s a place called Taco Bill all over the place.

Also not okay.

Anyway, so I find that brands more more these days are inserting ourselves in these places where we’re used to having conversations with our friends and family.

Sometimes they do it for very deliberate reasons like in the case of inviting my brother to the taco tasting. Sometimes you might notice, like on Twitter and stuff, brands just kind of inserting themselves in weirder ways. Here’s an example of Charmin, the toilet paper company, a guy over Twitter just to chat.

Probably a mistake but pretty funny.

On Twitter you might see brands even talking to each other as well.

So this happened right before the last time I gave this talk to wireless providers and this day it’s gone to this really weird Twitter argument.

That (laughs) was perfect timing for my last time doing this talk.

Here’s an example from my life that I thought was really funny.

I texted my friend and her car wrote back to me, which was pretty strange.

And so yeah, more and more brands are inserting ourselves in these conversational spaces, so what a time to be alive, I guess.

It’s never been easier to talk to our favourite brands, probably not something that you thought you’d want to do more of.

But I guess sometimes it can be convenient like, for example, if you’re talking to a friend about meeting up and you can order an Uber right in your conversation.

Or if you book a flight and you really opt in to get your notifications through text message, things like that.

So this idea of brands talking to us in these sorts of places is called conversational commerce, and the guy who coined that term, Chris Messina, said, “You and I will be talking to brands “and companies over Facebook Messenger, WhatsApp, “Telegram, Slack, and elsewhere before year’s end, “and will find it normal.” And this is something he said last year.

I don’t know if we necessarily find it normal just yet, but certainly are maybe a bit comfortable with it now than we were even just a year ago.

So I think the idea is that back in the day we used to actually go to stores to buy stuff, and we still do that now, it’s like not something that’s totally gone, and we go to stores and we’d have conversations with people about the things we are buying and we do our shopping that way.

And of course nowadays we do a lot of shopping online where maybe we don’t have these sorts of conversations quite as much.

And one of the product managers at Shopify who is working on one of the products I’ll talk to you about today has said that “Even though online shopping has suppressed “our natural commerce-conversation behaviours, “people were still finding ways to do it.

“It’s not something new that we have to teach people to do; “it’s something natural that we took away “with the rise of online shopping.” So this idea of speaking to brand representatives and stuff is not something that is totally foreign to people; it’s something we used to do quite often, now we don’t do so much, and there’s kind of ways in which this is sort of coming back.

Okay, at Shopify, we use conversational commerce to help merchants on our platform grow their businesses. And so I’ll tell you a bit about what Shopify is so that even makes sense.

So Shopify is an ecommerce platform.

Basically what that means is we provide like one sort of central place for you to keep track of everything related to running your business. So for example, if you wanted to sell jam that you make or something like that, then you could use Shopify to set up an online store for your jam, you could use our Point of Sale software to sell the jam at a retail location or at a cafe or something. You could also use a bunch of these other things we call sales channels to sell your jam in other places. So for example, you can sell your jam over Pinterest, or Amazon, or whatever, and one that we’ll be talking about today is Facebook Messenger.

And so my job at Shopify, I’m a UX researcher, and it’s really awesome.

I get to spend a lot of time going to people’s stores and hanging out with them and figuring out what’s working well for them and what’s not. So, I get to do cool stuff, like go to this pretty cool onesie store in Toronto and try stuff on and talk to them.

And I get to usability testing, and I get to do things like beta testing of new hardware, like for example, we’re releasing a new credit card reader and I get to go to people’s stores and watch them try to set it up and do that sort of thing. It’s really fun and I’m really excited to be going to a bunch of shops in Melbourne over the next couple of days and kind of do some of these interviews with people there. And so, at Shopify, yeah, we’re using conversational commerce to help our merchants grow their businesses, and the way we’re doing that is through two bots that we’ve developed that I’m gonna tell you about today.

and I’m going to kind of walk you through some of the lessons we’ve learned along the way of developing these two bots.

Okay, so just in general, like backstory, Shopify is kind of really into bots.

I don’t know if you guys use a lot of like Slack bots and stuff in your work, but we certainly do use a lot. For example, we have a division of our company that’s focused on getting really big brands to use our platform to sell and if someone lands a really big band, then they may get a little congratulations from our Justin Trudeau bot saying “I love you for getting that big brand “on the platform.” We also get lunch at work, which is really nice, and we can use Food Bot to rate our lunch and send comments.

We even have a bot for UX research that one of our other researchers developed so you can, if you book a research session, you can have it ping different Slack channels and say “Hey, we scheduled the session, ping this person “who set it up if you want to watch the session “and learn about running UX research.” So we’re kind of not foreign to the whole idea of bots in general.

And so, today I’ll talk to you about two bots that we’ve developed for external use.

They’re kind of like employees that you get to hire for your business if you have a business on Shopify. So the first one is called Kit.

I’ll run this little video.

So Kit is this bot, I suppose, who will text you throughout the day and help you set up ads to run on Facebook or Instagram or whatever for your business.

So through these kind of short little text message conversations that you might have with Kit, you can pick what product you want to advertise, set some basic parameters around your budget and what picture you want to use and that kind of stuff. And Kit will kind of go off, set up the ad, run it, and then get back to you at the end of the day and say, “Hey, you sold eight purses through this ad. “Do you want to run another one?” or something like that. And then our other bot is a little bit different. Oh sorry, this is the Kit team.

I’ve said that I would show them, they’re great. So Kit was actually a really small startup that we bought, probably when they were about a third of this size and they’ve kind of grown with us internally since then, this is really cool.

Okay, and so, we have a Facebook Messenger bot, which does not have a cool name like Kit, it’s just Messenger Bot.

And it’s not really for marketing so much, it’s more for helping our merchants communicate with their customers and sell to them.

So, as an example, here’s a website and you can see a little Message Us button in the bottom right corner, and if you were on this person’s store, you could tap that Message Us button and you could start an interaction like this. So, this is kind of fast and simplified but basically you’d be given some options related to shopping or communicating.

You could shop the store’s products, dig down through different options like sizes and colours and whatever, you could select a product if you wanted to, you could purchase the product right in Messenger, and then you’d get all of your order notifications and shipping stuff right in Messenger as well. So basically your whole interaction will be contained within Messenger.

At any time someone from the store could pop in and as like a human ask questions. They could say something like, “Hey, do you find everything you’re looking for?” or “Do you have any questions?” Or the customer themselves could just start typing stuff and someone from the store would see it eventually. Okay.

Oh, there’s the Messenger team.

I put myself as a little ghost ’cause actually what I’m talking about today I no longer work on, sad, but I used to be a part of this very important team. Okay, so the two bots, I’m gonna try to, hopefully, keep them straight throughout this talk, just like someone throw something at me if you’re tired of me jumping back and forth between them, they’re kind of hard to keep straight.

So, just to kind of lay it out, Kit is interactions between Shopify as a company and our merchants, so it’s us engaging in conversational commerce with our merchants. It’s for marketing and it’s for bot messages only. So Kit’s never a person, it’s always just a bot communicating with our merchants.

And Messenger is a little bit different.

So this is a case where we are enabling our merchants to have conversational commerce with their customers, right? So it’s between our merchants and their customers. And this could be bot messages or human messages, right? So people will go through this flow, it can be totally automated, but merchants can jump in and say things and the customer can bury them with questions if they want to.

Okay, so one thing that’s important to note about both of these, is these are really just baby bots.

Like I feel even uncomfortable calling them bots, there’s no Natural Language Processing, nothing really complicated going on, they’re really simple interactions.

It’s not that we couldn’t do that, we could try to add little bits of that sort of stuff in, but we’re just not really confident right now that that’s not gonna come across as creepy or weird, at least with the way we’re able to develop them right now, so maybe in the future. Okay, so for today I want to walk through a few different, I guess, areas that I’ve learned some stuff about by doing research with these things.

The first really basic is the idea of knowing your users and how that knowledge of your users can help you design better stuff.

The second is understanding the medium that your bot is interacting with people in, because that can influence the sorts of conversations that happen.

The third is to simplify the interactions between your bot and people as much as possible. We’ll talk about personality, which is the most fun part I think, and why it’s important to be thoughtful about that. And we’ll talk about making sure you use your bot to solve problems, like real problems that exist, and not create additional problems as well. Okay, so let’s start with the idea of knowing your users. Okay, so what do we know about Shopify users? We know a lot, which is kind of one of the reasons why I work there, we actually have 14 researchers, which is like a lot of researchers.

We know a lot of things about our users.

But today I was gonna focus on a few things. So, let’s first of all separate Shopify users into merchants and customers.

So merchants are the people who start their businesses on our platform, and customers are the people that buy stuff from their stores, right? And so, some things we know about our merchants. We know they are super busy.

I don’t think that I’ve ever talked to a merchant who isn’t like super stressed out ’cause often they’re solo people, they’re starting a business for the first time often, and they just have these to-do lists that never get smaller. And kind of related to this, they’re often lonely ’cause they’re usually the only person, at least at the beginning, working on their business, and so maybe they’d quit their job to start this by themselves, or maybe this is something they work on at night after work, and so they’re often kind of feeling a little bit isolated. They might not be super tech-savvy.

They have electricity, I put a candle there, it was the best emoji I could use for this. But they might not be as tech-savvy as you might assume. For example, like I’ve mentioned before about selling jam, we do have a lot of people on our platform who I would call kind of like hobbyists. They’re people who make things because they have a hobby, like making jam or whatever, and they might know a lot about making whatever they love to make, but not a lot about running a business, right.

And so usually a family member or a friend will be like “Hey, your jam is so good, “you should really sell it.” And they’ll be like, “I don’t know anything “about running a business but I guess I’ll try.” And they kind of try to bumble through it and they may not really be that business-savvy or tech-savvy.

And their number one struggle is definitely marketing. We’ve done a lot of research on this year after year and marketing is the number one problem.

Okay, so in terms of their customers we know that at least in the context of using bots, they’re gonna be shopping online.

We know that they don’t really know who Shopify is. Mostly just because we don’t really put ourselves front and centre.

So stores who run on Shopify, they might have a little thing at the very bottom of their footer that says “powered by Shopify”, but they might not and there might be no other way to know that Shopify powers the store.

So the customers might have no idea that Shopify is even a thing.

And they might be sceptical of bots.

I think there’re a lot of us who are kind of sceptical with how good an interaction with a bot would be. And we’ve actually done some research where we’ve surveyed customer and user sort of people and found that not only do they think that, they assume that bots are not gonna be as good as humans at answering questions, which is super reasonable I think, but also they don’t necessarily think that bots are gonna be any faster than humans are.

So they might just really think that interacting with a bot might be kind of a bummer compared to interacting with a person.

Okay, so how do we know this stuff? I kind of already mentioned this, but things like interviews and store visits, we host a tonne of events and meetups, so we’re really interacting with our merchants quite a lot. And then we have these other ways, that are maybe a little bit less face to face, of getting information.

I mentioned before that store visits are like the most awesome thing because it really lets you get into the user’s context. So I’ve mentioned before about merchants being super busy and it’s never more apparent to me than when I go to a store that’s like super cute, like hipstery boutique that looks gorgeous on the outside, and then you get in and you start talking to them and you go into the back room and like, there’s like of stacks of stuff falling over and like two dogs running around, and like everyone’s freaking out.

There’s often this illusion of everything’s put together and fine but really behind the scene everything’s usually really busy and crazy. Okay, so I’ve totally believe that understanding our users helps us make smarter decisions about how to design our bots, and hopefully throughout the presentation you’ll kind of see examples of that.

Okay, so let’s talk about understanding the medium that your bot lives in.

And so I’m gonna channel, I’m Canadian, I’m gonna channel Marshall McLuhan, Canadian philosopher, who famously said the medium is the message. And so, what I mean by that is that the medium in which you’re interacting with a bot in this case could influence, or even really define, what that interaction is going to be like.

So I’ll give you some examples of ways in which interacting over a conversational medium might be different than interacting over other media. So, as an example, this is something I heard from a merchant when I interviewed them.

They said, “We’ve learned a lot about our customers “when they share such personal stuff with us. “It’s even interesting to see which emojis they use.” And so this merchant had a pretty common experience compared to other ones that I’ve talked to, where basically they find that, compared to email communications or things like that that they have with customers, the conversations they’re having over Messenger, for example, are really really different, they’re a lot more personal and people reveal things about themselves, and they get to learn things about their customers that they didn’t know before.

So this merchant, for example, talked about how someone had bought something from his store over Messenger, up until that point the conversation totally automated, and then suddenly the person chimes in with a real message and they say, “I’m so excited to get these shoes, “I’ve been struggling with depression for a while “and this really gives me something to look forward to.” And the merchant, you know, to me he was like, “That was amazing because I never would have had “that level of interaction with that customer before, “like there’s no way they’d send me that over email, “and it gave me the opportunity to say “that I hope that they’re feeling better “and to send them a discount or something like that “for their next purchase.” This is an example.

I have a Shopify store, they really encourage all of us who want to to set up our own side businesses, and so, here’s what happened in one of the Messenger interactions.

So the first message is just totally automated, menu of stuff, and the person wrote this really nice message.

“Hey, guys! “Just dropping a message to let you guys know “I’m picking out a design for a T-shirt,” you know, blah-blah-blah, “I’ve been following you “since the beginning, I’m so glad “you have more good designs,” all that kind of stuff. And that really like touched me because otherwise, at my store totally online, I never would really have a way of interacting with a person like this.

And I thought it was really funny because he says things like “Hey, guys” and “you guys” and “someone” and it’s like just me. So it’s good to know that I come across as maybe being a bit of a bigger operation than I actually am.

So Erstwilder is an Australian brand who took part in a case study for us and they said that “Messenger is so one-on-one, it makes the whole “post-sales experience feel very intimate and human.” So compared to something like email, Messenger is giving them a more personal connection, I suppose, with their customers.

Another example also from Messenger.

Okay, so this is something another merchant said to me, “You’d never say thanks to an automated email “because you know it’s not coming from a human. “Messenger, even if it’s a bot, “feels more personal and human.” So you can imagine if you order something off of Amazon and you get that automated order confirmation from Amazon, you’re probably not gonna write back like “Thanks, Amazon!” or something like that to that email because you know that there is no person behind it, you know it’s just sent automated.

But a lot of our merchants are reporting that their customers are actually sending little messages like “Thanks!” and whatever to things like automated order confirmations because they feel like it’s a more conversational medium, I suppose. Here’s an example from my store again.

So, up to this point I had not had any interaction with this person, shipment’s on the way, the automated message gets sent out and they say “Nice” with a little sunglasses guy emoji and that gave me the opportunity to say “Hey, hope you like it.” An interaction that I probably would not have had otherwise. And because I keep mentioning my store I’m also gonna plug it.

So this is store.

If you’re into like psychology stuff, it’s like weird psychology theme, like mugs and whatever. This is the URL and there’s a 20% off code for people at Respond.

Alright.

Oh, and an example from Kit as well.

So people also interact with Kit in kind of funny ways, people say things like “I can’t help but want “to write back to Kit as if it was a person “even though I know it’s just an app.” So as an example, Kit says something to this person, they say “thanks kit”, Kit says “Happy to help.” The person did not need to say thanks Kit, they could just ignore the message.

So I believe that understanding the nuances of the medium your bot lives in will help you understand how users will interact with it. These interactions that happen over conversational media might be different than ones that happen over email or over letters or even in person.

Okay, so the third thing is simplifying the interactions. So there’s two different ways I want to talk about simplicity.

The first one is the idea of having a simple purpose for your bots, and then the other one is having a simple process for them.

So in terms of simple purpose, Kit is a good example. So Kit exists to help merchants with just one problem which is marketing, right? That’s kind of an oversimplification ’cause marketing is a lot of stuff, right, like I say marketing, but like marketing means emails and SEO and a whole bunch of other stuff.

But basically, I mean Kit’s been in development for, I don’t know, years and Kit still just focuses on marketing, because we want to make sure that we are really focusing on what we think are the merchant’s biggest problems and I mentioned before that a lot of our research has shown that marketing is their really really big problem. So the sorts of things that Kit can do, I just grabbed this from their website: writing ad messages, creating customer awareness, marketing on Instagram, retargeting visitors, sending thank you emails that you can put in like discount codes or whatever after someone buys something from you, and doing email marketing and that kind of stuff. And Michael Perry, who is the creator of Kit, the founder of the startup that we acquired a year and a half ago, said that “People have asked me “100 times over, When will you build Kit’s ability “to do customer service? “And I said not until Natural Language Processing “is at such a point that it can carry on “the conversation and feel very authentic.” And so, basically a lot of people bring up this idea of Kit being a really good vehicle, I suppose, for customer service, because Kit does such a good job with marketing, and Michael Perry’s view is that we’re just kind of not there yet.

And so, in the meantime, that’s where we developed Messenger, because it’s a way that we can try to automate some customer service without having to relate NLP just yet.

Okay.

So in terms of simple process, so let’s talk about Kit. When it comes to Kit, we try to ask ourselves where can we be proactive while giving the user enough control.

And so, what we mean by that is, if you’re a merchant and you want to create a Facebook ad and if you’re doing it yourself, you would probably spend a lot of time worrying about things like exactly which image to use, what font to use, how big it should be, what colour it should be, all these kind of little dimensions of your ad. And that’s certainly fine, I would be totally obsessed about that as well. But the idea of using Kit as to kind of off load some of those needy-greedy details and not have to worry about them so much.

And so, we have to do that in a way that merchants will start feel like they’re in control enough, that they’re not just getting some random ad, right, we want to make sure that they’re still getting something they’re satisfied with.

And so, one of the ways we can do that is by using data. So, for example, one of the many things we could have Kit ask someone if they’re trying to set up like a discount code ad, for example, is “Hey, do you want this discount code “to be one time use only or multiple use?” And that’s a totally reasonable question, that’s something that people can choose to do in other ways on Shopify.

But one thing we know from research that we’ve done is a huge proportion, I think it’s like 80% or 90% or something, of our discount codes that are created are one time use only; so once someone uses it once, they can’t use it again and again and again. And so, since we know that most people are choosing that option, we can decide to have Kit not ask that question and just make that assumption.

And it doesn’t mean that some small proportion of merchants might be like, “Oh man, I really wish “I could have made that multiple use.” But what it does give us is a simpler conversation so that the merchant, when they’re setting up the ad, has fewer steps to go through in order to create an ad that’s gonna be probably working pretty well for them. Okay, so for Messenger, we sure had to focus on making sure it worked really well out of the box for busy merchants. So I keep saying our merchants are super busy, they’re super busy.

And one thing that we found is that, when a new sales channel, like Facebook Messenger, is opened up to them and if it’s free, like Messenger is, then people will instal the channel and kind of be like “Okay cool, I did that, “now I’m gonna go off and work on something else “I have to do.” And so, because a lot of people are not spending a lot of time really even like reading up on what this sales channel is all about and making customizations and that kind of stuff, we had to make sure that the kind of the default messages and that kind of stuff worked really well for most people. So basically, if you’re gonna create a bot that you’re kind of unleashing for other people to use, you want to make sure that whatever the baseline settings are, are really kind of generic and gonna be useful for most people.

So you’ll notice, if you ever interact with a store that has the messenger channel installed, that the messages are really quite, I wouldn’t say they don’t have personality, but they’re quite generic, because you want to make sure that they’re gonna work okay for most businesses. And I’ll talk a little bit about customization towards the end of the presentation.

So I think that in general we kind of have a data-driven approach to simplicity.

I mentioned when we have like the simple purpose focusing only on problems that we know are real problems, so we really know the marketing is a big problem, so focusing on that for Kit is really important. Keeping the conversations simple.

So this idea of, if your bot’s gonna have to go through a multi-step process to set up an ad in this example, you want to take out as many of the steps as you can by making intelligent decisions based on what data you have.

And then setting smart defaults based on what you know about your users, and that was the last thing I talked about there. Okay.

So now my favourite part, which is personality. Yeah, so bots are weird.

You want your bot to have a nice personality that people are gonna react to nicely and it’s just kind of hard to even know it, like how do you start designing some thing’s personality. The thing about bots is like they don’t have any personality innately, right? Everything that they do you give to them, right? It’s important to make smart decisions about what those things should be.

And so, you want it to be consistent, for example, you don’t want your bot to interact in a very formal way in some situations and then be really casual in other situations, you want to make sure that your bot has a consistent voice.

And we actually have a person on our Kit team who was recently hired whose title is literally the voice of Kit.

So she’s the content strategist who is gonna be responsible for making sure that Kit has a consistent voice basically as we expand Kit’s skills.

And so, one of the ways that we do this is we actually start all of our new conversations that Kit’s gonna have by actually writing our conversations as humans back and forth with each other. So it’s kind of like we’re talking, we’re creating that conversation in which a human is talking to a bot using one person as the human and one person as the bot. And some of the things that we ask ourselves while this is going on are things like, “What approach would an assistant take to this problem?” ’cause Kit’s kind of like a marketing assistant, right? “What would you type to me if I asked you this questions?” or “What would a person do in this situation?” So it all goes back to this idea that even though kit is obviously not a person we want Kit to behave as much like a person as possible without being creepy and weird, and so we start with real conversations and real data like that.

And one of the most important aspects of Kit’s personality, I think, is this idea of collective success.

So collective success is the idea that Kit is a member of your team, Kit is your employee or your partner or whatever, and so when you succeed, Kit succeeds and when Kit succeeds, you succeed.

And so an example of that is here.

Here’s a message that Kit sent my friend Vivienne. “Congratulations Vivienne, we just made our very first sale! “Our first sale totaled $10.

“Happy day!” So, “we” just made our first sale, “our” first sale totaled $10.

This was an interesting and important choice that the Kit team made pretty early on and it wasn’t one without controversy, I suppose, and I guess the idea is that, because merchants are lonely and they’re often working on their own, that Kit would sort of act like the employee that they don’t have.

And so, it doesn’t always 100% of the time work well for all merchants, but probably 99% of our merchants are done with this idea of Kit being a part of the team. I think I have, yeah, I’ve got some examples. So I just kind of, I dug through the App Store reviews and found a bunch of interesting stuff.

“Each morning starts with a message “from my virtual marketeer, promptly at 9.01. “She’s never late and always eager “to get out there and market my business.” This is an App Store review of an app.

“It’s like having a virtual partner in my phone.” “As a solo maker and store owner Kit is great, “it does actually feel like I have someone working for me.” “Kit is the best employee on our team.” So it sucks to be anybody else on that guy’s team or that girl’s team.

“Great help, especially if you are just overloaded, “which is easy to happen.

“This little guy gets you out of the rut.” So people are generally responding really well to this idea that Kit is a part of your team, this idea of collective success.

Interestingly, so Kit importantly has a general neutral name so it could be a boy or a girl, right, and it’s interesting to see some people call Kit it, some people say “her” and some people say “him” and they’ll usually pick up and be consistent sort of throughout.

And I find, this is like totally non-scientific, but based on my reading of every single App Store review, it seems like women call Kit a guy and guys call Kit a girl.

I tend to call Kit a girl when I slip up and don’t say it ’cause I know that Kit’s not a person.

Oh yeah, things can get weirder too.

So ideally you don’t want people, I mean, of all the examples I’m gonna show you I don’t think anyone here really thinks that Kit is a person, but you can hear some interesting things like “Fortunately I found Kit, “or Kit found me.” Or like “I love her.” A lot of people say I love her, I love him. “Oh my God, I just want to give Kit a hug “every time I see the text.” We see that kind of stuff a lot too.

Or “Sorry hubby, but there’s a new man in town. “Kit is like having another person in my office “running my marketing.

“It’s as simple as texting hey Kit “and he’s there for me at my beck and call.” And so, again, like I’m very confident that none of these people actually think that Kit is a person, otherwise we would have a bit of a problem on our hands.

Here’s another, I love this example, so here’s another example from a customer service chat that someone had about Kit.

The merchant says, “AI future stuff, “I’ll wake up and Kit’s in bed with my wife.” And then the Shopify employee tries to be all business and he’s like, “It acts as a connection “between your Shopify store and your business page “on Facebook for that store, so it pushes products “from your Shopify store.” And the merchant is still in his little fantasy land, like “go back to sleep, I’ll take care of that.” (laughs) So we get those sorts of things sometimes, sometimes people are just kind of blown away, the extent to which conversations with Kit seem natural. Very very occasionally we do have situations in which people do actually think that Kit is a person. And sometimes, when we see those sorts of messages, we usually pass them around Slack and we’re like “ha ha, this is so cool” but then we’re also like “actually this isn’t so cool” ’cause we do want people to know they’re talking to a bot, right? So here’s an example of that.

“Hello Kit, I do have a social media manager “presently for the next three months.

“We are just beginning to promote on my social media. “I will see how things go for now.

“Thank you.

“Stay in touch at the end of three months.” So this person thought that Kit was maybe like a marketing intern or something like that from Shopify who is contacting her to help, and this person must have installed the app Kit but maybe they thought that the app was some sort of portal for talking to a real Shopify employee or something like that.

So this is alright, Kit stories, and we laugh about it and stuff, but we wouldn’t want this to be a consistent problem and luckily it’s not usually.

So some things to watch out for.

Yeah, lack of understanding on the part of the user whether they’re talking to a bot or not.

You definitely want to make sure people know what kind of thing they’re talking to.

Negative reactions to individual messages or overall personality.

So if you get a lot of negative reactions to a certain message or a lot of negative reactions to your bot’s kind of general tone, I suppose, then you would certainly want to revisit that. And then this one I didn’t really talk about, but context mismatch between the bot world and the real world.

So going back to this example with Vivienne, normally this sort of message goes over really well with merchants despite the whole “we” and “our” sort of thing.

But in Vivienne’s case, she’d installed the app but she hadn’t actually configured it to put up an ad. And so basically Vivienne got her first sale on her own through other means, but Kit was taking credit for that sale.

So in this case Vivienne was kind of like, “what the hell, Kit, you didn’t actually do anything.” And so making sure that the world that your bot lives in is completely in line with the real world, I suppose, where your merchant lives in this case, so that they don’t accidentally take credit for something they didn’t even do.

Okay, so last section.

So solving problems, not creating new ones. You know bots are really popular, everyone wants to make a bot.

It’s really easy to make bots, there’s a few different apps and stuff that you can use. And I think that companies often rush making a bot when maybe they don’t really need to.

This is not necessarily a big problem, but when that bot actually creates problems that didn’t exist before, then that’s not so cool. So like this idea that this bot is a tool and tools are usually good but if your tool, in this example, for example, has a really hard to understand manual then it might actually create problems in addition to the ones that it’s supposed to be solving. Okay, so the first type of problem I’m gonna talk about is not respecting the user’s time.

And so I talked a little bit really briefly about customization before with the messenger bot. Basically just a bit before I started on this team, the team was really excited about the idea of having merchants be able to customise the messages. So, you could imagine if you were a merchant whose store had kind of a very specific personality, like maybe it was kids’ clothes or something like that, maybe you wouldn’t want really kind of serious, kind of lacking in personality messages to be sent out through the messenger bot.

So maybe you’d want to go in and customise these messages. And so we were really excited to release this option for people who wanted it, this idea of being able to customise every single bot message that’s sent out. I guess the problem is, oh sorry, customizations, so the problem was that we forgot how busy our merchants are and so we released this, so excited, I go to do a round interviews with them, some big merchants in Toronto, and including some merchants that really do have a very distinct brand voice on their website, and I was like, “hey, have you checked out “the customizations” and they’re all like, “oh yeah, we saw that but we don’t have time for that.” Like basically the defaults that you guys have set are good enough and one day maybe I’ll get to this idea of customising my messages, but I’m already way behind on the other things that I know are gonna be more important for my business that I’m just not gonna bother with this right now.

And so, in this case we didn’t really respect our merchants’, I guess, jobs and the amount of time that they have and we released this feature, put a bunch of probably, well time for sure, and money probably into it, and then didn’t really get used that much.

Yeah, so our merchants are these really busy people who are usually barely keeping up with their to-do lists and we assumed that they had the time and the inclination to do this customization and they didn’t really.

Not respecting the user’s context.

So this is another example I’m gonna take from Messenger. Remember like when you’re shopping in a store and in a store that works on commission and you have a pushy sales person following your around and they’re like asking you like, “Oh, do you have everything need? “Oh, have you heard about this discount?” or whatever and you’re just kind of like, “Okay, I just kind of want you to get away.” We had this issue or we have this issue a little bit with Messenger where some of our customers who are interacting with Messenger have mentioned things about feeling like in a bot interaction versus just browsing an online store.

That someone is kind of following them around basically, ’cause when you send a message over Messenger the bot can respond right away, right? And so, you kind of maybe feel pressured to keep going down and digging into products when maybe if you’re shopping on someone’s online store, you wouldn’t really feel that pressure.

And we actually have heard from merchants as well that, because they’re able to see these interactions happen and they can choose if they want to jump in or not, they actually find it a little bit creepy, like they’re the creepy sales people following people around when they can see these interactions happen.

So this is something that we’re kind of working on, what our approach will be, I suppose, for how to change that.

But it’s interesting how the same kind of real life conversational commerce behaviour of being in a store and having someone follow you around is sort of existing in Messenger as well, because of the timing of the messages and things like that. And the final one, not respecting the user’s wishes. So you can imagine, what if you just had a bot that sucks. Luckily our bots don’t suck, but like sometimes people don’t like them and they want to turn them off, right? And so, it’s really important to, if someone like let’s say, for example, with Kit, if Kit asks someone if they want to do something and the person says no, let’s say that happens three or four times, then Kit will usually ask like, “Oh, can you tell me more about why you don’t want “to run this ad?” or something like that.

And we take that information in so that we can make sure that, first of all, we don’t serve this person the same sort of stuff ’cause they obviously don’t like it and it sucks for them. But also we can make better decisions about what sorts of ads we should have Kit create and whatever, we can use that data.

There’s also like what if your bot is good but maybe a little too good, like maybe a little too on it? So in the case of Kit, again, Kit could message you many times a day if you wanted to and have you set up all sorts of ads, and people love what Kit does but they don’t necessarily want to hear from Kit all the time.

And so one thing that we’re kind of constantly working on is things like whether Kit should message you a couple of times throughout the day with small things to respond to, or if people prefer to have all of their marketing stuff at the beginning of the day all at once and answer all the questions at once and kind of get it going.

So we want to make sure that we’re respecting people’s time and not creating additional problems. Okay.

So stuff to remember.

Know who your users are, it’ll help you make better decisions.

Understand what medium your bot is interacting in ’cause it’ll help you predict what sorts of interactions will happen.

Simplify those interactions as much as possible, take away questions that you really know what the answer is gonna be.

Be thoughtful about the personality that your bot is gonna have.

You are the only one who gets to decide what that personality is, bot doesn’t have it on its own, so pick a good one. And then make sure your bot solves these problems and doesn’t create new ones.

And hopefully, if you follow these sorts of, I don’t know, rules, guidelines, whatever, then you’ll end up with conversational commerce interactions that are delightful, like my brother getting invited to taco time, and not horrible. Just a couple of things really quickly at the end. The first is I recently, I guess a couple months ago, gave this talk, a version of this talk, a shorter version, at Interaction 17 in New York as part of a three person panel thing on bots, kind of like how we’re doing today.

And just a couple of weeks after that one of the other speakers was tragically murdered by her husband in a domestic violence incident. So this is from Melbourne, the domestic violence response centre.

If you feel like making donation, you can do that there and then there’s a 24/7 hotline if you need it. A lot of people have been talking about design systems today which is awesome and yesterday as well. Just a couple of weeks ago we released our public facing design system called Polaris. You can go to polaris.shopify.com and play around with it, it’s really cool.

We also have ux.shopify.com if you wanted to see what sorts of medium, posts and whatever that we’re writing about UX.

You can also follow us on Twitter and Instagram @shopifyUX.

And if you’re looking for a job, we’re hiring like everything.

This is not even all the jobs we’re hiring for so shopify.com/careers.

We are based in Canada, Canada’s really great, we sponsor visas and stuff.

We don’t have like an Australian office but, I don’t know, talk to me, we’ll work something out. Okay cool, thank you! (audience applauds) (upbeat music)