AI-powered Virtual Agents: Buy or Build?
TechStyle Fashion group automates millions of calls and chats each year with conversational AI. The first step in their journey was deciding whether to tackle internally as a DIY project or outsource design & operation to a partner. Hear their experience dabbling with both approaches and how the considerations are different for each company.
What You Will Learn:
- TechStyle’s key considerations for “Buy vs Build?”
- The journey TechStyle took to omnichannel self-service
- What TechStyle measured in first-year savings
- How virtual agent CSAT compared against live agents
Director of GMS Technology
TechStyle Fashion Group
Chief Marketing Officer,
AI-Powered Virtual Agents – Buy vs. Build
Brian Morin: Well good afternoon everyone, my name is Brian Morin, and I’m Head of Marketing here at SmartAction, appreciate you joining today’s virtual happy hour. It’s interesting to see how creative everybody has been getting with all of these stay in place orders. And so we thought, “Well maybe one way of tackling this is just to send you a GrubHub, allow you to support a local restaurant.” And I hope everyone did have a chance to order your alcoholic or non-alcoholic drink for today’s 4 p.m. webinar.
With me today is Aarde Cosseboom, and Aarde he is the Senior Director of GMS Technology Analytics and Product from Techstyle Fashion Group. And Aarde, that’s quite a mouthful for the title, but we sure are glad to have you on for this conversation.
Aarde Cosseboom: Thanks for Brian, thanks for having me. And yeah I feel like it grows and grows each time I absorb a new team or as the years pass.
Brian Morin: Well it’ll be interesting to see where that’s at whenever we do this again a year from now. We did the same presentation a week ago to a couple of different audiences with the Customer Contact Central, and so it went over so well that we came back with an encore performance and doing a virtual happy hour.
If for any reason you did not receive your gift card, we do have a team on the edge of their seat willing to jump in and uncover status and, hopefully, we can get some delivered out to you. You just chime in on the Q&A box, or chat box and someone from our team will jump in.
And today we’re talking about AI powered virtual agents, the very first decision that any organization makes when going down that path to expand self service capabilities to conversational AI which is, “Should I build out my own internal team and tackle this as do yourself project, or the idea of just a buy where we can jump on a subscription platform and work and outsource all of that need to a partner who delivers the conversational AI technology stack and the NCX services?”
Aarde, this is obviously a path that Techstyle, at the very beginning of your journey, you had to make your decision on what path you’re going to take. You dabbled a little bit on both ends so you can see the pros and cons of where an organization might fit in either side of this.
I don’t think we want to call it an argument as much as, “What’s the best fit for an organization?” Because we on our side, as much as we operate conversationally for more than 100 brands, we look inside organization and can tell very quickly whether or not the best approach for them is outsourcing, or it may be simple enough that they handle as their own do-it-yourself project.
So what I do want to ask is … I’m just going to jump in and give you just a quick plug on who SmartAction is. And then I’ll throw it back at Aarde, so Aarde will tell you just a little bit about himself, his background, facts and figures about the Contact Center Techstyle and then we will jump into the conversation.
The only need to know on screen about us is that we provide AI powered virtual agents as a service. And that means that we bundle the AI technology stack together with all the end to end CX services needed to run it, and that means everything: the design, the build, the ongoing operation. Most customers of ours will start in Voice first, typically, just because that’s where the biggest ROI is then scaled digitally to chat or text so it is an omni channel solution.
And the precipice around this is the idea that conversations with machines aren’t easy. It’s a little different if you’re just trying to do something simple, but if you really are trying to do conversational AI, you’re capturing grammars and then intents, and in handling natural language, it’s not just design, build, and poof and you’re done.
And that’s why you do need a team of specialists committed to tuning the application week in and week out. It means poring through data analytics, listening to call recordings, chat transcripts, meeting with your team weekly on stats and objectives and constantly chase that frictionless experience.
And we integrate with every Contact Center platform out there: Genesys, Five9, NICE inContact, you name it. And we would like to think that this approach has been working for us, we do happen to be the top rated solution of Gartner Peer Insight as rated by our own customer reviews which is a nice plug to have.
So Aarde, I’m going to throw it over to you, but as I do when we jump into this conversation, I do just want to ask those listening in if you have any comment or question, make sure you type it into the chat box or the Q&A box, and we’ll tackle as we go along. We have set content, but it’s nice to have interaction and to be audience driven as much as possible. So Aarde, over to you. Won’t you introduce yourself to the audience?
Aarde Cosseboom: Yeah, thanks so much Brian. And I’ll speed through a lot of this content because I know we only have about 30 minutes, and we all want to get back to our normal lives, and enjoy the tasty beverages if we have those. And really the meat of the content comes in a couple slides here, and it’s really good content, I don’t want to skip over that I want to spend some time there.
But a little background of who I am, I’m Aarde Cosseboom, Brian gave me a great intro. My title is extremely long, but really GMS stands for Global Member Services. So what I focus on is customer service, customer success, and customer experience. And I focus, specifically, on the technology and product side of the house.
I do dabble in a little bit of the operation side, but I really leave that up to my peers and my colleagues. But what I focus on is how do I enable the businesses, specifically customer service driven businesses, to use technology and preferably to use technology for the better for self-service.
And then here’s just a little smattering of companies I used to work for. MindBody is where my origins were. Cornerstone On Demand, which is SAS product as well. And then I’m currently employed at Techstyle, I’ve been there for a couple years. And Techstyle’s a conglomerate of five different online e-retail brands.
Brian Morin: Sure and Aarde, won’t you tell us just a little bit about those?
Aarde Cosseboom: Yeah, so the five brands that we support here at Techstyle … We’re drilling into this a little bit more because it makes more sense when we get into the Q&A and how we use SmartAction and their intelligent virtual agents. Really we’re five different brands, so we’re not just a single brand, we’re a conglomerate of five different ones. We partner with celebrities, high profile celebrities like Rihanna, Kate Hudson, Kelly Rowland.
We’re online eCommerce retail; so shoes, fashion, athleisure wear which is extremely popular right now with our Fabletics brand, with everyone working from home and those stay at home orders. And what we do also have, on top of that, is a monthly membership model. So not only is it products that we sell, we also sell a membership model where people get curated hand selected items within their eCommerce experience when they’re shopping.
Brian Morin: I take it that’s the cue for me to change the slide?
Aarde Cosseboom: Yeah, a little background of what the contact volume is and where we are around the world: six million phone calls annually, we do three million chats annually, so we’re very phone heavy; that is transitioning that is changing.
On average, we have about a nine minute handle time for our agents. We have five million plus members across the globe, we serve customers across 12 different countries, seven different languages, and we support our members via phone chat and social primarily. We do email, but email is only for our non-North American members.
And then on the next slide, talks a little bit more about where we’re headquartered and how many people support our members. Today we’ve got a little bit more than 1,000 team members globally. Darker circles are really just where our headquarters are, the blue circles are where we’re located.
We are heavy on BPO outsourced agents who are tier one and tier two agents; of course, tier three can be escalated to our corporate offices. And the biggest footprint we have here is in the Philippines in the bottom right hand corner. You know 600-ish right now, but that peaks to about 900 when we get to Black Friday and Cyber Week.
And then a little bit more background of myself, just recently published a book … I say, “Recently.” It was a year ago last March. Published a book about enabling better service, which is a customer service contact center story of breaking the norm through creativity, and technology, and innovation. Available on Amazon, it has a lot of content.
It’s not a textbook, it’s not going to have the answers, it’s not going to have a lot of hard questions in there, but it is a little bit more of a conversation around what is technology, and how can we enable better service utilizing technology.
Brian Morin: Yeah, and Aarde, I know this has been a busy couple months for you as you’re speaking with the different communities Execs in the Know, Frost & Sullivan communities and a lot of what you have are communicating and talking about is captured in this book isn’t it?
Aarde Cosseboom: Yeah, so it’s a smattering of technology, there’s also work from home best practices, contingency business plans; all sorts of different areas of business all the way down to workforce management and people management as well too. But the first third of the book, so the first five chapters, are all about what we’re going to be talking today which is data devs technology and intelligent virtual agents.
Brian Morin: So well then let me go ahead and just set the table for that conversation. First and foremost, what’s in the forefront is why conversational AI for Techstyle in the first place, which is generally true for most organizations.
The biggest problem that you do have on the left side of the screen, as you can see, there’s a simple IVRs and chat bots, they just offer very little opportunity in the form of expansion in automation to be able to improve the efficiency within your organization, to improve your operational expenses, and be able to tackle a lot of those conversations that your live agents are handling.
If you look over on the right hand side of the screen, essentially, from a 50,000 foot level, what does conversational AI give you? Gives you the ability to not just handle the most simplest transactions, but it really does begin to allow you to move a lot deeper in those types of interactions that your live agents are handling if it’s transactional in nature.
If it doesn’t require complex critical thinking or human judgment, then it’s likely a great candidate for automation. If there is enough, what we would call, high volume repetitive conversations that are occurring that makes sense to build a solution to handle that conversation.
Aarde for a lot organizations who haven’t really gone down this journey yet, we see for their live agents that it creates a very mundane experience. We were first starting working with … And I’ll mention somebody adjacent to you, it’s a retail manufacturer, Electrolux. Their live agents were handling 35 product registration warranty calls a day, and the work was so mind numbing agents would literally be reading a magazine while doing a call. In fact, you have a story that you shared with me once about flying your CEO out to the Philippines, and having him sit in a customer service rep chair for a day. How did that go?
Aarde Cosseboom: Yeah, so it’s really, really hard to convey, especially to the executive level, the importance of not only customer experience when you’re initiating conversations around bots and conversational AI, but also how it’s going to solve a lot of the transactional conversations.
And, like you said, we flew out our CEO and a couple of our C levels, and what we did was we sat them down with agents and we said, “All right, listen to four hours of calls and tell us your feedback.” And it’s literally four hours of the exact same conversation, almost like it was scripted, but it wasn’t actually scripted, just the agents had said it so many times-
Brian Morin: Right.
Aarde Cosseboom: It was ingrained in the in their head. And we came out of that and met in a room and said, “You know what? We have to try to automate a lot of these conversations, because none of these things really carry a lot of complex weight to it, they really fall into this kind of simple transactional conversations.” But the important thing to note here and what I love about this slide is that things that we think are simple can actually start to become complex.
Brian Morin: Yeah, isn’t that the truth.
Aarde Cosseboom: Yeah and it truly is a conversation so instead of it being just a simple request and having a robot or an automation process that request, sometimes there’s follow up, sometimes there’s multiple requests back to back that you need to process and that’s what makes it complex. It may seem simple at the surface level, but it gets complex pretty quickly.
Brian Morin: Yeah, matter of fact if we have time even in this webinar, we have one I think it’s like a 90 second call example that you would think is otherwise a pretty simple transaction. But as you can see as it comes with humans and conversations, things can get complex pretty quick. And so which is just to know how we would be able to handle that.
Yeah, so when you are stepping over this world of conversational AI on the right, it might first start with the low hanging fruit of the most obvious the simplest transactions, but you can begin automating some very, very complex stuff as well, and this is where the symbiotic relationship between virtual agents and live agents really begins to emerge.
We automate a conversation between body shops and State Farm to approve work, and there are 17 back and forth turns in that conversation involves capturing things like long alphanumeric policy numbers as fast as somebody can say them, I mean that’s hard for a human to do.
So you can imagine, if there’s a 17 back and forth turn in a conversation, that presents a lot of different forks in the road for the conversation to go. But the question is, is the virtual agent going to try to handle every single exception that can occur within 17 back and forth turns? And the answer is no.
What we’ve done is we’ve designed the swim lane for the virtual agent that follows the happy path, will be called the widest path that most customers take where we can guarantee a great CX. And it just isn’t always worth the effort to automate for every possible exception, or in some cases, the data is available.
And here’s where that symbiotic relationship comes in, because anytime a call goes outside of the happy path designed for the virtual agent within those calls, will just get transferred to a live agent along with the data gathered up to that point, so that way they can finish the call.
So the idea being here is that AI isn’t trying to automate everything, but rather it’s about using it the right way to automate the widest path that most callers take, and you use your live agents for exceptions. And I think, Aarde, you designed that symbiotic relationship into almost every single call or chat that we automate where the virtual agent handles the transactional pieces, while transferring to a live agent when it goes outside those lines.
So let’s just go ahead and just jump right into this conversation. It’s why, Aarde, those have jumped in for the virtual happy hour, it’s why they’re here. Those that have yet to go down the path of expanding self service capabilities to conversational AI, from the outside it’s all kind of a great big black box.
But when trying to look down that journey, the very first fork in the road on that journey is the one right here that’s put in front by this slide. Some call it buy versus build; to be a little bit more descriptive what does that really mean? It means are you going to take a do-it-yourself approach where you’re hiring out your own internal team? Your right brain thinkers, your left brain thinkers to run the application to design, build, and manage that application over time?
Or are you on the right hand side of the screen, are you going to look to that skill set through a partner who delivers it as a service and, at the same time, is also delivering the technology? So it’s two very different ways of looking at it. And, Aarde, you and I, we’ve spoke at length about this, we’re trying to truncate this conversation just something bite sized for our audience to be able to consume.
When you with Techstyle were looking at this fork in the road in front of you, can you just help us understand some of the considerations that went into making the decision, and why in this case you’re working with us we’re an outsourcer and, clearly, you’re working with us. But you have taken those steps on trying some do yourself on your own. Can you just step us through some of your considerations on how you pieced together your path on what was right with Techstyle?
Aarde Cosseboom: Yeah this is great. We only have 30 minutes really, but we could talk about this for a-
Brian Morin: Yeah.
Aarde Cosseboom: Very long time even if you hold long sessions. But I’ll try to sum it up in as short as I possibly can. So, when we were going through the process, we were thinking of either building it ourselves, which our company is called Techstyle, and we do have a … we call it a fashion OS. We have an ecosystem of tools that we’ve built, it’s all home grown in house, and we’re very, very proud of that.
So this was one of those products that we could, potentially, build in house because we had the right know-how, we had the right potential talent, we had the right energy at that time to do it, and the reasoning behind it made a lot of sense.
Once we started to do a little bit of research, we realized that this project wasn’t as simple as we thought. The complexity isn’t, necessarily, on the tool set; that is a complexity of the technology that drives the machine learning and the capturing of natural language. And all of those many tool sets that you would have to put together and create an ecosystem around; so that is a complexity.
But the layer outside of that was the layer that was much harder for us, which was making sure that we had the engineering talent to not only implement and develop, but also make sure that we had people who are managing and monitoring this on an ongoing basis.
Things like this pandemic that just happened with COVID-19, we wouldn’t be able to pivot quickly if we had a DIY product that we built out. We can lean very heavily on our partnership with SmartAction, not only with the people resources from SmartAction, but also the technology that has already been built out and is running across multiple different of your customers, and passing that same information into its machine learning algorithms, and helping us output proper intent, and proper responses, and proper call flows.
So there’s a lot more complexity, at surface level it seems like, “Oh yeah, I’ll just look at Google Dialogflow or a Amazon Lex and plug it in, you’ll quickly realize that there’s a lot of maintenance and implementation that has to go into it. And then once you do implement it, it’s not a set and forget it, you’ve got to maintain it and make sure that it’s current.
Brian Morin: So matter of fact, you have a slide that you’ll share on just some of the limitations that you ran into when you’re at least trying to do chat on, I think, in Amazon or Dialog Boy; I forget which, and we’ll get to that in a minute.
But Aarde, one of the things that we see most often is organizations have a tendency to under spec the full time headcount and the talent gap across different disciplines that are actually needed to pull off conversational AI on a do-it-yourself approach.
Aarde Cosseboom: Yeah, absolutely. And we realized that really quickly that if we went with a partner, we’re really looking at six months, maybe eight months max of an implementation timeframe. If we’re going to do a DIY just for the version one would take at least a year, and then we would want a version two, or version three, or version four, and it would just be an ongoing project that would never end.
Brian Morin: Now, Aarde, we have seen some organizations successfully use a do-it-yourself approach that typically it’s whenever there may be only doing the most simplest forms of chat. Where they’re on a single channel, they’re not doing voice, they’re certainly a lot more simpler transactions, and they’re not needing as many bodies to run that application. But how would you counsel somebody that they’re looking through a little more complexity, particularly also not just doing chat but also doing voice, and really trying to do natural language intent capture?
Aarde Cosseboom: Yeah, this is a great question, and there’s kind of like three different areas that I try to coach people through. One it’s omni channel, so you need to make sure that when you’re building these experiences, that customers or members can have the same similar experience, or do the same self-service actions across different channels: chat, social, phone whatever channels you want to provide as opposed to just deploying to one channel, so that I could add an extra layer of complexity if you’re doing it yourself.
The other thing is that you don’t want to sacrifice the customer experience. If you’re going to … I hate to use the word cheap, because it’s not a pricing thing. But if you’re going to water down a self-service tool, and have it only do one or two intents, but you really have a catalog of intents that you could have within self-service, your customers are going to really understand that your customer service bot or tool is not going to service them, and they’re going to just opt out and say the word operator instantly. They’re not going to-
Brian Morin: Yeah.
Aarde Cosseboom: Even want to try with something like that. And then the last one, it’s really natural language, and this goes to customer experience as well too. And it’s making sure that you’re speaking to them with a bot that doesn’t sound like a robot, that doesn’t sound like press one press two press three options. It’s really open ended questions like, “How may I help you today?” And they could say whatever they want.
Or they could say … But it could ask them, “Would you like me to look up your account by email or phone?” And they don’t even have to say the word email or phone, they could just start saying their email or their phone number, and it would understand that much like a real human would understand that.
<Brian Morin: Sure. So matter of fact, we are going to share a slide on our side that really shows what of best of breed CX team looks like. And so if you’re kind of in that camp, you’re looking what would be required from a full time headcount perspective in order to support conversational AI in an organization? We’ll at least show you the team that we use that supports conversational AI across more than 100 brands.
But Aarde, this is probably the most important, and maybe even the first step, for someone to really get a good grasp of when they are starting to go down this road or conversationally AI. We’ve had this talk before where conversationally AI is not a product, it’s not something where you just design, build, and then you’re done and its first day is also its best day.
And I think, to some extent, some people can walk into conversational AI almost like they walk into their experience with an IVR; just design, build, and then boom you’re done; when in fact, it really couldn’t be further from the truth.
Because this is not a product that you can buy, it’s a solution because it requires ongoing care are feeding, that means day in and day out analyzing intents, expanding grammars, tweaking conversation flows, analyzing data to identify friction points, to listening to call recordings of those friction points, finding new data sources, or sometimes it’s about cleaning the data sources that exist to increase containment.
We oftentimes will try to work in prediction, and I think one of the calls that we may listen to, Aarde, from you it has a predictive quality in it. And then every week it’s, “Hey, what is the number one reason for why data transfer in any given interaction or hang up?” That then you address and then you rinse or repeat the same process all over again next week, because there’s always an opportunity to tweak the application.
So Aarde, was this a difficult sell for you internally, when you were first speaking with stakeholders to help bring everyone to the table that this is not something that’s design, built, and done but it requires a lot of care, feeding, and attention in order to be successful?
Aarde Cosseboom: Yeah, I love this infographic. Because when you’re trying to do-it-yourself or sometimes you’re blinded by just the first two: project management, and reporting analytics. You have deploy, set it, forget it. How did it perform?
But when you’re really thinking through this a little bit more in depth, then definitely you have to really think about training the system and fine tuning it so that it perfects its mapping. And then also how are you going to have upgrades and enhancements?
It’s not a tool that’s going to just sit there and be there forever, there’s going to be new technology that’s going to be needed to be infused into this process or this virtual agent. And then further on from that, it’s really the conversation.
So we call these call flows, but really it’s just how do you get a person from point A to point F? There’s B, C, D everything in between. It’s not just service them with A and then throw them in the middle of nowhere and try to service them with F; you’ve got to get them to that point, that’s a full conversation that goes back and forth. So yeah it’s a great infographic because sometimes you could be blinded by just trying to do the first one, or maybe the first two-
Brian Morin: Right.
Aarde Cosseboom: And you can’t really see the forest beyond the tree.
Brian Morin: Right. Well, so this next slide just gives just a little bit of insight into the type of team that it does take to operate conversational AI in order to operate it well. We have eight different functional disciplines spread out across eight teams. Each face represents its own team behind it, both AI disciplines and CX disciplines.
And the main reason why we just show this slide is because sometimes, again, from the outside looking in, this is a great big black box. And you think, “Hey may just be a couple computer whiz kids with a pizza box in the basement pulling this off.” When in fact it does take a whole team of highly skilled CX professionals to really pull out those frictionless experiences when we’re talking on the channel, when we’re talking voice, when we’re talking anything beyond just the simplest chat that may only have one or two intents that you’re trying to capture.
So Aarde this is a great slide here, where you actually broke down just some of the different difficulties or limitations that you ran into as you were working a little bit with Amazon and Google Dialogflow. I know most of this is related chat not voice, but perhaps you can help take us through this.
Aarde Cosseboom: Yeah, absolutely. And I’m not going to go through each one of these, I’m just going to maybe highlight one of them, or two of them because I already talked about some earlier. Also I know this deck’s going to be sent out later on the recording, so it’d be sent out later. So if you’re not screenshotting right now, you could always fast forward to that recording and then screenshot.
What we do is every six months we reassess our build versus buy conversation internally. Every single time we’ve had this QA assessment internally, we always lean towards continuing to partner specifically with our partner SmartAction, but continuing to partner with our partners in the space.
Amazon Lex and Google Dialogflow are some of the newer, up and coming, best in class toolsets with regards to natural language, voice, and predictive, and engagement. So here are some of the pain points that we found on the voice channel … Well the ones that are very specific that are near and dear to my heart is number three and number four.
Is that whatever we’re asking someone to fully say something out like an email address. It’s going to convert that at symbol, because over the voice channel they can’t say, “ACE at symbol Yahoo.” Or if they do, it’s going to convert to whatever words they said.
And then four is a great example of this as well too. So if you’re trying to do an alphanumeric capture, when you say, “One, two, three, four C as in cat.” It’s going to convert that into C as in cat as an actual word or phrase, and not recognize that it’s just supposed to record that C.
And SmartAction has built a lot of their voice tools to catch these cases and solve for these cases, and these are the kind of issues that we would have if we were to deploy with a build or DIY product which is, in our opinion, not good enough from a customer experience to be able to ever deploy; and not at this point at least.
Brian Morin: Well and not to mention the lack of the ability to go into customize and train against grammars and intents. We know that that ends up, ultimately, being the number one pain point. So Aarde, I like how you step to this when you talk to folks. Someone’s looking at going, “Hey what is the difference between a bot and an IVA?” So I know you get this one as you do a lot of talks, this is one where you talk about that differential line maybe you can step us through just a little bit of from your perspective.
Aarde Cosseboom: Yeah, and everyone has their own opinions on this, and I don’t want to squash anyone’s opinions; there’s definitely a spectrum. You’ve got bot on one side and IVA, which is intelligent virtual agent or assistant on the other end of the spectrum.
So when I think of the spectrum, when I’m talking about bots, I’m talking about something that’s going to do something for me. So it’s kind of like … And I apologize if your Alexa goes off right now, but if I tell Alexa to do an action, it’s going to do an action.
It’s not necessarily going to have a conversation with me, it’s not going to know what action I just did and then make a decision on what action to do next, and next, and next, and next, and next, and piggybacking. Some of the skills for Alexa’s can do that, but it’s not going to intelligently do that or proactively do that.
When I think of intelligent virtual agents, it’s tools that do that in more of a predictive way. So if you’re asking for something … For our example, how we use SmartAction if someone’s asking where’s their product, we’re going to tell them where their product is, “It’s on the FedEx, it’s going to be shipped to you in three days.”
But we may want to ask a follow-up question like, “Do you need this to be expedited? Does it need to get to you faster?” It’s that kind of predictive ability in that conversation versus just feeding back information.
Brian Morin: So the next slide that we have in front of us here was actually a call sample, a 90 second sample. I know that we’re at the bottom of the hour, so we’ll skip over it for now, but if during QA, if anybody wants us to you hit the play button, you at least can get an idea of what the system sounds like, and how somebody interacts with it. It’s also on our SmartAction YouTube channel.
And Aarde, we were just talking just a little bit about your journey. So why don’t we give, just as we’re wrapping up here, just talk just a little bit about your challenges, and why even went down the route of automation. And then we’ll talk about what results you’ve actually seen.
Aarde Cosseboom: Yeah, so for us, specifically, these were the challenges that we had and what we wanted to solve with this toolset. We had non-revenue generating calls that we wanted to self-serve, so things that were less valuable like the call content itself was less valuable to us.
We have huge monthly volumes and spikes, you can replace that word monthly with a daily, or annual spikes, and it’s the same problems that we all have. Just not being able to have the right amount of workforce to handle those spikes, so we needed some sort of self-service in place to help us with that.
And then the last one was we wanted to automate, but we didn’t want to sacrifice customer experience, and that was really important to us. So, we wanted to look for best in class natural language voice agents, as opposed to it being very robotic or something that’s a synthesized text to speech.
Brian Morin: And now just to give the audience idea of what is natural language automation, what is it doing for you at Techstyle. On screen you can see some of the functionality, you had some intelligent front door, when somebody calls instead of a press one it’s open ended, “How can I help you today?” And listening for intent so someone can get to what they need done very, very quickly.
If it’s intent that is falls under order management or account updates, somebody will just stay in self-serve, but if it is an intent intended for live agent, then it will get transferred there. Aarde, we’ll skip over the timeline since we’re a little short on time here, but I didn’t know if you want to share just a little bit about any of these chat examples very quick.
Aarde Cosseboom: Yep. And it was actually on the previous slide as well too, but I’ll mention on this slide. So this is an example of how we integrated it with InContact. We also integrated it with Genesys, so it’s not only a toolset that can be integrated into systems that you already have in place, but it’s also a toolset that can be put in front of your IVR from a telco perspective. So it could go to SmartAction then transfer to whatever IVR or ACD that you have.
The live chat example here, I’ll let you guys read them when we send them out. I think the coolest thing here is that if you see on the right hand side, right above the green spot where Erica said something, the virtual agent said, “Is that everything for today?” And Erica said, “No that’s it, thank you.” So she actually said a double negative there. It caught that and recognized that the thank you was saying that there is nothing else that can-
Brian Morin: Yeah-
Aarde Cosseboom: Be done.
Brian Morin: That was a good point.
Aarde Cosseboom: So it’s just something that you can see the system recognizing that most regular bots cannot. And then I’ll skip over this one, this is the exact same example but hosted in-
Brian Morin: Okay.
Aarde Cosseboom: PureCloud. And then same with this one, but this one is showing you that in the chat experiences, you can also send links as well too. So you want to make sure that you chop up the conversation bot so it’s in digestible chunks and then also you could have rich links as well. And then this is customer feedback. This is all the feedback that we got, just curated a couple of them from Twitter. The coolest one is that the middle one that says that she just talked to an automated service-
Brian Morin: That’s good.
Aarde Cosseboom: And thought it was better than actual people. So kind of interesting, not great to share with your agents, per se, but it’s great to hear that we’re performing just as well or better than live agents.
Brian Morin: And then just a high level on the results side, Aarde, then I’ll just share next steps for those that might want to engage and we’ll jump into Q&A.
Aarde Cosseboom: Yep 2018, we had $1.1 million in first year savings, and the cause of that was 18.5% increase in self-service containment. We also had a reduction of 45 seconds because all of the calls and chats that were transferred to our agents automatically had case screen pops, and dispositions because the intents were being passed over.
So a lot of the work that the agents had to do to collect the information like, “What’s your name and how may I help you?” That was already captured. And then we had a 92% member satisfaction survey score.
Brian Morin: Good. So just here on the very last slide, if you are interested in engagement, most people start by doing one or two things. One is requesting a demo, whatever space you’re in, we support customers across 12 verticals, so we’re likely already automating interactions for other clients that are in your space. So if you’d like to see what we’re doing for others with like interactions, let us know and we can share with you what we’re doing.
Or on the other side, just sitting down with you and offering a pre AI readiness assessment. Not every interaction is perfect for automation, but it is about finding which has the right qualities, where you can deliver a CX that is good or better than a live agent, and while making sure that there’s enough volume there to deliver that very, very quick and kind of almost no brainer ROI. So, you can see the email on screen email@example.com.
We have a couple of questions that have come in so, Aarde, let’s go ahead and tackle these. If you haven’t gotten it in, please get it into your chat or Q&A box and maybe by now, Aarde, we’re now a little over the bottom of the hour so maybe our audience is past drink one, and drink two or three, so we have some catching up to do. But this is from Chris he said, “Aarde, do you have specific tools to aid in mapping out to flow that you want to build?” I think that he’s particularly talking about the conversation flow.
Aarde Cosseboom: Yeah, this is a great question. When we were trying to do it ourselves, we were trying to map this out using chart flows, and it got pretty complex pretty fast. It wasn’t as simple as we had thought, there was a lot of forks in the road with regards to conversations. And then also there’s loops as well, so when someone gets down a certain path maybe instead of it forking and going to two potential outcomes, it may circle back to two or three steps, and then go through similar questions.
So to answer your question specifically, no we didn’t to begin with, we worked closely with SmartAction. SmartAction’s customers success and implementation team helped us build out our flows, walked us through all of the potential outcomes of each flow, and what toolsets they had to either repeat multiple times, or reiterate, or fail out to a live agent.
Once we got those completed in a joint collaborative effort with SmartAction, we then archived, we then made our own call flows and mapped them, and then saved them, and archived them for record keeping internally. But yeah, that was a very hard part of the process, and if we didn’t have experts help us with that, it would be a much longer and probably some failure points as well too, some un-successes.
Brian Morin: And so matter of fact, Aarde, you have a conversation flow designer at SmartAction who handles your account? I’m not sure who that is that SmartAction that handles your account, but they fall under our human experience activist team. And of course they’re the ones who are mapping out that conversation flow, we do it in Visio style using our tools. And then, Aarde, I’m not really even sure how that’s handed back and shared with your client side. You guys do a screen share, and then you dig into those new conversation flows?
Aarde Cosseboom: Yeah, it’s actually amazing. So we will either send … So we won’t update or change a flow, we’ll either send just an email with plain text or a screenshot of where we want to affect the existing flow, and then it goes into the customer success team, and then at SmartAction, and then you guys just send us a updated, a PDF diagram of what has changed.
We sign off on it, it gets put into a QA environment for us to test, and then we test. And once we sign off on it in that QA environment, then it goes live to production; so it’s extremely easy on our side. And I just have one person who spends a couple hours per week managing that relationship with SmartAction.
Brian Morin: Yeah so I was just going to ask that next piece on your side. Clearly the other pieces working with SmartAction, just what is that? You have one person who is dedicated as your facilitator of this client side relationship?
Aarde Cosseboom: Yeah, so I have two people on my team, one is just there for redundancy. But there’s two people on my team, one person’s the main point of contact. And he really only works on the SmartAction partnership, maybe a couple hours per week. There’s a 30 minute meeting that happens every week just for us touch base, just a standing call. And then there’s probably emails back and forth throughout the-
Brian Morin: Yeah.
Aarde Cosseboom: Week as well too, but not even half of a full time equivalent is needed.
Brian Morin: Gotcha. So here’s an interesting question, I like it. It’s, “What is the biggest lesson learned by new customers? In other words, what are the same things that most customers are surprised by or stumble with as they’re trying to get out the gate?”
Aarde, I know that you are a data point of one, just speaking to us about your experience with us, but I’m not sure if you can travel back in time and think on any of the biggest lessons that you learned when you’re going down this path with automation. And was there anything that you weren’t surprised by, or anything that made you stumble, that you had to circle the wagons to overcome?
Aarde Cosseboom: Yeah, when you’re in the early phases, I’d probably say it’s in the intent mapping and intent capture.
Brian Morin: Mm-hmm (affirmative).
Aarde Cosseboom: It’s also really easy to try to boil the ocean; you want to solve for every problem with this automation. Because it’s a really cool technology, it’s conversational, so it could do a lot, it’s very powerful; it’s very easy to get distracted by trying to do everything.
So I’d say start small, smaller, try to capture the intents that are the simplest and are the highest volume. And then if you find out that there’s another intent that is slightly bigger that you could tackle next, kind of do that in phases.
What we did was instead of having hundreds of intents and hundreds of different potential call flows, we really narrowed it down into four different larger buckets, and solved for those larger buckets. And then as those change or morph, we can modify and separate those into sub buckets or sub call flows. So that’s what I would say is don’t stumble upon trying to solve for everything. And we were surprised by that, because we thought we had enough energy to be able to solve for everything. We were quickly surprised that it’s probably easier just to think simple.
Brian Morin: Yeah, and I think for us what we see what folks run into, what is the surprise to them it’s understanding how iterative the process really is. Because, again, these are the conversation than for any interaction that we’re automating, I think that maybe the surprise is understanding that once you hit your prod and you’re live and getting interaction, at that point, is kind of, as Aarde mentioned, you really are only automating the widest path that most callers take where you can deliver a great CX.
And now for, particularly, the next six to nine months, you’re really rolling up your sleeves, you’re really fine tuning that application, you’re trying to train for your intents, expand intents, your AB testing, and you’re really trying to chase that that frictionless experience and it certainly is an iterative process.
And now you do get to a point with any interaction where you get to a point of diminishing returns. And when you’re at that point, that’s a great time to look at other areas where you can grow efficiency within your organization.
Question is, “Who owns the natural language learning, is it shared?” So this is a we take a supervised machine learning model on our side, and that is one of the things that come along with our service. We do train all the grammars to handle all the intents; that’s two different AI engines.
The first AI engine is just on the speech rec side, just to make sure that we can take sound, convert to waveforms and objects, and we have layered algorithms that can deliver a confidence score and what that means. And then it’s matter of fact of piecing it together from an NLP standpoint, your natural language processing, natural language understanding to make sure that now we’re also extrapolating the intent of what was said.
Aarde Cosseboom: Yeah, and to pigtail on that one a little bit more so. My impression is that on the SmartAction side, a lot of the natural language is shared across all of your customers.
Brian Morin: Correct.
Aarde Cosseboom: But specific to Techstyle, we have our own ecosystem where it’s also contained. So we’re not plugging and playing other people’s call flows, we have our own customized and branded call flows that we’ve designed so that it’s not necessarily shared with others, but shared within our internal brands because we have five different brands that we support.
Brian Morin: So, Aarde, here’s one for you, interesting it sounds like the virtual agent is almost like an outright replacement for your IVR?
Aarde Cosseboom: Yeah, I get this a lot. It’s not to be confused with replacing the IVR, although we have replaced the IVR with the, we call AI Anna, but it’s the SmartAction tool. You can have it in front of your IVR, so it can do natural language, it could do intent capture, it can do self-service. But then if you still need to pass it into a traditional IVR, press one, press two, press three, or into a queuing part of an IVR where it repeats back how long the wait is, and you’re next in line, and maybe a callback feature, you can do that.
So it can be one of either, or it can be a hybrid. And right now we have it as a hybrid, as opposed to a complete replacement of. I think, Brian, you guys have customers that have it as a complete replacement of, and you also have-
Brian Morin: So we do.
Aarde Cosseboom: It as complete-
Brian Morin: We do.
Aarde Cosseboom: And separate-
Brian Morin: We do.
Aarde Cosseboom: As well.
Brian Morin: All of those that you mentioned, because some will want to go into queuing before speaking with a live agent, so in that case we’re transferring. After we do the natural language, and intent capture, and authentication, we will route to the IVR which will then route to the live agent.
We have cases … And I would say, when we first came to market, most organizations stuck is behind the IVR. It might be a press two for, I don’t know, claims high volume interaction. And then it would get transferred to our system where we would handle a natural language and mimic live agent behavior; one example is AAA.
We handle the entirety of all their emergency roadside assistance; pretty complex interaction. They still have the IVR upfront, but I think it’s a press two, and then it’s transferred our system. But most … And I think 80% of our customers will have our AI powered virtual agent on the front end doing the natural language content capture. And then will either handled self-service or route to live agent, then they wonder why do they even need an IVR.
So there might be some things that somebody is doing within their IVR that’s automating very simple transaction. If that continues to work for you, fine we can route it there. So I think that the answer is, it depends.
Aarde Cosseboom: Yeah and something to note here, and we don’t necessarily do this but I’ve seen you guys do this, it was in one of your webinars a couple of weeks ago, you can also channel switch. So SmartAction can take it over voice, and then instead of passing back to a voice queuing IVR, instead it could help assist the member customer transition into an SMS or another rich chat, and then continue the conversation via an asynchronous conversation versus having to transfer back to phone.
Brian Morin: Correct. And matter of fact, those who are listening if they want to see what that channel deflecting service looks like, you can call into the National COVID-19 hotline, and we’re powering that. And you will get greeted by a special greeting, and it will ultimately end up transferring you out to our digitally SMS, and then to a rich web chat experience where you can have an interactive FAQ bot, or even triage through a set of questions; and that’s all powered by our AI. So lots more options on the table than ever before to serve customers.
Well, I’m looking here through comments and questions, I’m not seeing anything else that’s coming up so far. So Aarde, I appreciate that you hung in, looks like the whole audience has hung in here for almost the whole hour. If I knew it was going to run this long, we would need to have sent them GrubHub certificates for two drinks instead of one. Maybe we’ll have to remember that for next time. Aarde, already any closing remarks for us before we close out?
Aarde Cosseboom: No, just thank you very much, Brian, for hosting. Thank you SmartAction for hosting as well. I know there’s a lot that goes into these types of webinars. You guys make it seem fairly easy not only for the attendees, but also for myself as a panelist; so highly appreciate that.
And anyone who is an attendee has any questions, or wants a specific demo of what we have at Techstyle, not only reach out to Brian and the SmartAction Team, but feel free to reach out to me directly, I love answering people’s questions around AI.
Brian Morin: Yep. And finally I would just like to close out by encouraging everyone to take a look on screen. If you may recall, Aarde has a book that is on amazon.com, encourage everybody to get a hold of it, steps through what they’re doing from a technology side and also just to customer service best practice side. Lots you’ll find in it, it’s called Enable Better Service.
And for those who have joined in, we appreciate you for holding on. We’ve had fun and look forward to trying to do something creative while we’re still all stuck in this work from home environment. So we’ll try to come back with another idea and do something similar soon. So thank you for your time and we look forward to continuing the conversation, thanks Aarde.
Aarde Cosseboom: Thanks everyone.
Brian Morin: Thanks everyone.