#26 How to make automated AI a reality

#26 How to make automated AI a reality

#26 How to make automated AI a reality

Guest:

Evan Kohn

A digital marketing and CX leader, Evan drove Pypestream's growth from stealth concept in 2015 to the market leader in conversational AI today. He is the creator of PypePro℠, a deployment methodology used by Fortune 500 firms. Evan has appeared in Newsweek, The WSJ, TechCrunch, and Forbes.

With the persistence of the pandemic, users are spending the majority of their time at home and avoiding physical interaction. How do leading enterprises reach customers in this new paradigm? According to Evan Kohn, Chief Business Officer (CBO) at Pypestream, it is about hyper-personalization by adopting AI and automation into the customer journey.

Keywords:
AI, customer journey, automated AI, customer automation, messaging, conversational AI, hyper-personalization
Episode number:
26
Duration:
32:24
Date Published:
September 24, 2020

KRISTINA PODNAR, HOST: Welcome to the Power of Digital Policy. Today I have the pleasure of speaking with Evan Kohn, the chief brand officer and head of marketing at Pypestream. Evan, can you share a bit about just your background on how you landed in your current role at Pypestream?

EVAN:  Absolutely. So I've been with Pypestream since we were just at the idea stage back in 2015. But before that was in the management consulting world working with large enterprise clients around emerging technology defining their digital strategy, especially as so many new software as a service application has come to market over the last several years. I'm helping them navigate that market before that was in Silicon Valley startups leading growth, including for an education technology company called Rocketship, but back in 2015, when I met Richard Smullen, our CEO, was just really taken by this idea. This mission behind Pypestream of creating really always on brands. To adapt to the way that consumers want to engage with real companies in their lives through messaging and the latest capabilities of automation and AI, and that's what we're up to today with some of the world's leading companies.

KRISTINA: So tell me a bit more about your AI efforts, and what does it mean for the average consumer? Because as an average consumer, I see really ill-formatted emails are coming into my inbox. I see companies struggling to help me out through chat online. I'm seeing the basics still kind of you know, struggling. They're not getting it quite right. How do companies even start to go about AI like what does this really mean for me as a consumer and for where companies are today?

EVAN: Well, it starts with one of the problems of customer experience and customer service that lead frustrations for so many consumers, you know, you think about the 10 or 20 minutes of any given week that are most frustrating often it pertains the something around technology not working, or you know waiting on hold the customer service. So, you know, when brands think about how they can foster an experience that's actually delightful and effortless, and that moves away from you know, consumers needing to call a call center and waiting on hold or sending an email only to hear back days later. And what does it look like to adapt to those expectations of immediacy that consumers they bring to each engagement with the businesses in their lives? So AI and automation are today playing a really key part in facilitating those types of customer journeys. So across industries, we've seen executives champion different initiatives to bring automation AI into their customer experience programs and, in particular, through messaging channels, so I mentioned phone and email. Believe it or not, 9 out of 10 consumers, according to global consumer reports, prefer to engage for customer service through messaging similar to how they might text with friends and family because it's asynchronous that allows them to do it on their time. It's it doesn't require them to you know, drop everything and just focus on for example, a phone call, but automation and AI allow companies to foster these digital experiences at scale because oftentimes, you know, eight out of ten topics that consumers are contacting the company about a really the same 10 or 15 questions. No, it's I need to transfer bank account funds, or I want to change my TV subscription, or I need to file an insurance claim, and the repeatability of those topics allow for companies to leverage Ai and automation in a way to design really kick-ass a customer experiences that are predictable that are the include rich features from a user experience standpoint. And when we're talking about the messaging paradigm, you know, that includes anything from video to carousels with swiping to gifts and list pickers to make it for a really effortless experience. So artificial intelligence and they've been a lot of advancements, in particular, the last years are enabling companies to adapt to the ways that users consumers are posing questions in their own words and natural language and to foster really personalized experiences at scale.

KRISTINA: So this is great because you know, I love hyper-personalization. The concept really resonates with me. It cuts down on a lot of noise as a consumer but one of the things that I'm always concerned about, and I think a lot of folks are becoming more aware of is the trade-off in terms of giving up privacy. How do you see that balancing out with sort of the offering that we're seeing today and the hyper-personalization?

EVAN: Yeah, you know achieving hyper-personalization to date has not been an easy feat for a lot of big companies because it requires systems talking to each other, you know, what's going on in the back end between all the different enterprise technologies that a big company has implemented, where data is being transferred in a way that the end-user the consumer can be recognized who they are in authenticated fact fashion and allowing them to go beyond just the basics of frequently asked questions or content navigation actually getting to those transactional use cases where they can get something done, you know get to a resolution in a way that doesn't require them to then have to go call a call center or send an email. You know, that's one of the impetus behind Pypestream being born was that we recognize that problem for big companies. What's the critical path, the path of least resistance for them to get hyper-personalization, and that includes a flexible architecture that you can plug in a variety of APIs, including the homegrown and legacy back-end systems, and breathe new life into them to include authenticated end-user data. So we recognize who they are and can adapt to the journey based on their history and also their desires but to your point, a critical element that is keeping all this data safe in particular sensitive data as it pertains the finances or health or even just preferences around what products consumers might prefer. So I think you see a number of platforms today in the SaaS world that is using Ai and automation to achieve this degree of hyper-personalization that are simultaneously being required to comply with very stringent protocols. Between CCPA, GDPR, PCI from for transactions HIPAA in the healthcare space and also looking at the human factors of cybersecurity, recognizing that a lot of platform companies that some of the biggest risks are the way they operationalize from a human element how they managed data beyond just the encrypted elements and how that data is being fed between different systems. So the balancing act there really requires platform companies to establish things like security committees to regularly review what the threat landscape is, you know, getting very clear on network security requirements, making sure that any sensitive data that is included in automated customer journeys includes redaction by that. I mean if you were doing a transaction through messaging to buy a product t' and you're entering in your credit card number, that number should be masked with stars so that if it escalates to an agent or that conversation log goes to a back-end system to a database for an audit trail no one should be able to see that sensitive data. So being really smart about proactively knowing when a user might be entering in sensitive details, but having the operational means as well to navigate and all those compliance requirements.

KRISTINA: So, how do you most of your clients deal with that it? Do you expect to have your clients sort of having their digital governance framework in place understand all of these issues or do you help them navigate that privacy space, and does the platform itself help with you know, CCPA, GDPR, LGDP in Brazil for next year looking ahead?

EVAN: So, we recognize quite early that in order to earn business and partnership from the world's leading brands, we needed to invest from an engineering and product standpoint quite early in these compliance protocols. So we've built within our messaging infrastructure compliance and deep security and encryption at rest and in transit, but as a young company, you know, we and I think a lot of young SAAS companies are faced this is a challenge to work with large enterprises, often you have to fill out a 3000 their questionnaire to get through an infosec team. So we recruited, for example, some of the best in the brightest from Palo Alto Networks and security engineers in the SaaS world to make sure that you know, we're ready to improve the security and compliance capabilities of our platform. I think a lot of SaaS companies are recognizing that that's a key element to earning business. But we also as well advise our clients on what we think they need to be given that many even Fortune 500 companies are still brand new to implementing conversational AI. It's important that we also educate them over the course of implementation but also even before we initiate a kickoff for a specific project that we call out how we're going to manage a consumer data sensitively what compliance protocols we need to have in place before we even begin developing API integrations, for example for systems to talk to each other and ultimately authenticate users.

KRISTINA: So yesterday, I actually had a conversation with a CMO. It was interesting. I shared with him that I'd be speaking with you today and his point to me was that there's so much hype right now around AI, especially in marketing and customer care. Do you think that AI will reach a peak of sorts of inflated expectations, and if so, when do you see that happening?

EVAN: Well, we've seen over the last few years a widening spectrum in terms of capabilities. So around 2016-2017, there was a flooding of the market of basic chatbots. So lots of companies raced to put a little chat in the bottom right of their website with a few buttons some very simple automation primarily around frequently asked questions, not necessarily transactional use cases, but there's been an evolution since then, we're today there's a lot of differences between basic chatbot you can build in your dorm room in two hours and the other end of the spectrum what you're going to put in front of millions of consumers in a secure, scalable fashion that has high accuracy from an AI standpoint, where by that I mean can recognize a user's utterance the way they pose a question in their own words and that It may vary based on their geography, they send their demographics, getting to a point where 90% plus of the time that a user is posing a question. That's within the design scope of an automated solution. We can actually support them in navigating and effortless journey. So there's certainly a big data element there, and there are a variety of ways that we coach and work with our clients to tackle that piece; one is looking at phone and email logs with consumers I'm using that as training data to recognize the way that uses that consumers pros questions, but also crowdsourcing data securely. So looking at representative ways that consumers would pose questions on specific topics, but looking at that data also of how it's going to be used to optimize a solution over time, so because we have a private platform by that meeting our clients a data is only seen in used by then. It's not used to improve the performance of other solutions by other clients. That's a big difference between a private AI platform that businesses deploy customer experiences on versus, for example, social media platforms where the data is being used to help advertisers target consumers according to you know their potential preferences for similar products. So where the solution is deployed, how your sourcing the data are all really critical components making this promise of AI meaningful for consumers and not just turning into another frustrating customer service experience.

KRISTINA: And with so much concern right now around brand safety, I've seen quite a few large multinational step away from social media and the way that we saw them using it a few years ago, but brand safety has become such a hot issue. Do you see more clients moving sort of to the private platform as a result of that?

EVAN: There's a number of ways that we can help big companies and to navigate especially that transition from on-premise in a localized data centers toward cloud experiences and between Google Cloud, Azure, AWS there's a lot of flexibility today to navigate that journey in a way that respects consumer privacy and security elements. For example, you know, we actively tap into AWS' KMS key management system, meaning enterprises can even manage their own encryption keys. So if they're deploying AI customer experiences on a platform that's not their own, they can still do so in a trustworthy manner by knowing that they have access to encryption keys. The vendor is not the one that has the operational control of the end of the day. It can deploy experiences through virtual private clouds. And I think even as you see insurance, financial services, healthcare organizations now deploying these types of experiences on platforms that are hosted in the cloud, recognizing they can do so with a lot of flexibility to quickly adapt an agile format to consumer preferences, but not without compromise. They're not forced to compromise security standards. In fact, in many cases, it gives them even more stringent compliance to follow.

KRISTINA: Got it. A few moments to you brought up the issue of sort of these chatbots that could be created in a dorm room and sort of moving away from them. I did want to follow up and ask you, you know at this point, do you see chatbots as being dead or do they still have a place in the industry?

EVAN: The chatbot as a term has received a pretty negative connotation, now given the fact that we saw that flooding of the market a few years ago of low utility, single task bots. So conversational AI, which is still a pretty broad category, is proving that you can take a conversational experience and some of those components of the early days of chatbots, bring them really to the next level with actually deflecting phone calls, the call centers deflecting your emails in a way that delivers on that promise of immediacy and making brands always on so I would say that the components of chatbots two-three years ago that you know led a lot of companies to race to just put them on their page. That's dead now because they didn't provide utility. They weren't providing true customer. Or service in a way that delighted and users, and I think the most advanced conversational I had solutions today proved the there's a huge difference from what we saw a few years ago to what some of the world's biggest companies are achieving though.

KRISTINA: How do large brands deal with that, especially on the multinational level you were mentioning having the platform, you know, and the ability to do multilingual call centers handle different cultures? When we start to look at local markets, things get fragmented really quickly. A lot of times, you know, I've heard people say AI can't replace the local agent because there's just localization nuances that AI isn't going to understand. What's been your experience with it?

EVAN: This is a few degrees of what the requirements are for multilingual to bring experience to life for consumers, it's reliable and trustworthy. So some of the types of traditional chatbots the low utility ones that we were talking about a few minutes ago. Those can easily be deployed and in really any language because it's primarily a rules-based program that you've deployed. There's no real parsing of the way that the users posing a question. Now when you begin to introduce natural language understanding, so that's where we're really focused on the precision of the solution where it's able to determine the intent through computational linguistics of the way users pose a question. That's where a number of AI solution providers today are investing a lot of resources to unlock new language packets to tackle. Not only languages across regions, but also the intricacies geographically for how dialects will change the way that end-users will communicate through natural language. So, you know, we've served clients not only in the US but in Europe in German and Spanish in other languages and that localization pieces important not to you know, undermined it's a key part of implementation requires. The same type of user acceptance testing monitoring and analytics that that precision is staying high and if there's a lack of data either to crowdsource or historical data for training purposes, it's important to go first through deployment a conversational AI were you're collecting data around the way users are posing questions and then only in phase two activating natural language understanding in that language.

KRISTINA: What are the really big challenges that I see for multicultural, multilingual localization is this notion that we're becoming more and more fragmented around the internet. We have data localization laws in the EU through GDPR, China, Russia both have their own data localization requirements, you can't take data, at least personal sensitive data outside of the country. When we get to intent capturing personal intent around a user, how you deal with that from a platform perspective? Do you actually fragment the data? And if you do, do then worry about anonymization, pseudo anonymization? What does that look like?

EVAN: Well, it's important not to take a one-size-fits-all approach, and that really comes back to the implementation before anything is actually deployed. So you're going through design thinking exercises on user persona definition, empathy mapping, so taking some sample personas of end-users that are semi representative of the broader audience and really thinking hard about what are they saying, thinking, feeling, doing today? So, you know, pre-deployment of this new AI automation solution and what problems can we solve and how are we then going to use that to inform the design of this journey by that, I mean not only the AI training component but also the automated elements of the integrations and how different conversation flows they're offered up to the user based on how they're navigating a journey. So, you know, one key element of the AI that we've seen that's a real differentiating factor, I think compared to how a lot of other providers are navigating this, how AI can actually respond to the disposition of a user. So looking at sentiment in tone and even being able to read emojis to determine, you know, what's the sentiment and tone of the user so that the automation can adapt accordingly. So, for example, if a user is very frustrated they're experiencing an outage, let's say with their streaming media, streaming content provider, they are trying to figure out what's going on. We're going to want the automation to provide a really fast path to resolution, you know, cheat chat and humor, that's not going to be the right place where If someone just went through, let's say an e-commerce experience, just bought a skincare product that they're excited about. They have a we're be able to read a high sentiment of joy, the brand may look to have the automation teaup a promotional offer for another product. So, you know, those are examples to respond to a user's disposition that I think will see increasingly play into this concept of hyper-personalization where we go beyond just that the indicating a user knowing who they are and even what their preferences are but also adapting to that human element of you know, what's their emotional state right now and how can we best serve it

KRISTINA: And that I think actually spans beyond sort of the customer care the call center is right. It starts to really embed itself into products and services. I'm thinking specifically about my elderly father, who is 80 years old, has dementia Alzheimer's, and he's a member of AARP and AARP did this thing where they basically released a conversational but through AI to help seniors who are isolated, lonely trying to help them sort of overcome the loneliness that came with this pandemic and it was an interesting use case I thought, it went beyond just the traditional customer care into let's check in on you, but there's still this really heavy reliance on humans because yes, it is a conversational AI component, but they still have agents who are calling or checking up and sort of tapping folks on the shoulder. At what point do we become independent or is there always sort of the need for that human touch?

EVAN: Yeah. There's a myth out there that these types of solutions are really today only ready to serve Millennials or gen Z, you know, those that you may be in particularly digitally oriented, especially as we see the growth of companies like Lemonade that have deployed conversational AI to really bring in new generations to traditional products like those in insurance. But as we see, for example, the trends around telemedicine right now that's across demographics, across generations, the ability for digital conversational experiences to provide the new access, especially during COVID, to important things like patient care, I think is a good reminder that automation can really apply to a really wide variety of different topics and use cases, but it comes down what should be automated and what shouldn't, and we definitely find ourselves as we work with a variety of clients pushing back at times on what shouldn't be automated. We often go through an exercise with the executives and will whiteboard in what are the topics that consumers are contacting you about today. What APIs you have available where we can actually embed transactional elements to a customer journey, and let's use that to determine what are the use cases that we should automate out of the gate that will pay the highest evidence for consumers and for the brains and what shouldn't we automate that is really best still going to agents and a lot of companies going to in this race to adopt conversational AI don't pause to go through that really critical exercise to determine really where can automation in a play, can really tackle the most meaningful use cases.

KRISTINA: One of the things I'm hearing as you're speaking is there's an opportunity to automate aspects of work that historically have been basically managed by humans and to get some efficiencies out of that and also to redirect humans to more valuable tasks within the enterprise and so it's not so much displacement of workers as it is a retooling of workers. How are you saying enterprises deal with that?

EVAN: Well, oftentimes, brands are adopting conversational AI to fuel growth so very much not in the context of oh, this is a human replacement, but rather they have customer service representatives they are really skilled at communicating with consumers on the frontline and representing the very best of the brand, and they want to augment that with automation and not in just a one to one replacement but just taking out the first minute of interaction even I'm so that those upfront automation to identify: who's the user, what's their problem right now? Can we help them troubleshoot it? So it seems like this is a really curveball topic, this requires a human touch. That type of hybrid approach to automation and humans, I think, best empowers customer service agents to represent the brand for where their maximum value-added and we've seen that type of mindset in CX initiatives across insurance, e-commerce and also travel, industries that are challenged with business continuity planning, what does the next normal look like post COVID where if they get inundated with customer service volume. How can they use automation to prevent their users from having to wait two or three hours? Take an insurance company, as you know, we have hurricanes right now coming, unfortunately, through the Gulf of Mexico. We work with flood insurers who sometimes have 90% of their volume for an entire year concentrated and just a day or two because of natural disasters and want to position them in a way where policyholders, people who have a flooded basement or might have just lost their home and are trying to figure out you know what to do with this potential financial disarray in their lives having to wait on hold is really the last thing a brand wants to do to a consumer. So using automation to kind of get things started even before a representative can assist them can really go a long way but and showing consumers that the brand cares.

KRISTINA: I'm looking forward to this automation and conversational AI coming to absolutely every aspect of my life but still as a consumer, I tend to wonder about data and privacy, and I see that as a bigger challenge may be in the EU than in the US. At this moment, so just coming back full circle to that aspect. Do you think that consumers should fear for their data and privacy?

EVAN: Well, they should absolutely be mindful of how companies are using their data and should take comfort in the fact that there is a trend toward empowering them to have the right to be forgotten or, at minimum, to speak with the company about and have more clarity than just having to read, you know, 70 pages legal and how their data may or may not be used. So CCPA represents a first step in the US toward a model like GDPR. It's also led to some confusion among brands of what exactly does it mean but at minimum, you know, and how should it be interpreted but at minimum, it's bringing up the discussions within C-suites and at the executive level of big companies. How should we handle our data in a way that puts it's consumers in the driver's seat and respects their preferences of how we use or don't use their data. If you're engaging with the platform, that's free. You can probably assume that it's being used to some degree and just be mindful of the business model behind how that data might be used to retarget you, whether it's through display ads or to be fed other companies to advertise to you. So I'd caution any end-user, particularly around financial services or insurance or health care, to take a second, think about, you know, what data you're sharing is critical in are you on a secure platform.

KRISTINA: Those are great tips, especially for individuals as they look to leverage the conversational AI in their life, but also for businesses and big brands in terms of weighing the risks and opportunities that conversational AI brings today. Evan, thanks so much for taking the time out today to hang out and share insights. This has been really useful, and I appreciate you sharing your time with us.

You can reply to this podcast here: