By clicking “Accept All Cookies”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.

Get a demo

Build or buy AI? You’re asking the wrong question

Jul 31, 2020

Tonya Hall: Build or buy AI? That’s the wrong question. I’m Tonya Hall and joining me is Evan Kohn, Chief Business Officer at Pypestream. Welcome, Evan.

Evan Kohn: Hey, Tonya, thanks for having me.

TH: Of course. What are the problems that Pypestream helps their customers solve?

EK: Sure, well, we’ve seen a two-fold challenge for a lot of the largest corporations today and organizations where on one hand, a lot of their consumer base has really shifted in their expectations of how they engage with the brands in their lives. Away from waiting to hear back in days or navigating clunky web portals really preferring to engage via messaging the way that they engage with their friends and family. Now, for many organizations delivering on this always-on presence, making their organization really an always-on entity, the fastest path for them is really a combination of messaging and automation. But to achieve that, and over the last few years many Fortune 1000 companies have had to work with five, seven, ten different vendors that do a 6-12 month implementation, it really becomes a heavy lift in terms of a digital transformation initiative, to operationalize different parts of conversational AI. So, Pypestream was born out of this notion that consumers want to engage in a different way with brands and for brands to operationalize that they need an all-in-one cloud messaging solution. So that includes a web conversational interface, omnichannel offerings of where users can engage, a patented, secure messaging carrier that will relay data back and forth between businesses and consumers in an encrypted fashion, PCI, HIPAA compliant, the ability to escalate to live agents, assuming that automation isn’t going to tackle 100% of topics; so, finding that fine line of what should be automated and what should be escalated and how do you make that a seamless experience for the consumer. Also, plugging into a company’s backend systems via APIs, that allows for transactional conversational experiences as opposed to just navigation of content or question lookup. So, all those ingredients brought together along with a powerful AI engine that can really parse any user intent, understand emojis, sentiment, tone, those are what we’ve seen as the necessary ingredients to bring conversational AI into a manner where an organization is going to get real business value out of it – upwards of 30% increase in customer satisfaction and 90% cost savings compared to what they might be paying agents today to handle very basic calls like password resets or account management questions that today are costing companies $6-$15. So, this shift is really a no brainer for a lot of companies and we spend a lot of time helping executives think through what are the right use cases to automate so that they can stay ahead of this paradigm shift.

TH: Why is asking whether to build or buy your organization’s AI solution the wrong question?

EK: Yeah, you know, a lot of companies have been in a race to showcase innovation as – first different messaging platforms were emerging, maybe 5 years ago, in 2016 there was a real flooding of chatbots to the market where a lot of organizations parked a very basic chatbot on their site just to show that they were using that type of new technology. Unfortunately, regardless of whether there was actual utility there for end users. So, today now that so many capabilities can be brought together to deploy a personalized conversational AI at scale for customer service, marketing, and sales use cases, executives are thinking about it differently. Some are establishing centers of excellence within their organization around AI where they can take the latest advancements and really get the maximum business value out of them. So, rather than just seeing AI projects as technical playthings where they’re poking at capabilities in a sandbox, I’d say the most forward looking leaders are actually investing in resources to take advantage of AI initiatives from a design and customer experience standpoint and knowing where to draw the line between what they should do in house as an organization and where they should work with external vendors, startups, partners that all day, everyday are focused on, you know, one unique or several components that can really augment what they’re doing in their organization. So, out that friction is coming this question for a lot of executives should I build or buy AI, and I would argue that’s not the right question because it’s really a matter of what’s going to advance my product and service to best serve customers, so how do I become an AI capable organization instead of AI centric. So, AI capable meaning I have the resources and skill sets in my company in terms of analytics, in terms of design, to leverage the latest and greatest AI, but I’m not necessarily going to focus on rebuilding what’s already out there in terms of AI platforms, that requires really deep information security capabilities, DevOps capabilities – just to manage platform uptime. So, that’s where we, at Pypestream, spend a lot of time with companies, helping them find that balance between building a center of excellence and leveraging the capabilities of AI from partners.

TH: So, let’s talk about how you do it. How do you build the internal core team of experts who tailor the AI solution to fit the organizations? What skills, experience, and attributes do they need?

EK: Yeah, so even before an implementation begins a prerequisite exercise is really mapping out what are the most optimal use cases for both customers and the organization to start with, so, we’ll go through that exercise looking at, you know, what’s the volume of inquiries across different topics today that are maybe hitting their call center or email or social media care, what APIs do they have available that could actually be plugged into conversational AI for more transactional use cases, what are their businesses goals, is it revenue generation, is it cost containment, increasing net promoter score, or maybe all of the above including gathering actionable insights around user journeys so they can actually evolve their products and services. From there, once we have that North Star established of what’s the business behind a conversational AI deployment, then it’s really critical to do a kick off enroll the right stakeholders, have executive sponsors talk to what’s at stake with this initiative and if we don’t get it right what are the risks to the customer then, there’s huge opportunity in this so you want to have everybody aligned. And then the skill sets required include design thinking expertise, folks who can tap into empathy mapping exercises, user persona definition so that the design of a conversational AI is really solving for user pain points. On top of that API integration specialists, so this is the plug into backend systems, even homegrown legacy systems, into a really modern front end conversational interface to securely relay data around those more transactional use cases. On top of that, operational and business analysts to make sure that a deployment is really reflecting the best of the brand both in terms of compliance but also in showcasing all the capabilities and anticipating what customers might actually be looking for through this type of experience, whether it’s more customer service oriented or on the marketing side around promotional offers and sign ups. So, we actually created a certification program, there aren’t really many in the world yet, around how to deploy conversational AI at scale in a repeatable way that aligns the capabilities of AI technology with what are the business goals of an organization.

TH: What should organizations understand about building their data sets?

EK: Sure, well, there are a number of different ways to leverage data to increase what we call the precision score of AI, really the accuracy of it, so, an example: let’s say a rock hits your windshield and you need to get in touch with your auto insurance to file a claim, so you go to their website and you see they have a conversational interface there and they have maybe a few questions that you can just tap on that might be common questions but you have maybe a curveball one that you want to pose through free text entry like “hey a rock hit my windshield, I don’t know if I need to get it repaired or not, it’s just a little knick, does my policy cover that”, so to be able to parse the way the user, that policyholder, poses that question there’s a number of different elements there; but what we want to do is understand what geography they’re in, they may have a different way of posing that question, which we call an utterance, and they way to source that data could be a combination of historical phone or email logs or many of our customers have operationalized agent led live chat without automation. Those types of historical logs can be used to ingest into AI engines and by running different combinations of classifiers we’ll be able to get to a precision score, typically we’re targeting north of 90% accuracy before recommending a go-live to a client. Now, for those companies that may not have those historical logs that reflect the way their consumers or audience might pose certain questions, you can also crowdsource data quite securely through companies like Innodata or Amazon Mechanical Turk, to crowdsource utterances around specific topics. And a third approach we take with some clients is to go-live with agent messaging or very basic automation and use that first phase to collect more information from customers around the way they pose questions and then in a phase two really train for NLU classifiers so that precision score can go up, in a manner that executives can really trust in a predictable way, knowing that customers will be satisfied with really an effortless experience.

TH: Evan Kohn, Chief Business Officer at Pypestream. Thanks so much for joining us. If somebody wants to connect with you Evan, what’s the best way they can do that?

EK: Thanks, Tonya. They can go to or they can DM on Twitter, I’m @evankohn. Thanks for having me.

TH: Of course. And find more of my interviews right here or at

Originally published on ZDNet.[/vc_column_text][/vc_column][/vc_row]