Real ROI from AI

Paul Nashawaty:
Hello, and thank you for joining us at the Six Five Summit: AI Unleashed. Welcome to the session, Delivering Practical Applications with AI. I’m Paul Nashawaty, Practice Lead for the Application Development Practice at the Futurum Group, and I’m excited to introduce you to Andy from Twilio. Andy, would you like to introduce yourself?

Andy O’Dower:
Hi, Paul. Thank you for having me. Yes, my name is Andy O’Dower. I’m the VP of Product for Voice and Video here at Twilio.

Paul Nashawaty:
Andy, welcome and it’s exciting to have you here. Twilio. Let’s tell the audience what’s Twilio and why should they care?

Andy O’Dower:
Twilio is a customer engagement platform founded about 15 years ago from the invention of what we call CPAS, communications platform as a service. That’s how the company started is abstracting away all of the complexity and fragmentation of telecom so customers can communicate via voice messaging, OTT channels, email, video, every channel that might arise where a business needs to communicate with consumers. Now, we’re a global company, over 300,000 customers, building on all of these channels, created this CPAS category. Through acquisition, acquired a company called Segment, which is key to our data strategy to power where we’re going is the next level of customer communications, where we see this blend of best-in-class, world-class, global-scale communications channels along with data to really power more personalized, more effective communications channels. Then you mix in all the advancements with AI, both AI in-house at Twilio, and also with partners, as that landscape changes to bring what we call a trifecta of data, communications, and AI together to move even beyond CPAS, where we then into new levels of customer engagement to be more effective for customers and consumers alike.

Yeah, it’s been a adventure. I was a long-time customer of Twilio for many years before joining Twilio on the product development side.

Paul Nashawaty:
Andy, it’s really great to have you here because was it last week or so I was in person with you at the Twilio executive event. Lots of activity going on there, lots going on with what developers are working on, the market efficiencies of what we’re seeing in the market. So there’s a lot happening there. But let’s jump right into that first question and think about how we want to answer this for the audience. When we’re looking at investments, and we look at AI at every layer of the stack, we’re seeing that we’re talking about AI to production and customer facing, particularly gen AI, and when things get real. How do you see a lot of this hesitation due to security and compliance?

Andy O’Dower:
There’s no shortage of amazing new updates of language models and developments in AI, as well as, like you said, investment up and down the stack, from hardware layer all the way up to SaaS apps and things that are even wrappers on top of LLMs that are getting massive amounts of funding. But there is the difference between that and real customers solving real problems out in the real world. And yeah, those areas around security compliance, those starting to get into new territory, especially with generative AI. You might see a case where… We have many customers, hundreds of customers, live with a variety of capabilities that we’ve offered, and we think that Twilio’s in a unique position where you think about “How do I apply AI to improve a customer consumer experience?” I can use generative right out of the gate. Maybe I can deflect some calls, and I can stay out of introducing all of my consumer’s data into the mix, all of their usage data. All of that type of data, that segment at Twilio has been primetime for activating that data to improve the outcomes.

You do get that apprehension of, “But then I unleash it into the wild, out in front of my customers. What might go wrong? What other data do I need to bring into this virtual agent, for example, to make it smarter and better and faster, but not introduce security risks?” So we’ve taken a couple approaches. One is transparency. We’ve launched what we call even nutrition facts. Quite literally took from a transparency construct that exists out in food of labeling, what is in my food, what is in my AI? What models is Twilio using? At what layers of the stack? What data am I sharing to those models at different layers of the stack to get the outcome that I want?

First things first is you have to have that transparency to be able to then get the trust of your customers to be able to not just engage at these early stages, but then to unleash it out into the wild so you can avoid those things that you see in headlines of hallucinations or leaking of customer data out to a bot that you didn’t want to leak out to a bot. Those types of things. So we think that transparency is key, and we’ve introduced a whole trust center and trust layer at Twilio, as well as… Because we’ve had many customers in large variety of industries and healthcare and financial services that even in communications, they need to trust Twilio with that. As we layer on the AI piece, we feel like that transparency of what’s under the hood, of what’s happening, is a thing that builds trust and at the same time controls and visibility and analytics and insights. So when you do unleash these out to end consumers, you can quickly have a feedback loop to improve those types of interactions and capabilities.

Paul Nashawaty:
I like the way you’re addressing how you’re helping the customers solve their problems. One of the things we’re seeing in our research is about nine months ago, I ran a study about using AI and production workloads, and we found that 18% of the respondents indicate that they were using AI in their production workloads. But then I re-ran the study nine months later.

Now, currently. And we see that 54% of organizations are running AI in their production workloads. So it is happening. It’s happening really fast, which is interesting to me because when I think about it, and actually from our briefing, we were talking about the impacts of using tools and acceleration of these app production and workloads. How are customers measuring their success?

Andy O’Dower:
It’s a variety of ways, I think. When you look at some of the automation categories, for example, many of our customers have, over the years, built up bots using predictive AI for handling all sorts of customer support. “How do I handle hundreds of thousands of customer support inquiries? And how do I do that without hiring another 200 people into a contact center?” For example. They’re looking at measurements of success in terms of resolving customer issues and support type areas and service type areas while maintaining high CSAT that customer satisfaction. That’s absolutely critical. It’s one thing to just throw a generative AI bot out there that’s wrapping a chat GPT, for example, and hoping for the best. It’s a whole other measure of success to be able to say, “Out of all of these types of inquiries that my business received, these that were more transactional, that could really be handled by a virtual agent, were handled successfully and CSAT was maintained or even increased. So we’ve seen many cases where customers are having those exact same results. That’s one way when they look at it.

Another way is, “I have all of this data and I know they’re in all the communications that I have back and forth across all channels. How do I turn that into actual structured data so I can learn from it, too, then learn how I might even optimize a virtual agent?” For example. So we have products like Voice Intelligence that we launched that will be in general availability here in a couple weeks, where we have hundreds of customers automatically securely recording, transcribing, doing language analysis on all of their calls. They’re turning that into very structured data that they can then use for analysis to improve an automated process, then know where their products might need improvements and where their competitors might get mentioned. So they’re getting that analysis served up on a silver platter, if you will, from previous mounds of data so that they can improve their processes in that way.

It falls into those make money side of the house and the save money side of the house. Both need to be secure, and we’re trying to look at that communications and unlocking that data, putting enhancements on top of it with AI, using an AI as a tool, so they can drive outcomes like that.

Paul Nashawaty:
Now that makes sense, and it sounds from the productivity gains and the savings, but also the differentiation of making money really does play into the measurement of success. One of the things that I found very interesting, and you touched on it just a little bit, I want double click a little bit more on it, is what AI solutions you have to offer that help with addressing these concerns? I know we had a lot of discussion last week, and I think that that would be great for the audience to hear.

Andy O’Dower:
Starting in things that we’ve been doing for quite some time on the segment side of things is analyzing all of this customer behavior and predicting who’s going to churn, who has propensity to do this activity, who are my highest LTV customers in these types of campaigns across all these channels that I sent might drive more ROI for my business. Those are the types of things that are typically in these type of CDP, customer data platform, offerings that we’ve been confident that segment’s been the best in the market at that. Then you look at the other side, on the communication side of the business, offerings in messaging or for things to just simply reducing fraud. Messaging is one that’s fraught with some of those issues and in pumping SMS volume and things like that that our customers don’t want have to pay for and shouldn’t have to pay for, so we’ve got AI in those areas to help you also be compliant in your messaging, but also introduce cost savings there.

Like I said, in Voice Intelligence, launching things with speech recognition and language understanding so you can really turn all these ephemeral calls and that you might have with all of your customers, usually at high lifetime value and high customer acquisition cost moments that then you can turn that into data for insights. Then the other things that we’re doing on virtual agent side are both with partnerships with OpenAI, with Google as well in the virtual agent category and language understanding and really conversational understanding that we have as well out there. The umbrella really is what we call customer AI, and that’s all of these insights activated across all of your channels to be able to be much more effective. You’re not just sending more messages or making more calls or taking more calls, you’re being a lot smarter about each one of them.

Because we look at this future as not just more, it’s more effective and those things that can delight end consumers and drive repeat business and things like that. So those are the offerings that we’re bringing under that umbrella and to, again, make them much more tangible and real. So you can bring in the three-horizons-out, really cool technology into the here and now. We’re also launching those types of things. We’ve launched an AI assistant builder within our alpha team as well, so you can extremely quickly get something up and running and bring in vector database and bring in all sorts of other assets to make that agent faster and smarter. So we’re putting things out in the market that are more “I can drive ROI today,” as well as things to show what’s to come over the years.

Paul Nashawaty:
Very cool. But one of the things we can touch on real quick here is… And you kind of double clicked down a little bit on this already, but there’s a lot of data out there. Twilio has data, customers have a ton of data, but generic AI produces generic results. With these large language models and whether it’s private in instantiations or public, how are you able to leverage all the data to basically create effective outcomes and decisions for your customers?

Andy O’Dower:
Yeah, that’s key. As we’ve seen all the generative AI pop up and yeah, these interesting… These big language models are really good, but when you say, “As a business, I need that to work for my customer segments with these problems, with these products or services, that’s what I really need to solve here today.” In many of these cases, they can also be small language models and smaller models that are more discreet and highly, highly trained. The missing piece though, is that personalized data of the customer and the consumer that you’re interacting with to actually resolve issues in a call or in a virtual agent chat that can also connect into systems to reschedule, reship. An application that can actually produce an end result of a task that an end consumer wants to accomplish. That’s why they contact you in the first place, and the personalized data is key.

We were fortunate to acquire Segment a few years ago, that leading customer data platform, and very publicly, we’ve talked about this merge of data and communications, so what we see is high volumes of customers that have been using Segment for many years for their customer data platform to activate it and bringing in data across every channel. From Facebook acquisition campaigns to website, mobile app behavior to support requests to any number of service, CRMs and everything else, data warehouses and everything else, to be able to activate that data, then really baking that into communications use cases that our 300,000 customers over the years already do, and make that connection really seamless for them to activate it versus having to do development work. That is very, very top of mind for us, to bring these things together.

All the way from interoperability with, you can call it legacy data warehouses, to zero copy, for security reasons, for large banks and financial institutions and healthcare to activate that, then to make it seamless with the communications. That’s really the unlock that we see is just AI alone or communications alone don’t really hit that mark. All of us selfishly, as consumers, want highly personalized experiences, so that’s what we really see as the value add is bringing those key pieces together. As the new models will come and go, we want those foundational elements to be active for our customers.

Paul Nashawaty:
Yeah, it makes a lot of sense. How would you advise the audience to get started?

Andy O’Dower:
To learn, go to Twilio.com and look at what we have. Like I said, we’ve got over 300,000 active customers, 10 million developers. We’ve really put the emphasis on the ability to get started and see that magic moment, that aha moment with Twilio, with just quickly and easily signing up and connecting these things and getting a proof of concept out in minutes rather than days and weeks. Coming into Twilio, you as a customer can automatically quickly try things then do a paper use model. And it’s usage-based, so you can get started and ease into it and then quickly scale. We’re the partner that you can scale globally with, as well, not just easily get started. So that’s where I’d encourage you to go, and you can learn a lot more about what we’re doing, the offerings that we have, both live right now and things that are coming. And you’ll see more and more from us over the coming months in this space.

Paul Nashawaty:
Great. Thank you, Andy. I want to thank you for your time today. I want to thank you for your perspective, and I know it’s just the tip of the iceberg here. Lots to think about, lots to consider for your organizations. I also want to thank the audience for attending our session today. There’s a lot of data here, a lot of information, and a lot to consider. Go to Twilio.com or TheFuturumGroup.com for additional information. Thank you very much, and have a great day.