How Telecom-Grade AI Personalization Can Solve Streaming Retention with Hemant Soni, AI Architect

Why AI Value Gets Debated

Tim Rowe, State of Streaming 0:00

There is a lot of debate about whether or not AI is actually creating enterprise value. That's what today's conversation is about. Today we are meeting with our AI architect, Hemant Soni. Hemant has been a contributor since day one to State of Streaming. We chatted previously, what telco is looking at, how that ultimately ladders up to the streaming conversation. And today, we're looking from the telco perspective back into media, specifically how companies like T-Mobile and Comcast, how they are using AI to create a unified view of their customer profile, how they're using that to combat churn and improve retention. Today's conversation is not only packed with insight but actionable takeaways. Amon at the end of the episode, he describes exactly how to get started applying these frameworks for yourself. So, if this is something that you're trying to figure out, how do we combat churn? How do we improve retention? How do we create a unified view of our customer? How do we actually use AI? Well, this is the conversation for you. Enjoy.

Telecom Outsmarts Media On Personalization

Hemant Soni, AI Architect 1:31

Thanks, Tim, and uh really appreciate it. And love being on uh Straight of Streaming. This article that I was thinking about and writing, uh, specifically when I talk about AI personalization at scale and what media can learn from telecom parallels, I would say that this sounds sometimes counterintuitive at first because most of the people A little bit. Yeah. So most of the people think about personalization. They immediately think about media companies like Netflix recommending content or Spotify building playlists or new platforms, tailoring specific articles for you. But in this conversation, what I would like to highlight upon that how the industry that has actually mastered personalization at scale is not media, it's it's uh telecom out there. And telecom didn't build this capability out of curiosity or innovation for its own sake. They built it out because I would say it had no choice. It built it out because survival depended on it. And when industry is basically, I would say, forced to solve a problem under pressure, and I always say pressure is privilege, it doesn't improve incrementally and creates systems that are fundamentally more advanced. So media doesn't need to reinvent personalization. Yes, there are a lot of opportunities out there, which we can definitely talk about in around media space, especially about like generative AI in production, hybrid journalism, synthetic media avatars, content hypergrowth, and others. But the what what opportunities do we have to learn from telecom?

T-Mobile Signals And Predictive Retention

Tim Rowe, State of Streaming 3:05

And one of the things that really stood out to me was you had a data point about T-Mobile reacting to signal in kind of a mind-blistering time to T-Mobile reacting to signal in under 200 milliseconds. And just to describe the significance of that, what you found is that T-Mobile has essentially unified their customer stack. So they have a clear sightline into that customer journey, and they're able to react to signals that could indicate the customer leaving or maybe an upgrade opportunity. What did you find with T-Mobile?

Hemant Soni, AI Architect 3:40

So this is like considering I do a lot of research and going through a lot of media articles about it, this is this is all about moving from reactive to more of predictive. For a long time, customer experience in telecom in a lot of industries have been reactive. A customer would call to cancel, then they would comply to retain them, and then the billing issues would occur or they would would have to be addressed. So everything was built around responding after something went wrong. But AI out here has completely changed that model. Today, telecom companies can detect early signals, subtle changes in behaviors that can indicate dissatisfaction, maybe a usage drop, maybe an engagement pattern, maybe a sentiment shift. And instead of waiting for customers to take action, systems or these AI intervened applications can provide or recommend early intervention. That's what I was reading to you while being a great example out there. Their AI systems process customers context, I believe in milliseconds. That's what I was looking into. And this actions could be tailored or plan adjustments or even proactive outreach can be done. Imagine the results uh being powerful, like customers don't feel like they are being retained, they feel like they are being understood. Companies companies with media, again, same thing, streaming platforms, they can collect similar type of signals, what the user watches, what journey they skip, when they disengage and then respond often to delays, like emails come days after, or campaign coming weeks later, and then by by that time the movement has passed.

Tim Rowe, State of Streaming 5:21

So I canceled, I moved on.

Hemant Soni, AI Architect 5:23

Exactly, exactly. So the fundamental shift out here is reacting to the behavior, predicting the intent, and that's the game changer.

Comcast And The Convergence Advantage

Tim Rowe, State of Streaming 5:33

And Comcast is an interesting company. We just had a conversation with Seth Mitman from Ampersand, who represents Comcast, Cox, Spectrum from an ad sales standpoint. And he brought up a similar point of we know more about the household. So think about a business like Comcast that has the infrastructure, that has the streaming footprint, that has broadcast all of these elements. What do you see for a company like Comcast? It seems like they would be well positioned to understand and interpret those signals and respond in real time.

Hemant Soni, AI Architect 6:08

I think one of the lessons in the article I do talk about Comcast and the convergence advantage. The intent was that Comcast operates pretty much in both the worlds. It's a telecom provider as well as on the other side, it's a media company. So it's understand the business, and the combination kind of creates a powerful advantage out there. If I take a step back, Comcast had spent years building um AI-driven systems for telecom, managing millions of customers, optimizing retention strategies, personalizing experience. And what it's doing is now is not just now it is applying the same systems to its streaming site. That's what my assumption and belief is. And so instead of experimenting, Comcast is deploying those proven intelligences. It's using data across services, integrating customer journeys across platforms, it's not just engagements, but the entire life cycle out there from acquisitions to retention to upsell. This is how the overall convergence practice looks like as of as on date.

Tim Rowe, State of Streaming 7:12

It's the thing we spend the most money on, acquiring customers. So if it could have dual purpose of also creating signal that we can collect to make the feedback loop smarter, to be able to respond to things in more real time. Obviously, that seems like a noble goal. We're investing heavily in all of these things.

Four Pillars Of AI Personalization

Tim Rowe, State of Streaming 7:32

Let's talk about the four pillars. You you identified kind of four themes across AI personalization. Could you could you take us through those? I think you touched on the shift from reactive to proactive or predictive. Yeah. What are the other lessons?

Hemant Soni, AI Architect 7:48

So quickly, I if if I number them and I walk you through in terms of telecom and how what results look like and what takeaways would be, is the first one, as I mentioned, is shift from reactive to predictive, which I was referring to earlier as like T-mobile deploying a pegger, customer decision hub, identifying customer context under, I would say somewhere around 200 milliseconds. Again, incredible. Give and take it out there.

Tim Rowe, State of Streaming 8:12

Now and that's across 140 plus million US subscribers. Just to give some scale to this, we were looking at the Comcast figures, and that's north of 40, somewhere between 40 and 50 million. That's what I believe, though. Pretty significant scale here. We're talking about.

Hemant Soni, AI Architect 8:27

Yeah. I think from a result perspective that I was reading in that article was lowest postpaid phone churn in the US wireless. I think it was somewhere around Q2 of 2023 or so. And the net promoter score going up eight points. And again, I feel like people well knows me as a customer when they help understand me. And again, from a media takeaway, if I talk about is act on disengagement signals, as I was mentioning, at the individual interaction level, not just at a campaign level. And that's that's what the the takeaway for from a shift to reactive to productive is. The next one, the key lesson is centered around the unified intelligence, replacing a lot of siloed data, or other way something I talk about is data democratization. And I was looking at like how would a phone had built up a centralized AI intelligence layer using Azure, OpenAI. And I think uh there were other examples of handling about Super Toby, it handles like 45 million customers across 13 countries and 15 languages. I talk about that in that article. And I've if I talk about the results out there, again, it is the it was it had a high resolution rate. There was also, I think, a 50% jump in customer.

Tim Rowe, State of Streaming 9:42

50%, a 50% improvement in customer satisfaction scores.

Hemant Soni, AI Architect 9:46

Yeah, yeah. And then it's huge. Yeah, and again, I would say seamless context across apps. Now, what media can learn again, same thing from this unified intelligence, is that unify a fragmented audience data into like one layer. That's that's the first thing. And that's where I would say personalization improves exponentially and and not just incrementally. So that's the key takeaway that I was looking into that. The third one, which was I was looking into Comcast bridging as a telecom and media company, and I think Comcast platform was XLR8.

Tim Rowe, State of Streaming 10:22

XLR8, yeah.

Hemant Soni, AI Architect 10:23

Which yeah, which runs retention.

Tim Rowe, State of Streaming 10:24

It'd be a good license plate, XLR8, the Comcast XLR8.

Hemant Soni, AI Architect 10:30

Correct. Yeah. And then this was mentioning about full lifecycle of customers like across video, AI, and there are others like data beeps applied to Peacock. And that's what I was reading into that personalization, redefining across 32 million broadband customers, ported directly into streaming. That's that's an advantage out here. No media company has a foundation, but they have always have an advantage to basically something it was talking about earlier in terms of how to use generative AI in production, or or are there opportunities in terms of synthetic data or avatars? These are like looking into how media can utilize those AI opportunities out there, more like content creation, it seems like the conversation for AI and media.

How Media Teams Start Using AI

Tim Rowe, State of Streaming 11:17

Let's break it down. You're the AI architect. If I'm a media company, how do I start? Where do I start? I want to, I I hear you, I want to combat churn and I want to use AI to do it. Where do I start?

Hemant Soni, AI Architect 11:30

First, you would need to identify that which segment of that churn is coming from. And and again, even before that, you would need to you need to understand what kind of media company you are. You are into voice, like there, there are certain media companies which are centered around only on voice. So from think from a perspective like Cirus XM. Okay. So different different media organization or Netflix. It it works into more of a video content as well as gaming, and they they are trying trying to widen their their net out there. I don't know, like most of the people that look into like going into Netflix, playing, watching videos or web series, but how about playing games, being being interactive or there?

Tim Rowe, State of Streaming 12:11

Oh, yeah, we've been doing that a lot in my house. My son loves play, we play Pictionary, and also it teaches me how differently we communicate. We've been having a ton of fun on the Netflix games, and it seems like there's obviously a lot of conversation, there's significant investments being made in all of these different types of media. So, okay, step one is understand the types of media that you deliver, the types of media that your customers, your users are interacting with. All right, then what? What's next?

Hemant Soni, AI Architect 12:42

Yeah. So I would say the first perspective would be to understand what your organization is, then build up a better unified data foundation. Who's your viewers? Are what's your subscription data? What is an engagement, engagement pattern out there? Now, data is important. We need to clean it, connect it, it needs to be reusable. Then we need to look into other areas in terms of what AI platforms we are looking at, especially around the cloud AI platforms. Are we looking for Google Gemini or or there are others out there? What kind of recommendation engines are we looking forward to? I mean, I'm getting into technology, but again, speech to text, translational API, generative AI tools, those would be steps. And then think about batch to real-time analysis, analyzing data layers, act through campaigns. Start small. Uh, that's how I always recommend uh real-type recommendations on like homepage, trigger notification based on behavior. Again, that's that's where these will evolve into like real-time churn prevention or dynamic pricing offers. One thing I always believe, and if if if if I'm consulting a company, I will say start personalization early. I believe is the highest ROI use case.

Tim Rowe, State of Streaming 13:56

Why is that?

Hemant Soni, AI Architect 13:57

Yeah, so the reason is simple thing personalized recommendation, personalization on homepage, content ranking, personalization of offers and pricing, because something what works for me, like a six for you is uh is a could be a nine for me. That's how I would I would look for it. And like if going onto a Netflix, I prefer a Netflix version, which is with added gives me an advantage. I can step away, grab some snacks, come back, and continue with my movie. It gives me a break. My Apple Watch tells me that hey, step up. Yeah, I've been sitting down for the last 30 minutes or so. But some for someone they would want to binge watch the whole series without an advertisement content. That's one area. Then the other one could be like, okay, what kind of genre are you watching? Or something you and I were talking about is somebody who's more interested in watching movies across languages. As was mentioned, like Money Heist was based in Spanish and English, but eventually, with the overall success it had, it got translated into different languages. And and something which you have which I was talking to you about is overcoming language barriers, advanced NLPs will allow media companies to localize content and instantly break global language barriers. That's that's an advantage that AI will give to them. And then it's incredible. Yeah, and then that's that's another thing which which is audio subtitling, AI dubbing, language translation, these comes into picture. That's where I was referring to you in terms of content hypergrowth, like AI enabling uh smaller teams to compete with larger studios because the the servers of content will be there. Then AI also, these are like different use cases I'm talking about in different areas. Like when you talk about journalism, like newsrooms will be human-led, machine assisted, where AI helps gather, analyze, draft content, while journalists focus on verification, context, and overall oversight out there. Understand what success means to you, measure business impacts, not just technology, like increase in watch time, reduction in churn. So a 90-day starter plan could be like month one, could be identified use cases, clean content data. The second month you can go on and build upon AI models to start testing, do some small palettes, measure them, then you can optimize on scale on it. The final the takeaway from this is that how do we adopt AI everywhere? The need is where can I AI can create measurable value? That's what I was referring to. You need to understand out there because the reason is that once you show the impact in that area, then scaling AI becomes more easy.

Tim Rowe, State of Streaming 16:39

Come

Where To Connect And Final Challenge

Tim Rowe, State of Streaming 16:40

on, I couldn't think of a better way to end off in this installment. This is obviously going to continue to be an ongoing conversation. I can't thank you enough for being here. Folks want to learn more about your work, connect with you. Where should they go? What's the best place to connect with you?

Hemant Soni, AI Architect 16:55

I'm pretty much active on Linden. I do uh engage, I have an active newsletter. Every month I think about different areas, try to look into this month's feature. I've looking into where organizations are kind of had adopted here. They are thinking of scaling it down to basically think about something which was referring to that. Okay, where success matters. So look forward to this this month's edition as well.

Tim Rowe, State of Streaming 17:20

Excellent. We will link to your LinkedIn, we will link to your newsletter, but on top of all of that will be the link to this article. Encourage everyone to go read that, check it out. We talked about a lot of it here today, but there is a lot more in it. So, Hemant, thank you so much for being here. And we'll we'll see you again soon. Thank you. Appreciate it, Tim. Absolutely. And if you found this to be helpful, please share with a colleague, share with a client, start a conversation today. What's your first AI project? Think about that as you head into the week ahead. Wherever you are, we'll see you next time.