The AI Advertising Podcast: S2
Episode 3
AI and Audience Intelligence: Unlocking Hidden Segments at Scale
About This Episode
How do you discover audiences you didn’t even know to look for?
AI is transforming audience intelligence, making it faster, more precise, and more predictive. From probabilistic modeling to clean room collaboration, today’s experts break down how brands and agencies are redefining how they understand, segment, and engage with their customers.
Jeremy Lo | Managing Director, AlikeAudience
Karan Saggi | Director of Client Services, StackAdapt
Transcript
Diego Pineda (00:00:00)
What if you could predict which customer segments are about to emerge—before your competitors even see them? That’s the promise of audience intelligence powered by AI.
For decades, we’ve grouped customers by broad demographics or surface-level behaviour. But those days are fading. Now, we have machine learning, large language models, and real-time optimization, all working together to generate deeper insights and smarter campaigns.
In today’s episode, we explore how AI is advancing audience intelligence, from segmentation and data quality to predictive targeting and creative personalization.
Joining me are two voices driving innovation in this space: Jeremy Lo, Managing Director at AlikeAudience, a data science company building mobile-first solutions for the cookieless world in APAC. And Karan Saggi, Director of Client Services at StackAdapt.
Let’s get into it.
Podcast Intro (00:01:01)
Welcome to the AI Advertising Podcast, brought to you by StackAdapt. I’m your host, Diego Pineda. Get ready to dive into AI, Ads, and Aha moments.
Diego Pineda (00:01:17)
Audience intelligence is evolving fast. What once involved surveys and focus groups now includes billions of behavioural signals processed by AI. Karan Saggi gives us a look at the tools driving this shift.
Karan Saggi (00:01:30)
There’s supervised learning models for customer segmentation and prediction. There’s also Gen AI models, which you may have heard of. These are for creating personalized content or experiences that are more tailored for defined audiences. There’s also natural learning processing or NLP models for understanding customer sentiment and their intent through text. The more commonly known might be machine learning. This is a subset of AI and deep learning, which is a subset of machine learning. You may be using one of these. You may be using multiple, but they’re all meant for different purposes.
Back in 2018, in fact, this reminds me of a survey that eConsultancy did. They asked advertisers how they are using AI. And the top-ranking answer was using AI for audience targeting and for audience segmentation.
Over 80% of respondents said that they were already using AI or planning to use AI for this. And mind you, this was back in 2018. So we’ve come further since. The reality that’s now quickly settling in is each org must ask themselves, is AI a must have or is it a nice to have? And what the right place for AI is within the org.
For us, we’re enhancing the very core of advertising strategies with AI. From real time campaign optimizations to uncovering new audience segments, which we’ll touch on a little bit, machine learning optimization models are really turning what was once guesswork or more probabilistic into a lot more precision these days.
Diego Pineda (00:03:04)
Jeremy Lo, coming from a data science perspective, explains how AlikeAudience processes more than 20 terabytes of mobile data each month—requiring far more than spreadsheets.
Jeremy Lo (00:03:15)
We have over 3 billion mobile IDs globally, offering partners like Stack Adapt and DSPs and SSPs access to our audience or our data pool.
In terms of technology, what we do is, you know, in the past it’s really sourcing a lot of that data, applying our proprietary data modeling. So we use, you know, in the past past two years, generative AI and LLMs. Large language models to process our data. So if you can imagine, you know, it’s a lot of data we’re talking about. And, in a sense, it’s around 20 tbs of data that we process monthly. You know, Word and Excel will not get that done. So for us, in terms of technology, we’ve been deploying, you know, advanced models, and also applying, you know, I know AI gets thrown a lot, but from a generative AI perspective, help us process that data for efficiency in terms of give us that bit more power and speed because as you know, in data, we need to keep everything fresh.
Diego Pineda (00:04:15)
We’re talking about a new level of agility here. And that brings us to the foundation of any AI model: data.
Diego Pineda (00:04:25)
Sophisticated AI models don’t matter if the data feeding them is fragmented or outdated. Marketers today face a major challenge in unifying diverse sources of audience data.
Karan Saggi (00:04:37)
AI is only as good as the data it learns from. So data hygiene can be a challenge. Say if your data is being collected from multiple sources, like your CRM list or your email, your website, pixels, social media, surveys, and so on, the format and quality of audience data can really differ. It can really vary.
So struggling to map that and unifying everything can be a challenge. The other challenge relating to data as well can be the freshness of it. How real time is it? How relevant is it as your audience evolves and how the consumer evolves?
Plus, you may also be running into data governance issues or compliance issues. This is more common in certain verticals or certain regions of the world where we have different laws around them.
ACM Interactions actually published a paper back in 2022 called Data Excellence for AI, where they advocated for this exactly. They advocated for data quality, for data hygiene, because the efficacy of machine learning really depends on that.
Diego Pineda (00:05:37)
At AlikeAudience, Jeremy emphasizes that data integrity starts long before AI enters the picture.
Jeremy Lo (00:05:44)
We believe it’s noncompromised in terms of being compliant. So that’s the number one importance. AI, or no, AI, I would’ve put that straight, is for us is, is that’s number one when we’re dealing with data. You know, there’s zero tolerance. So for example, we wanted to, um, be compliant in terms of, you know, internationally different privacy laws, GDPR, for example, in Europe we need to be compliant. Other ways, we actually apply the opt-in, opt out for our audience.
So for us, it starts from the source. So rather than before the AI. AI is more of a, in terms of the analysis and activation, but for us, it starts with the source. So when we source our data, when we collect our data, that is the steps and the process in place to make sure we’re definitely compliant. For example, PIIs, right? you don’t want to be collecting emails and phone numbers. That’s a big no-no. So that everything, in short, all those requirements come actually before the AI piece as well. It comes from the source.
Diego Pineda (00:06:40)
Clean, compliant data is the fuel. But what can AI do once the engine is running? That’s where discovery comes in.
Diego Pineda (00:06:50)
Marketers often focus on their existing audiences. But AI is expanding the lens, helping brands find new customer segments they wouldn’t have predicted on their own.
Karan Saggi (00:07:01)
Think of it like giving your strategy peripheral vision. As marketers, you don’t just see who’s in front of you, but you also kind of want to peek behind the curtain. You kind of want the big picture. Today, the latest AI models can allow for a lot more granularity in audience segments that are richer, that have more nuanced personas.
In practice, marketers can use AI to find more net new customers, more net new consumers based on their first-party data than ever before. At Cannes last year, this reminds me of a campaign that Monks presented. They ran this campaign for a wellness company called Hatch. And they used AI to help generate three key audience personas for this campaign. They call them the stress professional, the biohacker, and the wellness enthusiast.
So cases like these show how brands are going after emerging audience segments that they may not have thought of or may not have thought of with that level of granularity.
Diego Pineda (00:07:55)
For Jeremy and his team, this kind of discovery hinges on blending deterministic and probabilistic data.
Jeremy Lo (00:08:02)
When we define data, I think deterministic data, it’s really about the precision, right? So it’s more exact and it’s more verifiable. Tends to be sourced from individual, but not necessarily, but very likely.
So for example, I’ll give an example would be in term when we say, you know, deterministic data, it’s really something you can determine, right? Like a custom ID, for example. You go to the supermarket, custom ID. Email? email address for example, credit card details, but it gets a little sensitive when it gets more PII stuff, right?
You don’t wanna identify. So that’s get a little sensitive in terms of identifiable data, but, you know, compare that to probabilistic. I think it’s more, you know, I, I’d say more patterns we want to find. So different types of data sets, different usage, but I think it, it provides different types of, you know, value.
It’s important to have both, you know, deterministic and probabilistic in mind to activate against, right? There’s pros and cons. So, for example, a downside of deterministic, the scale. It’s not necessarily scalable, so you have your own segments. But what if I’m looking to do more, you know, model and I want to build on those segments? What can we do? So that’s what something like AlikeAudience, we see value in terms of third party data with more sizable data sets that can, you know, predict, you know, certain trends or attributes and build on those deterministic data. So that’s what we’re, you know, third party identifiers come into play. That gives us a bit more scale, for example, in terms of products would be, you know, built lookalike modeling on top of certain data segments to, for, you know, advertisers or brands to build against or to target against. That’s always something that we we’re trying to fine tune, um, and to drive performance.
Diego Pineda (00:09:52)
Understanding your audience is only half the equation. The next step is getting your message in front of them at the right time, in the right context, with creative that resonates.
Karan Saggi (00:10:01)
At StackAdapt, we actually patented a technology called PCAI or Page Context AI. This reaches people when they’re in the right receptive frame of mind for your brand. AI finds the right context first and then places the ad within it to get in front of the audience.
Diego Pineda (00:10:18)
Jeremy sees this kind of intelligent activation growing across emerging channels like audio and programmatic out-of-home—but attribution is still a puzzle.
Jeremy Lo (00:10:27)
In terms of multi, you know, channel, you know, for us is making sure the IDs are relevant first and foremost. Specifically, we’re talking about, for example, CTV. We’re talking about different channels. Different channels work differently. At the end of the day, that’s the holy grail of programmatic, right? But that’s the catch, because there’s no single source of truth from an attribution perspective on an ID perspective, if you face out cookies, for example, there’s no ID specific with CTV.
But you could use emails, right? Hash emails as an identifier. So now as an industry, we tend to move towards more preferred, IDs to activate, I guess cross channel. Um, and also we’ve got technology, DSPs that actually have more in-house or in platform ID graphs that connects built on data providers, and then that can activate across multi-channels. For example, using emails. Hash emails through the ID graph to power multi, multi-channel campaigns because they have a better view of certain audiences. As a data play, I think that’s what’s exciting is we’re still growing that piece because other channels are quite new, but we’re definitely seeing growth in specifically in the US but also in the past six months, more conversation around, how do we activate, how do we smart about program out of home? What about audio is coming? So I think that’s the bit that ‘s getting everyone excited. But I think the key takeaway here is not necessarily just the targeting piece, but the attribution and the measurement.
Diego Pineda (00:12:08)
Let’s bring it all together with some clear takeaways:
Audience intelligence is deepening thanks to AI—marketers can now see not just who people are, but how they behave and what they care about.
Data quality and privacy compliance aren’t optional. They’re the foundation of any effective AI model.
Emerging segments can be found and activated faster using real-time data, lookalikes, and AI-driven modeling.
Creative and media execution are evolving too—AI personalizes content and optimizes it across new channels like audio and CTV.
As AI takes more of the heavy lifting, marketers need to level up in new ways. Less data crunching, more insight creation and decision making.
Podcast Outro (00:12:53)
Thanks for listening to this episode of The AI Advertising Podcast. This podcast is produced by StackAdapt. Visit us at stackadpat.com for more information about using AI in your advertising campaigns. If you liked what you heard, remember to subscribe, and we’ll see you next time.
You May Also Like
Infographic
GLOBAL
Nov 17, 2025
2026 Programmatic Advertising Trends Sneak Peek
Webinar
GLOBAL
Nov 13, 2025
Innovation Forum 2025: Breakthroughs In Marketing Technology
