The Distillation of Data: Alembic’s Causal AI Solution

Hey, it's E.B. Moss, and this is another episode of Insider Interviews, where you get the insider scoop on the business of media, marketing, and advertising. Today's scoop is coming from Miami and Possible.

It's the fourth annual Possible conference, and I decided I would do four different POV Possible interviews, the little mini-series for insider interviews. And today, the great mind I'm bringing you is that of Hitesh Wadwani. And he's new to Alembic. Hi, thanks for joining me.

Thank you so much for inviting me. Yeah. To come to the conversation. Yes, absolutely.

And you got off the stage yesterday and now you're in the hot seat. No pressure. No pressure. I'm always in for measurement talks.

Excellent. Yes. And, you know, there is a little bit of pressure because you're kind of brand new to Alembic, right? That's correct.

I joined the company three weeks ago. Oh, okay. So still trying to figure out a lot of moving pieces, but super excited to be with an amazing group of people to contribute to the measurement innovations. Yes, I am so intrigued by it.

You know, I've been talking to a few different companies here as part of this POV Possible, and they all have interesting names like Human, the cool company. Now we have Alembic, and that's like some scientific device what is it yeah that's correct and that was actually one of the first things i had to also learn about ellumbic when i heard the name it's actually a distillation device which can help you identify and refine different chemicals it can help you clarify things okay and the way it kind of relates to measurement is in the current context there is so much data these big organizations have and you have to really go granular and refine the insight which advertises and take into account and action upon so it's very relevant from that context especially from an ai evolution perspective beautiful all right i'm gonna like drill down a little bit more on that in a second but i should say that another funny name that you can relate to is this thing that became a verb called Google.

Right? So you just joined from 20 years there, I think? Oh, not so much. By the way, 20 years in Google is called a dinosaur.

I spent 12 years, which was definitely a very interesting, enriching journey. Overall, I've spent 14 to 15 years in measurement, And majority of that time was with Google building global measurement products like Google Meridian, contributing to Google Analytics in terms of budgeting, and also thinking through in the latest context of AI agent-like workflows. Okay, so you are the right person for the job now as GM of Alembic. That's correct.

GM of Alembic, focusing on causal AI and applied AI implementation. Okay. Causal AI. Let's just start there.

Actually, I, of course, researched you because I'm very anxious about speaking with you. You're so smart. And I read something on your announcement on LinkedIn, and I want to read it because I think it's sort of like a little microcosm here. You wrote, we're at a fascinating inflection point where generative AI captured everybody's attention over the last few years, but we have to agree that LLMs are great at text and language, but they're not designed for the rigor for enterprise data sets.

And this is my favorite line. He wrote, you cannot hallucinate your way through a multi-million dollar strategic budgeting or pricing decision or supply chain affecting inventory during holiday periods, etc. so now based on that that we all agree that llms are not for the job of marketing and measurement what the heck is causal ai and how is that more apt for the job yeah this is like a such a fascinating topic i'll say for me to deep dive into um the way i think about this is in the current context because LLMs were not really designed to predict next number.

They were designed to predict what the next word should be. The way causal AI is approaching this space is thinking of how can we build a real-time granular graphical representation of your whole enterprise based on all the data you have And that data could mean all of your marketing data all of the pricing data all of the promotions data Think of everything from a dimension perspective which can impact your sales So if you really care about how is my sales driven you have to create a digital twin of your enterprise from a data perspective.

And that's what causal AI really is trying to solve for by creating this graphical representation, you have a real-time causal AI framework, which can be used in many ways to identify what worked, why it worked, and what should we do next to make it work more. So that's the kind of predictive, not next word, but next number, next set, etc. So, and does it kind of graphically depict it for you when you say, it it wow yeah when you think of how you can represent down the data you can almost identify what are the nodes and then what are the edges of causality between those nodes and each of those nodes can represent different events which happened okay yeah that makes sense then i also want to know why would you want to join a company that's doing this now?

What's the problem that you see Alembic solving specifically and that gets you excited about being on board? Yes, yeah. So I've been on the quest for causality throughout my career in a way. You saw this coming.

You saw the cause and effect that you needed to jump into. Yes, 100%. Based on all the work which I've done previously, whether contributing to Google making or building its own measurement stack, building products like Meridian, which have been used by thousands of advertisers globally, whether it's enterprise segment or market segment. One thing which is common in order for these models to be actionable and accurate is a way to identify causality.

In the current context, again going back to the amazing capabilities now we have both in terms of the granular data available big organizations globally and the compute power we have to think through new innovation to identify causality in a much more real time in a granular manner and why those two phrases are super super critical yes because if you have to really action on things in real time to make sure you are being efficient in your decision making, as well as resource allocation across budgets or any other time investment for your teams, you have to be very confident in causality.

So it really goes to the inherent nature of how can we maximize returns and make sure teams become more efficient. So in terms of that confidence, is it too good to be true? Are marketers really able to rely on this? It seems like the holy grail.

I agree. And that's really led me to this journey because I truly believe it is the holy grail. In terms of really trustworthy outcome from any measurement solution, it's super important to have the right transparency in the approach which is applied to the modeling stage, which is why the way we are approaching this space and building causal ai is with full transparency so it's not a black box okay uh you can identify you can go much deeper into the outputs of the model and really figure out why the model is recommending what uh where to invest into uh you can understand the causal change yep um few examples i can give is uh the model will be able to identify if a particular sale at a particular time happened, was that only due to someone searched on Google with an intent or it was also preceded by a big TV campaign which might have influenced them to actually in the first place search for that product.

There could be multiple reasons for someone to buy a product and this causality can define this whole journey. so i'm guessing that cfos are a little happier with this if you guys are actually proving the value of investment is that true yeah that's very relevant to the whole uh challenge industry facing right now yes every cmo uh cfo conversation is usually uh two sides of the coin two sides of the same coin and they try to come to common terms in terms of defining what success means and the underlying requirement for that what success means or to even define that is to be able to talk in the same language And that where the outputs which we provide through our causal AI framework are designed to enable that conversation both for the CMO and CMO.

Okay. So all of this almost sounds too good to be true. And it might be the holy grail because we're always looking for correlation and everybody's buzzword is about attribution. You know, it's like this campaign ran and these results followed.

But can you guys really prove an outcome? And give me an example. Yeah, of course. I think one of the earliest examples which our team was able to test our framework on was with Delta Headlines.

Oh, yes. That's on your website. Read about that a little bit. Great example.

tell. Yeah, Delta Airlines had like this huge investment of almost 30 million dollars for the Olympic campaigns. Oh. And they were really interested in understanding, did that campaign drive impact directly on their sales, which is a common question for every marketer.

Especially if you've spent 30 million dollars. Exactly. Okay. But it's very difficult to prove impact from these events where you can actually not run any kind of experiment.

usually the way to approach causality in the industry is to run geo experiments where you can run do a b testing across different geos but that was not possible in this particular event because it's it's olympic and you cannot have a counterfactual for that so our model our framework was able to actually identify to the minus granularity of what drove majority of the sales for Delta. One of the examples which we shared with the CMO and they were blown away by the insights was that during the medal ceremony Yeah, when they're on the podiums and stuff.

Yeah, in front of the Eiffel Tower during the medal ceremony the placement of the Delta logo Yes. It said, drove most of the flies to Paris than any other moment in time and that level of granularity really doesn't come through from any other measurement tool I have at least seen in the market. Sure. Wow.

I'm so fascinated by how you're able to connect those dots quite literally, but the emotion of watching the ceremony and the Eiffel Tower. So I guess you can't prove emotion, but you can attribute moments. Exactly. Yeah.

So the way the framework really works is you, these moments are identified as nodes. And again, if they're having an effect in the graph, which is representation of your whole enterprise business, that graph and the weight of that edge will really define causality. So we can prove if a particular node is having a higher impact on your final sales or not. Okay.

Now that's, I get emotional watching the medalists or something, but you're not really identifying emotion so much as good opportunities and bad opportunities, and then you're feeding that back to the marketer. You can have definitely nodes which can have a negative impact. You can have changes in your pricing strategy where it might completely have a negative effect suddenly on your sales. maybe you started adding discounts and over time that impacted the value of your brand itself and from a consumer perspective in their mind they are always expecting a discounted rate which might lead to lower revenue in the long term so that is a very common example which we see through data.

There could be other ways of identifying what is having negative impact inventory management is a prime example of that where during holiday period if you're not managing your inventory and logistics accurately or forecasting the requirement in terms of demand accurately, that could lead to an opportunity loss. And our model is able to pick up those kind of deeper insights for improving your business. Okay, that's brilliant. We touched on this the cmo cfo thing but there's still kind of a disconnect there um and is this clarity helping close the gap because you spoke about that on the stage and you also talked about the challenge of silos how are we using alembic to can't we all just get along kind of solution Yeah that it I mean one of the biggest roadblocks for majority of the organizations right now especially marketing teams are these silos across the brand teams the performance team and then maybe there might be a central measurement team And because they are approaching the measurement problem from a completely different lens, usually it's very difficult to take those different approaches and then talk to a CFO in a manner where they can also justify the investments.

The way Alembic approaches this space is there's no differentiation in the causal ai graph between brand versus performance it's all connected in a way where you're getting a common uh understanding of how measurement works it is able to incorporate the the requirements for long-term brand impact and short-term performance impact but everything is translated in a way where it is comparable and that's what really matters for a cfo where they don't want to see different measurement approaches representing different Sure.

You know, there's a lot of buzz, influencer marketing, creator marketing. I mean, that seems sort of nebulous, but even that goes into the Alembic, the refiner. And so it doesn't matter if it's direct response, which is easier to assess, or the sort of esoteric creator marketing, you can identify all of that causing. that's correct the only requirement is you should have uh data to represent that in the model and the model will be able to pick up and include uh that as part of the causal ai graph so we have seen examples of where many companies already have started including their earned media data or influencer data directly to take decisions on whether they should be investing more in paid or invest more in organic to drive value out of the influencer marketing.

Interesting. So AI moves pretty fast. I think we all know that. Like yesterday was different from tomorrow.

But if you were going to project out, what do you think is really going to change in the next six months? and what do you want to impact as change over the next year now that you're three weeks in? Yeah, that's something I've been thinking around quite deeply. But I must admit, one year down the line in the current context is decades maybe to predict.

But the goal for me coming to Alembic is really improving how our customers are seeing the platform. whatever the customer really touches in terms of the outputs, how we can distill those insights for them to make the outputs actionable and enable them with specifically in the marketing context, two key use cases. How are they allocating their budgets on a regular basis to maximize ROI? And then how are they optimizing their campaigns on a daily, weekly basis, whether during holiday periods or others?

So that's just like the marketing domain example we are starting with. But we have already seen many organizations asking us questions outside of marketing. Yeah. Whether it's related to pricing strategies, whether it's related to inventory management.

So my focus will be to make sure we are solving all of these use cases for these organizations to be more effective. Yeah, that's cool. I know that you've gotten some nice investments and you have some pretty big companies that have already come to the Alembic table. So I anticipate you're going to be able to keep that growth going.

Yes, we are very thankful to our partners. NVIDIA is one of our biggest partners providing the level of compute, which is really a core requirement for this kind of approach to be successful. We recently bought our EITS supercomputer. So I'm very excited for the team to invest the time into really going into the details of the data and then the compute and how we can keep improving our causal AI.

You're like a kid in a candy shop right now. I feel like that definitely. That's awesome. Hitesh Wadhwani, thank you so much for speaking with me.

It was brilliant. Thank you so much, Ivi, for taking the time. And I hope the insights were useful for the audience. Definitely.

Well, I hope you got some good insider scoop from this episode of Insider Interviews with me, E.B. Moss, and Media and Marketing Pros. If you did, give it a like, give it a share.

And if you'd like to learn more about how you can get your own podcast, reach out to me at podcasts at mossappeal.com or follow Insider Interviews anywhere. Thanks again for listening. Thank you.