How MarSci's Causal-Driven Attribution Builds Trust and Reduces SaaS Churn

Marketing measurement used to be simple. Advertisers could observe immediate performance changes and allocate their media spend successfully. Then GDPR, the iOS 14 updates, and the rise of video-first platforms changed everything. Today, users consume content differently. They navigate from platform to platform and get exposed to far more messages along the way. It's not just how advertising operates. It's deeper. The way consumers buy things has changed. As a result, advertisers have no idea where to allocate their budget. Every advertiser faces the same challenge, "Facebook ads report 50 sales, Google ads report another 50 sales, and I only have 20 sales." This is the marketing attribution problem. For me, it became personal while I was working at Facebook. I met so many advertisers who were confused with their media strategy. I kept seeing advertisers who genuinely didn't know where and how their customers were finding them, and that frustration is what pushed me to build MarSci.

There's always this assumption that everything can scale. There are so many case studies that claim the same thing. The reality is, it can't. Paid ads completely flopped for us. As an open-source company, we simply don't have the unit economics to support paid activity. Events, on the other hand, are where we excel. We organize educational sessions where we teach advertisers how to measure their marketing without any bias. We do that for other open-source models too, and it brings advertisers closer to the ground truth of running models themselves because that's where the industry is heading. Especially at the beginning, many businesses forget that growth tactics need to align with the business model. Not just the unit economics and how much profitability you have, but how a business delivers value to customers. That was something that I realized with MarSci, and now I believe we have a solid GTM which helps us grow nicely.

It's a great question. It primarily comes down to trust. How can they trust the results we deliver? And this has been a challenge for the industry for a while. Other competitors can mitigate that with their sales and CS teams. We - in the other side - can't do that. In some cases, model diagnostics are enough to prove accuracy. But in many cases, they're not. For a model to truly be trusted, it needs to validate some intuition the advertiser already has. For instance, the marketing team knows that TikTok Ads benefit the business, but they can't see that on their Google Analytics. This is a problem that happens very often. By using MarSci CDA (Causal-driven attribution), advertisers can see that indeed TikTok benefits the upper funnel activity and has an indirect effect on the business. When they validate their hypotheses, that's when they start trusting the platform. Of course, this also depends on the advertiser's understanding and background, but generally speaking, we try to educate as much as we can while users are running and setting up their models.

MarTech has very high churn rates. The whole industry struggles with it, and yet most people misdiagnose the root cause. The reason behind it is the service model. Most SaaS companies in this space are actually consulting firms using their platform as a vehicle to deliver services. The technology itself often isn't sticky enough to retain customers on its own. I believe that's exactly where the problem comes from and why churn remains so high. Advertisers stay when there's real, tangible value, not because a contract tells them to. The moment they stop seeing results or feel like they're paying for hand-holding rather than a product that works independently, they leave. With the rise of more AI-native solutions and MCP, I believe this dynamic will be massively disrupted. Advertisers will be empowered to run models and extract insights on their own, without relying on a services layer. That changes everything, because ultimately, advertisers are driven by the value they receive, and when the technology delivers that value directly, the old consulting-wrapped-in-SaaS model simply won't survive.

We're constantly deploying new features and re-evaluating others on almost a weekly basis. Nothing is sacred. If something isn't delivering value, we revisit it. We're committed to democratizing marketing measurement, but we'll do so by listening to what our customers have to say. The product is shaped entirely by the feedback we receive, and that's by design. Too many companies build in isolation, shipping features based on internal assumptions rather than real user needs. We refuse to do that. Every conversation with a customer is an opportunity to learn something we didn't know before, and we take that seriously. Our approach is simple, listen, build, reiterate, and keep doing that until we make MarSci a global measurement leader. The roadmap isn't set in stone, and that's a strength, not a weakness. It means we stay close to the problems advertisers actually face, rather than the ones we think they face. That responsiveness is what will set us apart.

We have more than 145 active advertisers on the platform, and we run roughly 300 models on a weekly basis. This is just the beginning. We aim to have 10,000 advertisers within the next 2 years.

I've had the chance to work with some brilliant minds in marketing measurement. I started as Head of Performance for a travel startup called Welcome Pickups, then moved to Skyscanner, where I was responsible for paid activities across various verticals and markets. After that, I moved to Facebook, and most recently, to TikTok, where I served as Lead Marketing Scientist. I've always been in marketing analysis, using statistical modeling to understand how and where to allocate marketing spend. My PhD was in this exact area. I published research around the nuances of the consumer journey, covering both performance and branding activities.

Not iterating fast enough. MarSci sits at the intersection of AI agents, open-source models, and decision-making for marketing managers. As a result, everything around us is moving incredibly fast, and we have to iterate even faster. In hindsight, there were moments where we spent too long refining a feature when we should have shipped it sooner and learned from real user feedback. In a space that evolves weekly, perfection is the enemy of progress. The companies that win aren't the ones with the most polished product. They're the ones that adapt the quickest. If I could go back, I'd push us to move faster, break things more often, and trust that our users would guide us toward the right direction. That's the lesson, and it's one we apply every single day now.