Tasfia Tasbin, CEO and Co-founder of Markopolo AI

San Francisco, CA – January 22, 2025 – Two years ago, Markopolo was an ad automation company. We helped merchants run better Facebook ads, optimize their Google campaigns, and maximize their ad spend.

We were good at it. Customers valued the product. Revenue grew steadily.

But something kept nagging at us.

We were building on rented land. Every time Meta changed an API, we scrambled. Every time Google adjusted their algorithm, we adapted. We weren't building infrastructure—we were building a nicer interface for someone else's.

Last year, we made a decision that carried real risk. We pivoted. Not a small pivot. A fundamental repositioning.

Markopolo stopped competing with ad tools. We started building the behavioral intelligence infrastructure for commerce.

Today, Markopolo is an AI-native customer engagement platform. And ATHENA—the first behavioral foundation model we released—represents exactly where we're headed.

The question nobody was asking

The entire industry is building agents right now. Shopping agents. Marketing agents. Support agents. They're all trying to answer the same question: What should I do next?

But nobody was asking the question that actually matters.

When AI agents flood the internet—clicking, browsing, simulating interest—how do merchants distinguish real signals from noise? How do they separate genuine human intent from automated behavior?

That's the gap Markopolo identified. That's why we built ATHENA.

How Markopolo got here

When we started Markopolo, we built something called MarkTag. Most people assumed it was another tracking pixel for analytics. It wasn't.

MarkTag was quietly becoming one of the most advanced behavioral intelligence systems in the market. It doesn't just track clicks—it captures hesitation patterns, comparison behaviors, scroll depth, reading patterns, even frustration signals. Every micro-interaction tells a story.

We were sitting on something powerful. We just hadn't realized what to do with it.

Then we asked ourselves a question that changed everything: What if we could train an AI on all this behavioral data—not from one business, but from hundreds?

Google, Meta, and Amazon have all built remarkable recommendation systems. But they made the same architectural choice: train within their own walls. Their models understand users on their platforms exceptionally well. The moment a user steps outside? Cold start. Blank slate. Start over.

Markopolo had something none of them had: behavioral data from 603 independent businesses. E-commerce stores, streaming services, SaaS products, mobile apps. All different. All real.

So we trained ATHENA across all of them.

What we discovered

Something remarkable emerged. The patterns held.

A person deciding whether to buy running shoes exhibits the same micro-behaviors as someone evaluating enterprise software. The hesitation before commitment. The comparison loops. The trust signals they seek.

Behavior isn't really about the product. It's about how humans make decisions.

ATHENA learned the grammar of human intention.

When ATHENA predicts what a user will do next, it's accurate 73% of the time on the first prediction. Nineteen out of twenty times, the actual action falls within our top five predictions. And it delivers these predictions in 0.01 milliseconds—a hundred times faster than typical recommendation systems.

The entire model runs client-side. User behavioral data never leaves their device. We compiled ATHENA into a package smaller than 10 megabytes. Privacy isn't something we added afterward. It's embedded in the architecture.

Why Markopolo made this bet

Consider what's happening in commerce right now.

Shopify, Walmart, Target, and others recently announced the Universal Commerce Protocol—an open standard for AI agents to transact across platforms. Google launched shopping agents for Home Depot, Kroger, and Lowe's. The agentic commerce era has arrived.

But those protocols coordinate. They tell agents how to communicate. They don't predict what humans actually want.

That's what Markopolo built. We're the prediction layer.

Within two years, agents will negotiate with agents. Shopping agents will interact with merchant sales agents. When commerce becomes autonomous, the winners won't be those with the best coordination protocols. They'll be the ones who can predict human behavior accurately enough to act on our behalf.

That's the future Markopolo is building toward.

What this means for our platform

This pivot doesn't abandon what we built. It puts a foundation under it.

MarkTag now creates a 384-dimensional behavioral fingerprint for every user. It understands hesitation patterns, comparison behaviors, and decision momentum. Our AI no longer segments people into buckets—it creates individual strategies for each person.

Here's what that looks like in practice.

When Sarah abandons her cart at 2:47 PM on a Tuesday, Markopolo knows she's in research phase two of three. We know she responds to social proof, not discounts. We know she engages with WhatsApp around 7 PM. We know she needs validation, not urgency.

Instead of sending a generic "You forgot something!" email, Markopolo's AI waits. At 7 PM, it sends a WhatsApp message with customer reviews. If she engages but doesn't purchase, it schedules an AI voice call for the next day at lunch—when she's historically most receptive.

This isn't personalization. This is individuation at scale.

Traditional marketing tools recover 10-15% of abandoned carts. Markopolo recovers 30-40%. Because we're not guessing.

The moat

Investors ask about defensibility. Here's Markopolo's.

Cross-domain behavioral transfer learning requires cross-domain data. We trained across 603 different businesses. That dataset doesn't exist anywhere else. Google has deep data on their platform, but not across independent merchants. Meta faces the same constraint. So does Amazon.

Every new business that joins Markopolo makes ATHENA smarter for everyone. Network effects, but for intelligence.

We also have a two-year head start on the core technology. MarkTag has been collecting and vectorizing behavioral data while the industry focused on campaign optimization. That compound intelligence effect matters. In month one, the AI learns baseline behaviors. By month six, it identifies micro-patterns. By month twelve, it predicts with 85%+ accuracy.

Our mission, updated

We started Markopolo with a straightforward goal: help merchants run better ads.

Our mission now is larger.

We're building the intelligence layer that makes every interaction count.

Where others see abandoned carts, we see individuals with unique needs. Where others apply templates, we craft journeys. Where others guess, we know.

By 2030, Markopolo will recover $100 billion in lost revenue by ensuring no customer is ever treated as a segment again.

To our stakeholders

To our team: When you joined Markopolo, you didn't join an ad automation company. You joined a mission to build AI that understands a billion humans as individuals. Every line of code you write, every model you train—it brings us closer to a world where technology truly understands human behavior.

To our customers: You believed in us as an ad platform. We're asking you to believe in something larger. The campaigns you run with us will become smarter because they're powered by ATHENA. The audiences you create will be more precise because we understand behavior at a level nobody else does.

To our industry: At 0.97 AUC-ROC, ATHENA outperforms Google's published recommendation benchmarks by a margin the industry considers transformational. At 72.67% Top-1 accuracy across 603 independent businesses, it more than doubles BERT4Rec's single-domain performance. This isn't incremental improvement. This is new territory.

The bet

In five years, every e-commerce transaction will be guided by an AI that understands the person behind it. The question isn't whether this will happen—it's whether merchants will use Markopolo's intelligence layer or attempt to build their own.

We're three years ahead. And we're accelerating.

ATHENA isn't just a product. It's infrastructure. The behavioral intelligence layer for the agentic economy. We trained it on how humans decide. Now we're deploying it to help machines understand us.

That's the bet. That's Markopolo. That's where we're going.