Vijay Kumar, Founder and CEO of Mile – Interview Series

Vijay Kumar, Founder and CEO of Mile, has focused his career on helping digital publishers extract greater value from programmatic advertising through advanced machine learning and data-driven optimization. Since founding Mile in 2013, he has led the company’s strategy and product vision around improving open-exchange performance, developing systems that dynamically manage pricing, optimize auction mechanics, and enhance signal quality to drive measurable revenue gains while reducing operational complexity for publishers.

Mile is a New York–based ad-tech company that operates as an AI-powered yield optimization layer at the “last mile” of monetization — where auction dynamics ultimately determine publisher revenue. Its platform integrates with existing programmatic infrastructure to deliver dynamic price flooring, traffic shaping, bid enrichment, and real-time analytics, enabling publishers to maximize fill rates and CPMs through intelligent automation. By focusing on precision optimization rather than adding more ad stack complexity, Mile helps publishers capture more of the value already present in their inventory.

You started Mile in 2013, long before AI became a buzzword in ad tech. What were you seeing firsthand in publisher monetization and open-exchange dynamics that convinced you the system was fundamentally broken and worth rebuilding?

When Mile started, the open exchange was scaling fast, but publisher-side intelligence wasn’t keeping up. Most decisioning on the sell side was static—manual floors, coarse rules, infrequent changes—while buyer behavior was becoming increasingly dynamic and strategic.

What stood out early was that auctions weren’t inefficient because of a lack of demand, but because publishers lacked a real control layer. Price discovery was effectively outsourced to buyers and intermediaries, with publishers reacting after the fact.

The system wasn’t “broken” in a dramatic sense—it was imbalanced. As auctions became more complex, publishers needed adaptive, data-driven decisioning on their side of the auction. That gap is what made the problem worth rebuilding from the ground up.

Today, Mile operates as an AI optimization layer on top of existing publisher stacks. How would you describe the core problem the platform solves for publishers right now?

The core problem is coordination.

Publishers already run sophisticated stacks—Prebid, Amazon, AdX, multiple SSPs—but each component operates independently. There’s no native system that looks across auctions and asks: how should this inventory be priced and routed right now, given actual market behavior?

Mile acts as an intelligence layer above the stack. We don’t replace Prebid or existing demand. We help publishers make better, real-time pricing and eligibility decisions using auction outcomes as feedback.

In practice, that means protecting value when competition exists and avoiding unnecessary restriction when it doesn’t—all without destabilizing delivery.

Mile’s AI trains on each publisher’s own auction data before activation. Why does site-specific training matter so much in live programmatic environments?

Because programmatic markets are highly local.

Two publishers with similar audiences can have very different demand elasticity, bidder overlap, latency profiles, and revenue sensitivity. Generic models trained on pooled data tend to learn averages that don’t actually exist in production.

By training on a publisher’s own auction history, Mile learns how their demand responds to price, competition, and segmentation. That allows the system to operate conservatively where needed and assertively where signal strength justifies it.

Site-specific training isn’t a nice-to-have—it’s what makes AI usable in live auctions without introducing risk.

Mile uses machine learning to continuously adjust minimum ad prices in real time, responding to demand signals without harming fill rates. Why is this kind of adaptive pricing so difficult to get right in programmatic advertising?

Because pricing mistakes are asymmetric.

If you underprice inventory, the cost is hidden. If you overprice it, the penalty is immediate—lost fill, lost revenue, and broken trust. That makes most pricing systems either overly aggressive or permanently cautious.

On top of that, auctions are non-stationary. Buyer strategies change throughout the day, across geos, and in response to broader market conditions. Static rules break quickly.

Adaptive pricing only works if the system understands uncertainty, tests safely, and knows when not to act. That’s less about maximizing CPMs and more about maintaining stability while capturing upside when conditions allow.

Trust is critical when AI is influencing pricing decisions. How do you think about transparency and control for publishers using Mile?

AI should assist publishers, not override them.

At Mile, publishers define the boundaries—floor ranges, inventory scope, rollout pace, and performance thresholds. The system operates within those guardrails and provides clear visibility into what’s changing and how it’s affecting outcomes.

We deliberately avoid opaque automation. Publishers can segment, pause, or roll back at any time, and they can see how pricing decisions correlate with win rates, CPMs, and revenue.

Trust comes from observability and control, not from asking publishers to “trust the model.”

You sit at the intersection of AI and publisher-side ad tech through your work with Prebid and the IAB Tech Lab. How do those perspectives shape how you build and govern Mile’s technology?

Working closely with Prebid and the IAB Tech Lab reinforces the importance of ecosystem-level thinking.

Short-term optimization that distorts auctions or exploits edge cases tends to backfire. Long-term value comes from strengthening open, transparent systems where publishers retain control.

That perspective shapes both architecture and governance at Mile. We integrate cleanly into existing frameworks, respect auction mechanics, and avoid logic that would create hidden advantages or undermine trust in the open exchange.

Our goal is to make publisher-side decisioning smarter without weakening the system it depends on.

Many AI-driven monetization tools promise uplift but fail in production. From your experience, what separates systems that deliver sustained results from those that don’t?

The difference is whether the system is built for production reality.

Many tools optimize toward a static objective and assume conditions will hold. In live markets, they don’t. Demand adapts, strategies shift, and yesterday’s signal becomes noise.

Systems that last treat production as a continuous learning environment. They measure constantly, adapt cautiously, and degrade gracefully when confidence is low.

Equally important is restraint. The best systems don’t intervene everywhere—they act selectively, only when the signal is strong enough to justify change.

As privacy changes, signal loss, and market volatility continue to reshape programmatic advertising, where does AI provide the most leverage for publishers today?

The biggest leverage is in understanding market behavior, not user identity.

Even as addressability declines, auctions still generate rich signals—bid density, price dispersion, response patterns, competitive overlap. These signals are difficult for humans to interpret at scale but well-suited to machine learning.

AI allows publishers to optimize based on how demand actually behaves, rather than relying on identity proxies that are increasingly fragile.

Mile works with premium publishers across multiple Tier-1 markets. What differences do you see in demand dynamics or optimization strategies across regions?

The fundamentals are consistent, but the risk profiles differ.

North America tends to have deeper, more volatile auctions, which rewards adaptive systems that can move quickly without overreacting.

European markets are often more constrained and stable, placing a premium on precision and conservatism.

In APAC, fragmentation and latency variability make downside protection and delivery stability just as important as yield optimization.

Across regions, the common lesson is that rigid logic fails—and adaptive, publisher-specific systems perform better.

For publishers evaluating AI optimization layers now, what should they understand about what Mile offers and when it delivers the biggest impact?

Mile isn’t a replacement for your stack, and it isn’t a shortcut.

It delivers the most impact when publishers already have demand but lack the ability to price and coordinate it intelligently in real time. That’s typically when manual rules and static floors start to plateau.

Publishers should think of Mile as an intelligence layer that compounds over time—learning their market, operating within their constraints, and improving decision quality as conditions evolve.

If you’re looking for quick hacks, it’s the wrong tool. If you’re building for long-term resilience and control, that’s where Mile fits best.

Thank you for the great interview, readers who wish to learn more should visit Mile.