Everyone has a retail media strategy deck. Almost nobody talks about what happens when that deck meets production reality.

The second episode of our Media Outlook series goes where conference keynotes stop. It covers operational friction, internal resistance, and the product decisions that determine real profitability. The core question: does your commerce media network generate profit, or does it quietly bleed money?

Hosted by Danilo Tauro, Co-Founder of Cartograph AI, the conversation features Juuso Alho, Global Head of Product for Retail Media at JustEat Takeaway, and Nagarajan Chakravarthy (Naga), Chief Digital Officer at iOPEX Technologies. Together, they dissect the structural breaks, product tradeoffs, and operational realities that leaders must navigate to scale commerce media successfully.

Fixing the Foundation: Avoiding "Middleware Hell" 

A recurring theme in the episode is the danger of skipping fundamental groundwork in a rush to launch a Minimum Viable Product (MVP). Juuso Alho highlights that new retail media networks often face internal skepticism from colleagues who view ads as detrimental to the organic user experience. Beyond cultural buy-in, the technical hurdles are steep; integrating ad servers is far more complex than just dropping a JavaScript tag onto a page.

Nagarajan Chakravarthy points out that the real challenge, and where most "scars" occur, happens in the execution or "run phase" rather than the initial build phase. Nagarajan put a sharp edge on that last point: "Using yesterday's data for tomorrow's decision is absolutely useless" when your account managers are trying to optimize live budgets. The lag between what happened and what your team can act on is where early revenue dies. 

He identifies "middleware hell" as a critical operational trap. When campaigns take weeks to activate due to manual legal approvals, fragmented ad servers, and siloed data, organizations suffer from severe "time to revenue realization" delays. To combat this, Naga advises that building a strong "operational tissue" and execution velocity from day one is far more critical than simply buying big-rock components. Ultimately, failing to standardize data pipelines and processes leads to a compounding loop of "data debt" and "operational debt".

Build vs. Buy: The "Shoe" Analogy 

When deciding whether to build or buy core technology, Juuso offers a highly practical framework: selecting a vendor is like buying a pair of shoes. You can buy many different pairs, but if you don't know whether you are going to a fancy party or preparing to run a marathon, you will make the wrong choice. Companies must evaluate their in-house engineering skills and specific go-to-market needs before locking into a technology stack.

His point was that digital ad servers are now commodity software. The differentiation comes from how you connect that tool to your proprietary first-party data. If your engineering team is not ready to build a custom ad server from scratch, partner with an established vendor and invest engineering effort in the data integration layer that creates a competitive advantage.

Naga echoes this sentiment, noting that ad servers and measurement tools are largely available as commodities today. The real differentiator is how well these components are orchestrated, intertwined, and made native to your platform without leaving behind dozens of manual reporting steps.

Danilo reinforced this from the investor side. The networks attracting capital are the ones that ship working products quickly, not the ones that spend 18 months perfecting proprietary infrastructure before generating a single campaign dollar.

The Evolution of AI: From Prediction to Agentic Operations 

Cutting through current industry hype, the speakers clarify what AI actually means in today's ad tech landscape. While machine learning and prediction models have been standard in bidding algorithms for decades, the current shift is moving rapidly toward autonomous, agentic workflows.

Naga explains that AI is evolving from simple pattern recognition (e.g., merely alerting an account manager that a campaign is underperforming) to becoming a true "autonomous operator". Modern agentic AI can independently evaluate creative performance, check available inventory, generate reallocation scenarios, and seamlessly collapse the decision-to-action cycle.

Juuso shares a cutting-edge use case from JustEat Takeaway (backed by Prosus), where agentic workflows leverage Large Commerce Models trained on billions of tokens of shopping behavior. An agent can ingest an advertiser's brief, autonomously generate targeted campaigns with unique imagery and messaging for specific audience segments, and continually monitor and adjust the campaign without human intervention.

Measurement: Use a Compass, Not a GPS 

The industry's endless debate over perfect incrementality measurement often leads to "paralysis by analysis". Juuso argues that 100% accurate measurement is a flawed pursuit due to complex consumer behavior, such as the "halo effect" of seeing an ad on a phone and then purchasing in-store later, and the prohibitive expense of running constant control and variant groups. Instead, networks should treat measurement "as a compass, not a GPS" - a probabilistic tool that provides directional guidance rather than absolute certainty.

Naga completely agrees, emphasizing that an 80/20 directional measurement combined with high operational velocity will always beat a "perfect" measurement system plagued by slow execution.