#6. AdTech Stack — Architecture and Integration for RMNs
📋 Disclaimer
This article series is designed for CMOs, CPOs, and CCOs in retail and e-commerce. My goal is to help you convince your leadership to create and grow a dedicated Retail Media division.
Why listen to me? I built Retail Media from scratch to 2.5% of GMV in 14 months. I have worked with over 50 FMCG brands, from Coca-Cola to local manufacturers, and hold certifications in Amazon DSP, Criteo, and other AdTech platforms. Before this, I spent 7 years consulting on retail digital transformation in Russia and CIS, witnessing firsthand how large retailers monetize their data and traffic. I know the pitfalls, the KPIs that matter, and those that don’t.
What this article is about, in short:
The Tech Stack Reality Check
Ad Servers: Your Make-or-Break Component
Here's the thing about ad servers - they're like the heart of your operation, except if your heart stops for 100 milliseconds, you lose money. Not kidding.
I remember our first Black Friday. Our ad server couldn't handle the load and we watched conversion rates tank in real-time. It was brutal. Now I tell everyone: CitrusAd handles 50 billion requests annually for a reason. They've been through this pain already.
Amazon Ads ' system processes over 10 million queries per second. Sounds insane, right? But when you're dealing with that scale, every millisecond literally translates to revenue. We learned this the hard way.
Bidding Engines: Where the Magic (Should) Happen
Real-time bidding is where you either look like a genius or completely embarrass yourself. The system has 20-30 milliseconds to decide which ad to show. That's faster than you can blink.
The smart move? Let AI do the heavy lifting. PromoteIQ 's platform delivers 3x better MROI than industry averages, and honestly, their ML is way smarter than any human trying to optimize bids manually. I spent way too many late nights trying to outsmart algorithms before I accepted this reality.
Attribution: The Part Everyone Gets Wrong
Remember when we all relied on cookies? Those days are basically over. I've watched too many retailers panic about iOS updates and cookie deprecation.
Le Pixel by Lebesgue figured out how to track conversions using IP addresses and cross-device matching when cookies fail. It's not perfect, but it works. The key insight? Stop trying to track everything perfectly and focus on tracking what actually drives decisions.
DMP vs CDP: The Decision That Keeps You Up at Night
This is where I see retailers make expensive mistakes. DMPs work with anonymous data - fine for basic targeting. CDPs use your actual customer data - way more powerful.
Here's my take: if you're serious about retail media, go CDP. Yes, it's more complex. Yes, it costs more upfront. But you'll have better match rates and actually own your customer relationships. I've seen too many retailers regret the cheap DMP route.
Data Architecture: The Unsexy Stuff That Matters Most
Processing at Ridiculous Scale
TripleLift processes 4 billion ad requests daily. That's not a typo. We're talking 13 million database rows per hour. When I first heard these numbers, I thought they were exaggerating. They weren't.
The lesson? Your data pipeline will break. Plan for it. Use microservices so when one thing crashes, everything else keeps running. I learned this during a campaign launch when our monolithic system went down and took everything with it.
Storage: Hot, Cold, and Everything in Between
Storage costs can eat your budget alive if you're not careful. Captify keeps 30 months of data readily accessible, then archives the rest. Smart move.
Here's what works: keep recent campaign data and active customer profiles hot (expensive but fast storage). Archive everything else (cheap but slower). You'll thank me when you're not explaining massive AWS bills to your CFO.
Build vs Buy: The $2 Million Question
Building In-House: The Control Freak's Dream
Walmart Connect went full in-house and it's paying off. Complete control, custom features, no vendor dependencies. Sounds perfect, right?
Reality check: Amazon started in 2012 and didn't see real growth until 2018. Six years. Mirakl took two years just to build their platform. Unless you have Amazon's resources and patience, think twice. When to build:
SaaS Solutions: The "I Want to Launch Next Quarter" Option
As we discussed earlier, marketers use 5-10 different tools. That's a lot of vendor relationships, but sometimes it's worth it.
Criteo gets you standardized workflows from day one. CitrusAd reduced their release cycles from six weeks to two weeks with cloud-native architecture. Kevel lets you plug in your own ML while handling the infrastructure headaches. When to buy:
Hybrid: Having Your Cake and Eating It Too
Target 's Roundel and Kroger do this well. Build what makes you unique, buy everything else. It's probably the smartest approach for most retailers.
We went hybrid after my initial "build everything" phase nearly killed our timeline. Built our core targeting algorithms, bought the ad serving infrastructure. Worked much better.
Conclusion
Look, retail media tech is complex. But don't let perfect be the enemy of good. Start with what works, iterate fast, and don't be afraid to change course when something isn't working.
In the next part, we'll explore Go-to-Market Strategy...
Sources
VP Growth | 12+ years | Applied AI | 10x AI-Powered Revenue Scale | Retail Media Strategist | Cross-functional Product Marketing
3moOn medium https://nelepko.medium.com/6-adtech-stack-architecture-and-integration-for-rmns-03f908abe487
Content Writer @ Triple Whale 🐳 Senior Copywriter, Tone of Voice & Messaging Expert ✍️ Ecommerce, SaaS, AI 〰️ The O.G. Em Dasher
3moCan’t wait to dive into this! So glad my research helped shape this. ☺️