TechFlow’s 3.2x ROAS in 2026: Data Feeds Win

The whole game changed in 2026 with the move to agent-driven video discovery platforms, forcing a complete rethink of how we prep product data. Generic product feeds that were perfectly fine for old-school e-commerce listings are now a major handicap when AI agents are building personalized video experiences for shoppers. We put this to the test with a three-month campaign for “TechFlow,” a mid-sized electronics retailer, to see if obsessively detailed product data would actually pay off in this new environment. So, did all that granular work give us the returns we were hoping for?

Campaign Overview: TechFlow’s “SmartFind” Initiative

TechFlow needed to move more of its mid-range smart home devices, and they wanted to get in front of early adopters and tech fans in the 25-45 age bracket. Our main bet was that extremely detailed, context-rich product data would crush traditional broad-match campaigns when fed into agent-driven video platforms. We ran the campaign with a $75,000 budget over a 90-day period, from July 1 to September 30, 2026, and our success was measured by Return on Ad Spend (ROAS), Cost Per Lead (CPL) for their newsletter, and, of course, straight-up product sales.

Strategy: Hyper-Segmentation for AI Agents

Our entire strategy was built on creating multiple product data feeds, not just one, with each one designed for a specific agent discovery scenario. A single “smart speaker” entry in a feed wasn’t going to cut it. We took a device like the “SmartFlow Hub” and broke it down into separate, specific entries for “SmartFlow Hub – Voice Assistant Integration” or “SmartFlow Hub – Home Security Compatibility.” Each of these sub-entries got its own keywords, super-detailed attributes (like “compatible with Zigbee 3.0” or “on-device AI processing”), and even tiny review snippets from actual customers. That level of detail was the whole point. We knew AI agents don’t just scan for keywords. They figure out semantic relationships and user intent from rich data. It makes sense, given the 2026 IAB Digital Video Ad Spend Report showed 68% of people now expect video recommendations to be hyper-personalized, something only AI agents can really deliver at scale.

We set up a custom data pipeline that pulled real-time inventory and pricing from TechFlow’s ERP, feeding it into Salsify, a third-party PIM. We configured the PIM to spit out dynamic JSON-LD schema for every product variant, which makes it incredibly easy for search engines and AI agents to digest the structured data. For the videos themselves, we used an AI-driven tool called Synthesia to assemble 15-30 second product clips on the fly, pulling images, specs, and even voice-over scripts right from our data feeds. In practice, this meant if a user’s agent was hunting for a “smart thermostat with geofencing and energy reporting,” our system could instantly build a video for a TechFlow model that showed off those exact features.

Creative Approach: Dynamic Video & Contextual Triggers

Our creative plan was completely tied to the data work we did. We didn’t bother with expensive, pre-produced, one-size-fits-all video ads. Our focus was on creating modular video assets, short clips of individual features, animated graphics for tech specs, and user testimonials, and then tagging each module with metadata that matched our product attributes. So when an AI agent picked up on a user’s intent, like “show me smart plugs compatible with Apple HomeKit,” our system would stitch together a unique video right then and there. This video might have a virtual presenter (we had a whole library of them) talking about TechFlow’s HomeKit plug, showing the setup, and then cutting to a user review that specifically mentioned how well it worked with HomeKit. The approach pretty much eliminated creative fatigue and made sure every ad was relevant. Our total creative production cost for the campaign was about $15,000, a 40% drop from TechFlow’s last campaign that used traditionally produced videos.

Targeting: Agent-Assisted Audience Matching

We didn’t target in the traditional sense of just using demographics. We went after platforms that use AI agents for discovery, like Google’s “Discover Feed” and some of the newer personalized shopping assistants. We pushed our super-granular product data directly into their ad systems, letting their own algorithms match our product features to user queries and what they inferred about a user’s needs. Instead of just targeting “tech enthusiasts,” for instance, we set up campaigns for users whose digital assistants were frequently looking up terms like “smart home automation” or “voice control solutions.” We also A/B tested how much detail to show in the ad copy itself. One version had a simple title and price, while the other packed three key features into the description. That second, more detailed version beat the first one with a 12% higher click-through rate (CTR) every time.

Campaign Performance Metrics

The “SmartFind” campaign gave us some really solid results that confirmed our data-first strategy. Here’s how the numbers broke down:

The big jump in CTR and the drop in CPC were direct proof that our super-relevant ads, which were only possible because of the granular data, were hitting the mark with users that the AI agents found. The final 3.2x ROAS was more than double TechFlow’s previous 1.5x average, proving this wasn’t just a fun experiment but a commercially sound strategy.

What Worked Well: Granular Data and Dynamic Creative

The single most successful part of this campaign was the insane depth of our product data preparation. We loaded up product entries with every specific attribute we could think of, from user sentiment we pulled from old reviews to cross-compatibility info (like “works with Alexa, Google Assistant, and Matter protocol”), which basically gave the AI agents a cheat sheet for making perfect recommendations. We were giving them structured answers to questions a user might have before they even thought to ask. For example, including data points like “average power consumption: 5W” or “installation time: 15 minutes” in the feed meant the agent could serve up a TechFlow device to someone who just happened to be interested in energy savings or easy DIY projects. According to eMarketer’s 2026 Consumer Personalization Report, 72% of people are more likely to buy when content is tailored to them, so we were just giving the machine what it needed to do that.

The dynamic video generation was also a huge win. Being able to spit out hundreds of unique video variations without a ton of manual work meant that we could serve almost every user a video that spoke directly to their specific intent, which kept them watching. We saw that videos highlighting a specific technical spec, like “Bluetooth 5.2 range: 30 meters,” had a 20% higher view-through rate than the generic, fluffy overview videos.

What Didn’t Work as Expected: Inventory Lag

Our biggest headache was keeping inventory perfectly synced in real time. We had a few painful instances (about 5% of the time, unfortunately) where a dynamically generated video would promote a product that had just sold out. That’s a terrible user experience and it torpedoed conversion rates for those products whenever the lag happened. We found out that even though our PIM updated every 30 minutes, the ad platforms’ own data ingestion could lag by as much as an hour. That might not sound like a lot, but for a popular product, it’s a lifetime. The experience drove home that you need near-instant, minute-by-minute inventory updates for this stuff to work. My firm opinion is that any brand trying this needs to invest in proper API integrations that push inventory changes immediately, not just on a schedule.

Optimization Steps Taken: Prioritizing Real-Time Feeds

Once we identified the inventory lag problem, we built a direct API connection between TechFlow’s inventory system and the ad platforms for the most important products. This let us push stock updates instantly and got our out-of-stock ad problem down to less than 1%. We also started tagging products with a “low stock” attribute in the data feed whenever the count fell below 10 units. This let the AI agent know to maybe promote an alternative product, or we could have the system slap a “low stock – order soon” overlay on the video. That small change actually bumped conversion rates by 7% on those specific products by creating some real urgency.

We were also constantly tweaking the product data itself based on what the performance numbers told us. We saw, for example, that products with “voice assistant compatibility” explicitly listed in the feed did way better than ones that just said “smart home integration.” That led to an audit of all our descriptions to make sure we were being as specific as possible. We even started testing a “sustainability score” on some products, and the initial tests showed a 5% higher CTR from environmentally conscious users, giving us another data point to build on for the next campaign.

Conclusion

The TechFlow “SmartFind” campaign proved it beyond a doubt: spending the time and money on highly granular, well-prepared product data is the only way to win in agent-driven video discovery. Brands have to get past writing basic product blurbs and adopt a data-first mindset. You have to give the AI agents the rich, contextual info they need to do their job of connecting the right person to the right product at exactly the right time. Commerce is personal now, and the only way to do that at scale is with precise data.