Show how Redis gives the ad decisioning stack the live context it needs to choose the right ad impression for yield, relevance, and viewer experience by activating Snowflake customer likes and dislikes at decision time.
In streaming media, the wrong ad decision is not just a relevance problem. It is a yield problem, a pacing problem, and a viewer-experience problem. Campaign systems, audience systems, and stream telemetry already exist. The issue is making them act together inside the ad-break budget.
This is built for media platform, ad tech, monetization, and data science teams. The visible demo stays customer-safe. The script is where the presenter carries the detail, the stakes, and the ask.
The data is already there: ad inventory platform, campaign management system, identity graph / CDP, content metadata store, Snowflake customer likes and dislikes, feature lakehouse, and Kafka streams for playback and ad events. The issue is not whether the data exists. The issue is whether it can be assembled and acted on in single-digit milliseconds inside the break while the moment is still live.
That is what this demo shows. Nine stages, one primary decision moment centered on Jordan Kim Household, and a decisioning pipeline that turns scattered operational context, including Snowflake preference data, into the winning action: Premium auto campaign with retail overlay.
At the top are the systems of record and the live signal sources for StreamSight Media. Nothing here is being ripped out or displaced.
The ingest layer has two jobs. RDI handles change data capture and operational sync from the core repositories. Redis FeatureForm handles the feature pipeline from the analytical and streaming systems into the Redis context layer. Two tools, two roles, one unified ingest layer.
The Redis context layer is the operational working set. Hot data and live session state stay in RAM for sub-millisecond access. Larger history, embeddings, and warm operational context sit in Redis Flex. And the Redis Context Retriever sits below those stores, assembling the Campaign 360 — audience state, inventory context, and bid history — and exposing it as structured tools for the decision engine. This is the layer that makes history and live state usable together.
The decision engine is where rules, eligibility, ML ranking, vector search, and policy arbitration come together. The output channels are where the business sees the result. And the learning loop makes every accepted or rejected action improve the next one.
Snowflake is important here because it holds customer preference data that is usually treated as offline analytics. In this architecture, Redis FeatureForm turns those likes, dislikes, and brand exclusions into online features. Redis RAM serves the hot ad-break working set, and Redis Flex holds the broader household history and preference vectors so the ad server can use them while the ad break is still open.
This is the architecture. Now let me show you what happens when the live customer moment actually starts.
Jordan Kim Household is the stand-in for the larger pattern. This is not one edge case. This is the repeatable decision moment StreamSight Media needs to handle every day.
The business stakes are simple: if the platform waits too long or acts on partial context, it loses the moment. If it decides fast enough with full context, it creates better yield, higher completion, and smarter suppression under hard policy and latency constraints.
We have one live moment to recognize Jordan Kim Household correctly and act before the old process falls back to something generic.
StreamSight Media keeps its existing repositories, models, and applications. Redis is not the new system of record. Redis Iris is the context engine — the operational serving layer that makes the existing systems act together.
The source systems in this use case are: Ad inventory platform, campaign management system, identity graph / CDP, content metadata store, feature lakehouse, and Kafka streams for playback and ad events.
RDI handles operational sync and change capture. Redis FeatureForm handles train-serve parity and online feature delivery.
In this stage, Snowflake is not replacing the ad server or the CDP. Snowflake remains the repository for preference history, while Redis makes those likes and dislikes operational for this impression.
Redis does not replace the existing stack. RDI and Featureform make that stack operational in the live decision window.
The left panel is who the customer has been over time. The right panel is what is happening right now. The decision quality depends on both.
Use the left side to explain durable context, then the right side to explain the live trigger. Bring the room back to why the winning action is Premium auto campaign with retail overlay and why the alternatives are weaker in this moment.
Redis Context Retriever assembles the Campaign 360 — audience state, inventory context, and bid history — so the decision engine has exactly the live context it needs.
The important point is that this is not just personalization. It is contextual intelligence. History without the live state is stale. Live state without the history is shallow. Redis is the layer that serves both together at request time.
The key difference is that the system is not only asking what campaign pays the most. It is also asking what this household has liked, disliked, avoided, or responded to before, and it can answer that inside the live decision path.
A profile tells you who the customer is. Context tells you what the business should do next.
This stage is for the data and ML stakeholders in the room. The point is not the specific feature names by themselves. The point is that the same features used to train the model are available online at the moment of decision with the same definitions.
Explain train-serve parity clearly. Most teams can train a model. The hard part is serving the right features fast enough in production. Redis FeatureForm on Redis closes that gap and removes the drift between the notebook and the application.
Tie it back to the visible demo. These features are what allow the system to choose Premium auto campaign with retail overlay instead of defaulting to Generic brand spot or surfacing Competing travel ad at the wrong time.
Redis FeatureForm turns the Snowflake preference signals into online features. Redis RAM serves the hot decision state, and Redis Flex keeps the broader preference vector available, so the ranker can see that auto has a strong positive preference and travel has a negative preference before the ad is selected.
Your model is only as good as the features you can serve in milliseconds, not the features you can describe in a slide deck.
Now the decision is visible. Walk the room through the winner first: Premium auto campaign with retail overlay. Explain why it wins on relevance, economics, and policy fit for this exact moment.
Then compare it to the alternatives. Generic brand spot is usually the path the legacy process would take because it is simple or generic. Competing travel ad is the kind of action a model might surface if it saw only part of the context or ignored policy and operational constraints.
The point of this stage is to show that Redis is not just ranking what is most likely to be clicked. It is arbitrating across policy, economics, relevance, and timing in one place.
The winning ad is not simply the highest-paying campaign. It wins because it clears policy, fits the content, respects the household's Snowflake preference signals, and still delivers strong yield.
We are not surfacing random recommendations. We are ranking the actions the business already cares about and choosing the one that fits this moment best.
The winner here translates directly into business language: better yield, higher completion, and smarter suppression under hard policy and latency constraints.
The value is not one click, one claim, one quote, or one call. It is the compounding effect of getting these moments right at scale.
The visible economics are the reason the next step is a pilot, not just a technical evaluation of whether Redis is fast.
The math is not the single transaction in front of us. It is what happens when this decision gets repeated across the full book of business.
The left side shows the legacy experience: generic, delayed, or incomplete. The right side shows the same customer moment with the right action already staged. It is the same surface and the same business flow. What changed is the decision layer underneath it.
The winner is Premium auto campaign with retail overlay. The real product is not the UI redesign. The real product is the ability to put the right action into the existing UI before the moment passes.
Same surface. Same moment. Different decision layer. That is the product.
Come back to the architecture now that the room has seen the story end-to-end. The same five tiers are still there, but now the audience understands what each one contributed to the outcome.
Summarize the three takeaways. First, this is not a science project — it is a practical architecture for StreamSight Media. Second, it is additive, not disruptive — the existing systems stay in place. Third, it is a business story first and a platform story second — the reason to do it is better yield, higher completion, and smarter suppression under hard policy and latency constraints.
Close on the ask. The next step is not to admire the demo. The next step is a focused working session to map this reference architecture to the customer's actual environment and scope one ad pod, one premium inventory slice, and a pilot against the current ad-decision path.