Webinar Summary
The problem was never the model. Supermetrics is seeing five times more customers connecting their marketing data to AI tools than a year ago, and with that comes the trust question: can you believe what AI gives back? Francesca's framing was that hallucinations and confident-but-wrong recommendations are usually a data problem, not a model problem. A data warehouse is the one clean shelf instead of a dozen scattered boxes. It buys you three things: speed, because pre-aggregated data loads in seconds with no timeouts; ownership, because platform APIs often only keep granular data for 90 days and tokens break without warning; and governance, because you control who sees what and your history stays intact even when a live connection drops.
Warehousing doesn't make AI smarter, it onboards AI properly. Francesca's analogy: you don't hand a new employee a task with no context. You give them the history. Fragmented data forces AI to piece the puzzle together and guess at what's missing. Clean, centralized, traceable data gives it the accumulated context to be useful, and gives marketers less time in reporting and more time on strategy.
The real blocker inside organizations is language, not technology. Scarlett's observation from ten-plus years of consulting: marketing teams speak ROI, attribution, and creative strategy. IT teams speak APIs, schemas, and databases. Neither has the context to decipher the other, which is why data foundation work stalls before the business use case ever gets touched. Ready-to-use connectors remove three hidden costs: engineering time building and maintaining API connections, the learning curve of every platform's dimension and metric definitions, and manual data validation across time zones, refresh frequencies, and backfill differences.
Case study: DTC ecommerce, from quarterly reporting to timely decisions.
Scarlett's first client, a fast-growing brand, went from three ad platforms to more than a dozen media and social platforms, with a five-person marketing team pulling data manually. Cross-channel performance reviews had degraded into a monthly or quarterly exercise. Scarlett's team piped all platform data plus GA4 behavioral data into BigQuery through Supermetrics, built full-funnel views from impression to conversion, and layered on custom attribution. The outcome was faster decisions, deduplicated conversions across platforms, and, most notably, the confidence to invest in upper-funnel campaigns, which are usually the hardest to justify because lower-funnel channels take the credit.
Case study: gaming, adding AI capability without touching the existing warehouse.
A mature client with a decade-old warehouse needed new features for churn prediction and LTV models, but restructuring the warehouse would have meant months of evaluation, and competitors move faster than that. Scarlett's team ran a no-data-movement pilot instead: Supermetrics pipelines ad platform and user behavior data straight into BigQuery, models are built inside BigQuery with no extra servers, and Gemini connects directly to BigQuery AI so the marketing team can query and iterate in natural language. Two to three weeks to a validated proof of concept, no rip-and-replace, and a playbook the client can now replicate across other games and markets.
Standardized data is what makes the AI connection cheap. Scarlett's point on this was concrete: previously, connecting an agent to marketing data meant writing long documents explaining what every dimension and metric meant and how CTR and CPM were calculated. When the data arrives in a standardized format, that work mostly disappears. You explain the business context and the agent can work from there.
What Scarlett would do differently if she rebuilt a foundation from scratch today.
Traditionally, you decide which dimensions and metrics you need first, then build. With AI in the picture, Scarlett argued you can't know yet what use cases will emerge, so the sequence flips: export everything you can into the warehouse by default, then spend your thinking time on the business use case rather than the data model.
Scarlett's bet for the next 12 months: multimodal. Structured data, unstructured data, images, and video are interpreted together rather than in separate workflows, which is closer to how marketing teams actually work. That means getting creatives into the warehouse alongside the platform data.
How Scarlett proves value before clients commit. For proofs of concept, TrueMetrics validates on two layers: the modeling layer, using accuracy metrics like MAPE, and the use case layer, pushing the predicted audience into an ad platform and A/B testing whether it actually performs. Both matter, because the client has to show results internally too.
Key takeaway: the foundation is the competitive advantage, not the AI. Get clean, standardized, centralized data in place and the advanced use cases, chat with your data, churn and LTV prediction, closed-loop creative workflows, stop being multi-month projects, and start being weeks of work you can replicate.
