Webinar Summary
Platform-reported ROAS and recorded revenue almost never match. Ren Agarwal, Meghan Hodge and Aniket Deshpande walked through why, and what to use in a budget conversation instead.
The gap is easy to size. On $50,000 of spend, three ad platforms claimed $210,000 (4.2x), while the sales system recorded about $80,000 (1.6x). Each platform counts the same sale for itself. Add up the revenue your platforms claim, compare it with recorded sales for the same period, and that ratio is your gap.
Three monsters keep it alive:
- Truth: platforms grade their own homework, and joining sales to touchpoints is a monthly spreadsheet job.
- Resources and talent: when the person who built the model leaves, requests queue behind the data engineer.
- Usability: slow dashboards and wrong metric definitions get worked around, not fixed.
Models don't close it. MMM lags, MTA models what it can't see, and neither starts from a sale. Azalea starts from revenue and works back to the campaign.
Supermetrics is the foundation. 150+ sources at ad, keyword and creative level, with API maintenance handled for you. Data lands in BigQuery, Snowflake, Databricks or Azure with up to two years of backfill, and setup typically takes three to seven days.
In the demo, the ranking flips once ROAS is joined to sales:
- Google prospecting: roughly 6x on the platform, about 14x against sales
- TikTok: about 11x on the platform, about 7x against sales
- Prospecting out-earned retargeting on sales, even though retargeting looked stronger on platform numbers
Consistent PRISM tags and a UTM builder make campaigns comparable, and Ask Azalea gives everyone the same answer to the same question.
Six weeks to a fix. Weeks one and two map your setup and find where the join breaks. Weeks three to six build the sales-to-marketing connection. A forward deployed engineer then stays, so a new channel takes days, not quarters.
From the Q&A: Azalea sits alongside MMM or MTA as the check on them. It shows which campaigns sit in the path of real revenue, but proving what caused a sale still takes incrementality testing.
Key takeaway: if a model has nothing to check against, it isn't measurement. Start from recorded revenue, then work back to campaigns.
