Recast

Recast’s own Chelsea Parlett is back on Analytics Power Hour — this time diving deep into the statistical underpinnings of something we all encounter in marketing science: variability. In her latest appearance on APH, Chelsea explores: • How ANOVAs help partition variability in outcomes (like email CTR) into meaningful buckets like customer demographics, strategy, and seasonality. • How adding covariates (similar to CUPED) clarifies the signals we care most about in experiments. • What it means to run an omnibus test and why multiple comparisons aren’t just a theoretical concern. • The fact that — surprise! — ANOVAs are just linear models in disguise. And we can carry their logic into all sorts of analyses. If you're working in experimentation or marketing measurement, this one's worth a listen. Link to the full episode below 🎧

Feeling extremely inspired after having dinner with Recast and so many bright marketing analytics leaders here in Los Angeles last week. It reminded me how much I love talking with people who get the messy, imperfect reality of measuring marketing impact- and the fact that there’s no “one size fits all” answer. One idea that came up is how not every brand is ready for advanced analytics or MMM- and that’s completely okay. I like to think about measurement as something you build over time. It’s about recognizing where you are and what makes sense right now. The way I break it down: 1️⃣ Reporting – Are we even looking at the right metrics? Aligning on definitions, building reliable dashboards so teams know what’s happening. 2️⃣ Optimization – Can we make tactical, channel-level improvements? Continually test different creatives, targeting options, and bidding strategies. 3️⃣ Causal Understanding – Do we know what really drives results? Incrementality testing, geo experiments, MMM can answer "what would have happened if we did nothing?" 4️⃣ Strategic Planning – Can we confidently guide budgets and forecasts? Using learnings to plan across channels and quarters, tying marketing to actual business outcomes. Personally, I think the best approach is meeting teams where they are and building trust as you go. 🌟 Would love to hear how others think about this. How do you break down measurement in your work? What would you add or change in this framework?

We work with an iconic news organization that reduced CPAs 30%+ YoY and brought paid LTV:CAC into efficiency. Three things helped them achieve this: 1. Stronger use of first party data at different stages in the customer lifecycle. 2. Better matching of creative with different marketing channel contexts. 3. Introducing Recast to their marketing org. That company is Los Angeles Times, and while the efficiency gains are impressive, my favorite part of this story is actually how they use the Recast platform. The LA Times was new to MMM when they came to Recast, so took a methodical approach to building trust in the platform. They “stress tested” their models: making small budget changes based on Recast’s forecasts and then assessing how accurate those forecasts really were. As Recast passed these small stress tests, the LA Times started taking larger swings with the platform, shifting channel budgets and launching new experiments. A measured approach to model validation like this is hugely important to any MMM project, and it allowed the LA Times to build real trust in Recast over time. To read more on our partnership–including how the LA Times unlocked new upper funnel channels and used their MMM to inform their experimentation roadmap–check out our new case study (link is in the comments). And special thanks to Travis Bernard and Will Dobson!

Check out Recast’s own Chelsea Parlett-Pelleriti on her (second!) visit to the Analytics Power Hour pod. Chelsea and the team dive into: • Understanding the basic ideas of linear regression, and how they can unblock both data teams and non-technical stakeholders. • The importance of setting good priors and tactics for handling pushback during the prior setting process. • How to keep causality top of mind when building MMMs and using them to answer marketing science questions. There’s a whole lot more, and if you work in marketing science, you’ll find this conversation especially valuable. Link to the full episode is below 📺