Podcast page image

The Brandformance Podcast • Ep 73

There Is No One Number in Marketing Measurement

With Jessica Duncan

Director of Data Science & Analytics, AG1

Director of Data Science & Analytics, AG1

I think the desire for one number really comes to a desire to be confident in a decision. And we can get there without having to discount all the other numbers.

– Jessica Duncan, Director of Data Science & Analytics, AG1

Episode Insights

Transcript

In this episode, Pranav sits down with Jessica Duncan, who leads data science and analytics at AG1. She moved from hands-on modeling into leadership and teaches AI and data visualization at the University of Maryland Global Campus.

They ask what breaks in measurement when a one-product DTC brand adds new products and retail, how to align finance and marketing, and what to say when stakeholders want one number. They also take on how to keep analysts honest and how to make the case for brand.

Key takeaways

  1. Adding products breaks click-to-purchase attribution, since an ad can sell a different product. Once a catalog grows past one product, a click on an ad for a new product can end in a sale of the flagship. Teams have to redefine what success means for an ad before the old logic misleads them.

  2. Instrument the site and own your business logic before launches, so you can move fast. AG1 set up tracking for quiz views, compare pages and movement between products ahead of launch. Its analytics team also kept its own business-logic tables on top of clean engineering tables, so it could change definitions without waiting for a sprint slot.

  3. Measure DTC and retail as one business, not as separate slices. You can slice an MMM by retail or DTC to satisfy curiosity, but a channel that drives retail sales may also move DTC. Optimizing to one slice undercounts it, especially as retail offers fewer trackable touch points.

  4. Model a simple trackable KPI across channels first, then layer in DTC LTV. Retail purchases cannot be tied to an individual, so repeat rate and LTV are invisible there. AG1 models media impact on one simple KPI across both, then applies consumer-level LTV to the DTC portion afterward.

  5. Pick the measurement tool that fits each decision instead of forcing one blended number. MMM suits splitting total spend across channels, but not choosing which of six new creators to fund. For low-click creators, post-purchase survey gives the better, if imperfect, signal.

  6. Let finance test definition changes on real data before you ask for agreement. Before changing how a new customer is counted, the data team ran hypothetical forecasts and built dev tables so finance could see the effect on existing forecasts. That turned a definitional dispute into something people could test.

  7. Standardize KPI definitions in a shared semantic layer so numbers match everywhere. AG1 built a dbt semantic layer on Snowflake for CAC, ROAS and AOV, queryable through Claude and its BI tool. Every route to the data returns the same calculation, which limits cherry-picked or conflicting figures.

  8. Settle channel biases with a test, and favor platform lift studies when budgets are tight. To check a hunch like social beating search, run a test. AG1 leans on in-platform lift studies because geo holdouts split into DMAs and need more spend, though geo tests help when retail is in scope.

  9. Brand spend is hard to prove, but cutting top of funnel erodes bottom of funnel. Finance often asks for a clear return line that brand media cannot give. Jessica’s case is to agree on a measurement approach up front, because a dried-up top of funnel is much harder to rebuild than to keep stable.

The sharpest marketing conversations, straight to your inbox

By providing your contact info, you agree to receive communications from Paramark. You can opt-out at any time. For details, refer to our Privacy Policy