Episode 06 · Christina Johnson
Christina Johnson on retail media, unreliable reporting, and using AI to understand your ad spend.
The conversation
Christina Johnson founded Advisar after working in strategy at Walmart and advising retail-media businesses. She joins Noah to discuss inflated reporting, the questions advertisers should ask, and what she learned when an AI-built analysis failed in front of a client.
Three lessons
Sales and awareness need different scorecards. A retailer’s attributed sales do not tell you how many purchases the campaign caused.
“what are the goals of this campaign?”Christina Johnson · 11:35
Try this: Write down the campaign goal. Ask which metric measures it and what the report cannot establish.
Christina showed a client an analysis that looked credible but contained wrong numbers. The experience changed how she approached data expertise and review.
“it was a real lesson in, like, how convincingly wrong things can be”Christina Johnson · 34:51
Try this: Trace one headline number to its source rows. Check the date range, metric definition, and calculation.
Bring sales and marketing exports together, explain the business context, and ask AI what happened. Include factors such as seasonality that could change the interpretation.
“Learn how to get all your data.”Christina Johnson · 41:24
Try this: Choose a completed campaign and gather its exports. Ask for calculations, sources, and unanswered questions.
Notes save in this browser.
Using the sales and marketing exports I provide, explain what happened in this campaign. Ask for the campaign goal and missing metric definitions first. Check dates, seasonality, promotions, and stockouts. Show calculations and source references. Separate observations from hypotheses and explain what the data cannot prove. End with the three most useful questions to investigate.
Audit this campaign report before I share it. Trace headline metrics to source data. Check joins, duplicate rows, units, date ranges, and attribution windows. Flag discrepancies and show reproducible calculations. Distinguish attributed sales from incremental impact. Do not fill gaps with invented data. List what a human reviewer still needs to check.
Turn my reviewed campaign analysis into a one-page decision brief and publish it privately with Valet. Include the objective, findings, metric definitions, sources, caveats, next action, and reviewer. Distinguish evidence from assumptions. Exclude private identifiers and secrets. Return the live URL and verified access state. Keep it private unless I authorize public access.