Retail media has mostly been a look in the rear-view mirror: a shopper bought something, and advertisers paid to reach people with that purchase history. On September 17, Sam’s Club Connect launched a group of targeting products that try to flip the model—selling audiences based on what a household will buy, sometimes up to a year before the transaction.
What Sam’s Club is actually selling
The new Connect suite has four predictive products. The most aggressive, Predictive Precision Targeting, builds segments of households with no prior purchase record in a category. Procter & Gamble piloted it with Cascade Platinum dishwasher pods. Sam’s Club reports roughly 800,000 households were identified for upgrade propensity, with a claimed 94% accuracy rate and a 71% spending lift on Cascade Platinum Plus compared with what it calls typical campaigns.
Those numbers need separation. A high accuracy rate means little without the denominator, time window and base purchase rate. A 71% lift matters more, but the comparison is not defined: no media spend, duration, control group or measurement partner was disclosed.
“What we’re trying to demonstrate is that much more than historical purchase behavior, this machine learning model is an indication of what might come 12 months into a member’s life cycle,” said Harvey Ma, vice president and general manager of Sam’s Club Connect.
The other products are less speculative. One targets lapsed or staggered buyers—members who once bought a brand and then slowed or stopped. Another uses adjacent purchases as leading indicators for expensive replacement cycles: a holiday 2026 test will watch soundbar, gaming console and streaming device purchases to predict TV buying. The fourth enables conquesting against rival brands, with guardrails intended to keep it from becoming basic retargeting.
Why warehouse data changes the game
Sam’s Club sits on a membership file that ties cash transactions to accounts, reducing the anonymous-shopper problem that weakens many retailers. That gives years of longitudinal household activity across grocery, apparel, electronics and fuel. For performance marketers, this is materially different from a browsing history or view-through log: it is observed spend across categories, not inferred interest.
But the current evidence is one pilot with one CPG manufacturer. The architecture and post-campaign validation details are thin. And Walmart is pushing from the supplier side too: at Inspire, Walmart Data Ventures said Scintilla would add marketplace data for first-party suppliers next year, with its assistant moving from summarizing metrics to making recommendations.
How to test predictive audiences
Before folding “future buyer” segments into a campaign plan, run a controlled comparison against your best existing audience. Ask for:
- The denominator and time window behind any accuracy claim
- The baseline purchase rate in the category
- The exact “typical campaign” comparison behind any lift figure
- Holdout or control group design
- Post-campaign validation after the prediction window closes
The most useful framework is simple: treat predictive retail media as a hypothesis, not a guarantee. If the vendor can’t describe how the prediction was validated, test it with a small budget before scaling.
Source: PPC Land
