Reading engagement metrics that actually predict shelf behavior

A brand manager posts a 15 second reel for a new oat milk SKU. It hits 2 million views. Three months later, velocity (units sold per store per week) at the two largest grocery accounts has not moved. The team is confused, because the video "worked."

It did not, at least not for the metric that pays the bills. And the reason is rarely fraud. Platforms report the numbers that are cheapest to generate and hardest to disprove, so a buyer who does not impose their own definitions ends up buying theirs. This lesson does two things: rank engagement signals by how well they track off-shelf movement, and set the media-quality floor you write into the contract.

Why views lie and clip-throughs don't

Begin with what a "view" is. TikTok counts one when playback begins. Meta's cheaper video objectives report 2-second continuous plays. So 2 million views can be 2 million thumb-glances at a muted pack shot, from people who may not buy the category at all.

Contrast coupon clip-through rate: the share of people who see a digital coupon (retailer app, cashback app, the brand's own site) and actively clip it to their account for redemption at checkout. Clipping costs the shopper a few seconds and signals interest in a specific SKU at a specific price.

The predictive hierarchy, ordered from weak to strong signal of purchase, looks roughly like this in CPG (consumer packaged goods):

1. Video view or impressionimpressionThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.View full definition →, weak: passive, autoplay-inflated, no category filter.
2. Like or share, weak to moderate: social currency, often about the creator rather than the product.
3. Social save rate, moderate: saving a recipe or usage post implies future utility, which correlates better than entertainment does.
4. Coupon clip-through, strong: a near-checkout action tied to a SKU and a price point.
5. Sampling request and follow-up purchase, strong: tracked through unique codes or loyalty ID matching.
6. Retailer app engagement with the brand's digital shelf page, strong: browsing plus add-to-list sits closest to the transaction.

The mechanism: proximity to purchase

The ordering is not mysterious. It tracks proximity to purchase: how close the action sits, in time and behavior, to a basket. A view happens in a feed, disconnected from any store. A clip happens while someone plans or executes a shop. Sampling redemption often happens inside the store itself.

Where those actions sit in the stage-by-stage journey is the funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition → mapping lesson's territory (for general cross-category benchmark data, start there). What matters here is the FMCG (fast moving consumer goods) wrinkle: most purchases are low-consideration, habitual and price-sensitive. Nobody researches shampoo for a week. Metrics that imitate high-consideration research behavior (time on site, video completion) therefore overstate themselves.

Two edge cases keep this honest. First, the exceptions are real: infant formula, pet nutrition with health claims, and premium supplements do involve research, and there dwell time and comparison behavior carry genuine weight. Second, save rate can be gamed by content type. Recipe and hack videos get saved by people who like watching cooking, not by people who will buy your paprika. If save rate rises while the linked digital shelf page shows no lift in add-to-list, you are producing food entertainment, not demand.

Worked example: sampling ROIROIReturn on Investment: the ratio of net profit to the cost of an investment. A 300% ROI means each dollar invested returns $3.View full definition → vs. video campaign ROI

Two parallel Q1 2026 campaigns for a new snack bar, each with a $50,000 budget (illustrative).

Campaign A: paid social video

• 5 million impressions at $10 CPMCPMCost Per Mille: the cost to deliver 1,000 ad impressions. A pricing and benchmarking metric for awareness campaigns where reach matters more than clicks.View full definition → (cost per thousand impressions, estimate)
• 2 million completed views
• A generous 0.05% view-to-purchase rate (passive video in CPG is commonly cited at 0.02% to 0.08%, estimate)
• Purchases: 2,000,000 × 0.0005 = 1,000 units, so $50 per incremental unit

Campaign B: in-store sampling with redemption tracking

• 50,000 samples at $1 each including staffing (estimate)
• Sample-to-purchase within 30 days is typically cited at 20% to 40% (estimate, varies heavily by category)
• At a conservative 20%: 10,000 units, so $5 per incremental unit

Roughly a 10x gap on cost per unit. Now the second-order point the arithmetic hides: those two lines are not substitutes. Sampling scales with feet and stores, so 50,000 demos cannot buy national reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition → at any budget, and it only ever touches people already walking the aisle. Video reaches households that have never considered the category. Treat the comparison as a reason to fund sampling properly, not as a reason to zero out reach, or you will win on efficiency for three quarters and then run out of new buyers.

The media-quality floor a buyer should impose

Efficiency claims are meaningless if the impressions underneath them were never seen. The floor is contractual, not conversational.

• Viewability measured against the MRC and IAB standard, 50% of pixels in view for one continuous second for display and two seconds for video, verified by an accredited third party rather than platform self-report.
• Completed-view definitions written into the insertion order. Buy 15-second or completion-based units (Meta's ThruPlay) and TikTok's focused-view style definitions rather than 2-second plays, and price the difference. The same creative bought on the cheaper definition can look twice as efficient and sell nothing.
• An invalid traffic (IVT) ceiling with make-goods above it, typically in the low single-digit percent, again from an accredited vendor.
• Sound-on rate reported separately for any creative whose product claim is in the voiceover.
• A holdout on anything you intend to claim credit for: withhold a tenth or more of markets or households for six to eight weeks. Weekly velocity noise in a mid-sized SKU is large enough that a test unable to detect anything under a couple of percent lift will simply return whatever you hoped for.

Procter & Gamble made this a public procurement position rather than a measurement preference. At the IAB annual leadership meeting in 2017, Marc Pritchard demanded one viewability standard, accredited third-party verification, transparent agency contracts and anti-fraud certification, and P&G subsequently cut more than $200 million of digital spend, reporting that reach went up rather than down as low-quality and over-frequenced inventory came out. The lesson for a smaller buyer is the same and cheaper to apply: you do not need P&G's leverage to write the definitions into a $50,000 insertion order.

Building a simple predictive scorecard

Rather than trusting a single metric, weight several by their historical link to sales:

engagement_score = (
    0.05 * video_completion_rate +
    0.15 * social_save_rate +
    0.35 * coupon_clip_through_rate +
    0.45 * sampling_redemption_rate
)

The weights are not universal. Calibrate them by regressing your own campaign history against POS (point-of-sale) scanner data from a syndicated provider, and expect to need a dozen or more comparable campaigns before the coefficients stop moving. Until then, treat the scorecard as a ranking device, not a forecast. The principle survives either way: weight signals by correlation with verified purchase, not by ease of measurement. Video is the easiest number to pull and the weakest predictor, which is exactly why it dominates dashboards.

What retailers and CPG brands actually watch

Retail media has shifted the industry toward exposure-to-basket measurement, the closed loop the retail media lesson sets out, and that is where budget has moved: P&G and its peers now discuss retail media in earnings and conference commentary as a shift away from pure upper-funnel spend.

The failure mode is inheriting the vendor's grading of its own homework. A network reporting in-basket sales after exposure is describing correlation among its own loyalty shoppers, who were going to buy the category anyway. Ask three questions before accepting the number: was there a control group, what lookback window was applied, and does the reported lift include the brand's existing buyers. If the answers are no, 30 days and yes, the ROASROASReturn on Ad Spend (ROAS) measures the revenue generated for every unit of currency spent on advertising, calculated as revenue divided by ad cost.View full definition → figure is a shopper-loyalty report, not evidence your media worked. Vendors also have an incentive to publish flattering clip-through and CTRCTRClick-Through Rate (CTR) is the percentage of people who click a link, ad, or call to action out of those who viewed it.View full definition → (click-through rate) figures, so cross-check against neutral industry reporting such as the Association of National Advertisers (ANA).

🎬 [VIDEO: "How Retail Media Networks Are Changing CPG Marketing" - https://www.youtube.com/results?search_query=retail+media+networks+cpg+marketing - search results for current explainer videos on closed loop attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → in retail media, useful for seeing how brands like Walmart Connect and Kroger Precision Marketing structure measurement]

Key Takeaways

• Views and impressions predict shelf movement poorly because the platform definition is permissive: TikTok counts a view at playback start, and Meta's cheap objectives count two seconds.
• Coupon clip-through and sampling redemption predict well because they demand a small action close to the transaction.
• Save rate is a middling signal and an easily inflated one; check it against add-to-list on the digital shelf before believing it.
• Impose the quality floor contractually: accredited viewability against the MRC standard, an IVT ceiling with make-goods, completion-based view units, and a real holdout. P&G's 2017 stance showed reach can rise while spend falls.
• Cheap cost per unit from sampling does not scale past your store footprint, so read the 10x efficiency gap as a funding argument, not a licence to stop buying reach.