Loyalty data: what it adds on top of POS

What loyalty data is

Loyalty data is retailer point-of-sale data with one extra column: an anonymized household identifier attached to the basket, captured when a shopper scans a loyalty card, enters a phone number or logs into an account. That single column changes what the file can answer. POS tells you an item sold 132,700 units in twelve weeks. Loyalty data tells you those units went to 41,300 households, and what each of them did next.

Everything else people call loyalty analytics follows from that join. If the basket is not linked to a household, no amount of processing produces repeat rate, penetration or a genuine buyer count.

What loyalty data answers that POS cannot

The middle rows are the whole reason to buy it. Basket-level analysis on plain POS gets you a long way, and market basket analysis works fine on anonymous receipts, but a receipt has no memory. Repeat and penetration are household facts, and they need the join.

The read that only loyalty data gives you

Northbay Salsa is an illustrative 16 oz jar selling through a 970-store grocery chain. The POS file for the latest twelve weeks says 132,700 units, 11.4 units per store per week, $529,473 at a $3.99 average retail. Year on year that is up 8.2%, and in a category review that is a good slide.

About 72% of those units scan to a loyalty card, which is 95,544 units. Here is the same growth, read through the card:

The arithmetic closes: 19.0% more households each buying 9.1% less is 8.2% more units. But it is a completely different story from the one the POS number told. This is not a brand getting stronger with its buyers, it is a brand recruiting hard and holding a thinner share of each new buyer. Repeat fell three points while units grew. Nothing in the POS file could have surfaced that, and the two readings point at opposite actions: the POS read says press the advantage, the loyalty read says fix the second purchase before spending more on trial.

Penetration comes out of the same file. Against roughly 2.14 million loyalty households shopping the chain in the period, 41,300 buyers is 1.93% household penetration, which is the number that tells you whether the growth has room left in it. Household penetration and repeat purchase rate both go into more depth on the formulas.

Who holds it, and what it costs to get

Loyalty data belongs to the retailer, which is the first thing that shapes the market. Most large grocers monetize it through an analytics arm or a partner.

Kroger's is 84.51°, which states that it applies "data from over 62 million households in the U.S." and serves "more than 1,500 consumer packaged goods companies, agencies, publishers and business-affiliated partners", with Stratum and Prism as its insight products and Kroger Precision Marketing as the media arm (8451.com). Tesco's Clubcard, introduced on 13 February 1995 and past 20 million users by 2021, is run by dunnhumby, which stores the store, products and price for each Clubcard transaction (Tesco Clubcard).

The costs are not only the invoice, and the non-cash ones bite harder:

It is per retailer. Kroger loyalty data tells you nothing about Albertsons. There is no additive total, so a brand selling into eight chains either buys eight datasets with eight hierarchies or picks two and accepts partial sight.

It arrives in the retailer's language. Their category tree, their fiscal calendar, their household segments. Reconciling it against your syndicated view is a project, not a mapping table, and the same problem shows up in the Kroger data-source comparison.

It is licensed, not bought. The permitted uses, the retention window and what may leave the platform are all in the contract, and they are the reason most mid-size brands run it for one or two accounts rather than all of them.

The biases nobody adjusts for

Unlinked baskets are not random. In the example above, 28% of units never touched a card. Cash shoppers, hurried shoppers and one-item trips are over-represented there, so the linked file is a biased sample of the same store.

Households are not people, and cards are not households. One card gets shared across an extended family; one family holds three cards. Both errors run in the direction of overstating household counts and understating repeat.

Demographics are usually self-reported at signup and never updated. A household that enrolled as a couple in 2019 with two children is still recorded that way. Modeled demographic overlays are a further inference on top of that.

Where Scout fits

This is the honest boundary and it is worth stating plainly: Scout reads transactions. On POS alone it will tell you units, velocity, distribution, price, basket composition and promoted lift, and it will not tell you how many households bought, whether they repeated, or anything about who they are. Those are loyalty-linked questions, and POS is transactions, not people. Where a retailer's own file carries a household key, the household questions become answerable, and where it does not, no analysis recovers them. A vendor claiming demographic insight off plain scan data is modeling, and the page should say which.

The short version

• Loyalty data is POS plus an anonymized household identifier. That one column is what makes repeat, penetration and buyer counts possible.
• It routinely contradicts the POS read. In the worked example 8.2% unit growth was 19% more households each buying 9% less, with repeat rate down three points.
• It is retailer-specific, non-additive, licensed, and delivered in the retailer's own hierarchy, which is why most brands buy it for one or two accounts.
• Unlinked baskets are a biased sample, cards are not households, and signup demographics go stale. Adjust the confidence, not just the number.