Hi Adib, that's great to hear, thank-you for reading.

Loyalty cards tend to be used more often by lower income customers, families, and those who are very regular shoppers, as they may be more keen to take advantage of the discounts and rewards provided by the scheme. This means if you're using loyalty card data to identify unique customers, any analysis you perform on this data will be biased towards the behaviour of this group, rather than the larger customer base. A successful loyalty scheme might capture 70% of weekly customers, but not every supermarket has one that well established, so any analysis must be extrapolated with this bias taken into account.

To deal with this bias, you can identify a smaller representative sample of loyalty card customers stratified by demographic data gathered through the customer's online profile (age, wealth, household size etc.). Analysis performed on this group can then be extrapolated to make conclusions on the behaviour of the entire customer base. However, depending on how biased your loyalty program is, your sample size may end up being quite small, making it difficult to do analysis on individual products or short time periods.

You can improve the coverage of the loyalty card data by also using the payment card to identify a customer. This means that the only transactions you can't attribute to a customer profile are those paid in cash (which is steadily becoming less common), without scanning their loyalty card. As mentioned in the article though, this has its shortcomings, mainly that if someone uses multiple cards, they will show up in the data as several different people.

I hope that makes sense. Thank-you for your question and I'll take a look at adding to that part of the article to make it clearer.

Richard