You have more ‘loyalty’ data than you think: leveraging Customer Data without a loyalty card

Foundational Customer Data Science

The arrival of loyalty data (in the mid-1990s) enabled dunnhumby – the first of the Customer data scientists – to follow shoppers through their purchasing over time. This so-called longitudinal analysis meant we could see which products were bought again and again by the same Customer, and what choices Customers made to substitute a product in the event of out of stock, change in assortment, or change in price or promotion, for examples. Understanding product substitutability became the bedrock of all Customer-led category science, unlocking the ability to evaluate, optimise, and predict retail decisions around assortment or price & promotion.

Exploiting every ounce of data

Of course, not all retailers desire to offer loyalty programmes or cards, depending on their strategy or Customer value propositions, and particularly in discount and convenience channels.

Moreover, even at best, the typical retailer sees 70% of sales going through on loyalty cards (and often considerably less for convenience formats or certain Customer segments). Squeezing every bit of available data for use is critically important, and so gone are the days when the remaining 30% of ‘ePOS’ data would be simply thrown away.

A blended approach wherein the best of ePOS science and the best of Customer science together is superior, and increasingly necessary for making difficult retail decisions in an ever-disrupted retail environment.

Accordingly, data science has had to evolve to derive product substitutes without the personally identifiable and longitudinal data that was originally fundamental to the science itself. Fortunately, science and machine learning has evolved considerably as well. For example, dunnhumby seized on the hot new science called ‘vector embeddings’, already used widely in natural language processing, via our Academic Partnership programme.

Cool and powerful new ePOS data science

Vector embedding techniques are used widely for search terms and document retrieval. The ‘vectors’ are like fingerprints in that each word has a unique numerical string (vector) depending on the context it is found it relative to other words in the same document or webpage. The idea is that words with closely matched fingerprints (similar vectors) have similar meanings.

Now here’s the cool bit – we take the concept and apply it to the world of grocery shopping, but instead of a word in the context of a document, we have a product in the context of a basket. In looking across the contents of all the baskets where we find a given product, we create a fingerprint for that product. For example – we find that a certain type of coffee is often bought with orange juice and croissant. In the simplest sense, orange juice and croissant, and the absence of other products, would become the fingerprint for this product. And if we look for products with a similar fingerprint, we are likely led to other coffee products.

Imagine greatly fine tuning this science on thousands of products in millions of baskets; not only can this science become extremely powerful, but arguably even more powerful than loyalty card-derived methods alone.

Our legacy in Customer data has made us experts in non-personalised data

This type of unsupervised machine learning is not straightforward to implement. In category science, for example, there is no absolute truth to point the machine to, no true north, no objective function against which to model or measure the new results. For example, some put Pepsi Max as the best substitute for product Coke Zero, whilst others argue for Diet Pepsi; either can be correct.

dunnhumby have two great advantages in 25+ years of experience with hundreds of category management experts in the top retailers all over the world, and in a well-proven substitution science based on personalised Customer (loyalty) data. This meant that we could compare ePOS substitutability versus our best-in-class Customer substitutability and challenge the results for both. The comparisons concluded that ePOS substitutability science well complements and supplements existing Customer science.

Cool Convenience channel applications of ePOS data science

Convenience store operators face unique challenges over their supermarket chain counterparts, like different ranges and sets by store or franchise, and this creates data complexities and a kind of ‘chaos’ across the network. The new ePOS data science turns this chaos to an asset rather than a barrier, by using machine learning techniques like recommender algorithms to learn relationships in volatile environments.

Collaborative filtering is a type of recommender algorithm used by many websites to suggest products to their Customers, e.g., “shoppers with similar interests to you also liked product X.” In effect, these algorithms learn from chaotic product interrelationships to understand “people like me”. In the new ePOS science, similar recommendation algorithms learn from other convenience stores how well a product performs in the context of a certain category range, and the resulting science delivers a “store like me” recommendation that outperforms the typical approach (which only looks at top selling items, for example).

Don’t underestimate the value of non-loyalty ePOS data: Key takeaways

• ePOS data accurately describes Customer behaviour if you know how to look for it – so you have more ‘loyalty’ data than you think
• ePOS data can be mined to improve the Customer experience and grow loyalty, even if you don’t offer a loyalty card programme
• ePOS data can be and operationalised and monetised in ways similar to loyalty data
• If you do have data from a loyalty card, ePOS data science can fill in the gaps
• ePOS science has become arguably even more powerful than loyalty card-derived methods, depending on format, channel, or value proposition.
• ePOS data science is a build on loyalty data, not a regression from it

dunnhumby have over 25 years of expertise in transformational ePOS customer data science, working with Discount, Convenience, and customary supermarket operators without a traditional loyalty programme. Get in touch with your local dunnhumby team to discover how you could maximise your ePOS data to better understand your Customers.