Transforming shopping history into customer insights
Reading customers’ minds is not easy. But with so many unique lifestyles, dietary trends, constantly changing preferences, and new brands introduced every week, it looks like retailers don’t have any other choice. The good news is that with all the available customer data, it should be a piece of cake. But is it?
Back in 2013, a group of researchers from Microsoft and the University of Cambridge created a model that was able to predict highly sensitive data of respondents such as race, religion, political orientation, relationship status and even sexual orientation. All of that just from 68 Facebook likes per respondent with a probability ranging from 82 – 95%.
Remember, this was in 2013 and the model only looked at 68 likes on average (which, for many, is roughly a 2-day dose). We can only speculate what can Facebook learn about its users today, with years of data on their online behavior, friends, messaging, location and shared posts.
Similar to likes on Facebook, shopping behavior tells a lot about us: the specific products we buy, the combinations and timing in which we buy them, the way we react when something is on sale at a discount, the frequency and size of shopping trips, and much more. All this data is freely available to every retailer that has some kind of loyalty program in place.
So it’s no surprise that retailers have been trying to leverage shopping data for decades, with AI being a synonym for innovation in the industry. But even today, after tremendous progress in machine learning and neural networks, most retailers keep using simple systems based on predefined rules. Such rules can look for example like this:
- No meat in any of the last 10 shopping trips = vegetarian
- Children’s products in the basket = household with at least 1 kid
- A purchase of a body lotion brand positioned for eldery women = age: 50+
There’s nothing wrong with this approach, and a rule-based system can achieve respectable results. Most of us have heard the story about how Target figured out that a teen girl was pregnant before her own father did. But observing big, life-changing events is something different than those small, almost unspottable nuances in the shopping behavior of people of different ages, gender, income or shopping potential. In these cases, rule-based systems are far from being precise.
Attribution algorithms derived from panel data
To predict customers socio-demographics as well as behavior inside and outside of retailers’ stores, a truly advanced AI solution is needed.
To prove that this is possible, we have developed Attribution Plus — a tool consisting of a set of machine learning algorithms trained on household panel data provided by GfK. This data includes both — personal information of respondents (e.g. age, income, food preferences…) as well as their shopping history across multiple retailers. This allows the machine learning algorithms to compare the shopping behavior between different customer segments (e.g. different age or income groups) and spot even the slightest differences.
These differences in shopping behavior are represent by models that are the key component of Attribution+. A retailer simply imports anonymized shopping data from the CRM database and the tool assigns to the customers missing information, such as age group, household size, household income, food or brand preferences.
Because the GfK panel data tracks purchases of respondents across all retailers where they shop, Attribution+ can also predict how customers behave outside of retailer’s own stores:
- Shopping potential – what percentage of the monthly grocery budget the customer spends elsewhere
- At which other 3 retailers the customer shops
- What top categories the customer buys there (aggregated)
Having this intelligence for each individual customer in a retailer’s database creates a wide range of possibilities. A retailer can better target customers and improve the ROI of brand promotions, resulting in better deals with CPG brands. With market insights on what their customers buy elsewhere, a retailer can design promotions to reduce churn and win over shoppers from local competitors. By grouping customers with high potential, retailers can also discover new segments and find ways to attract them.
Although the tool is only available in Germany at the moment, many other countries will follow. For more information, visit the official product webpage.
The SO1 Engine is an AI behind Attribution+ that also permits automated personalization of product recommendations delivered to a channel of the customers’ choice. Find out about this technology at www.so1.ai.