Customer profiling techniques must learn a customer’s changing preferences and maintain an up-to-date view to continue to make highly relevant decisions as to what to show to the user.
Profiling is everywhere, and many companies, Google, Meta, Amazon, et al., have proved that if it is done correctly, it can be extremely profitable. In 2021, Google’s parent company, Alphabet, generated a staggering $209 billion in ad revenue, up from $147 billion in 2021. This was a direct result of successful marketing and advertising models driven by data, which enabled users to show relevant ads to consumers. That, in turn, greatly increased the chances of an interaction or further purchases. One of the main reasons Amazon continues to be so successful is down to its ability to drive extremely high customer loyalty across its own services through its data-driven profiling capabilities.
Know your customer. Data shows us a consumer’s preferences. Whether that’s where they shop, what they buy, when they prefer to shop, or how they want to pay. By tapping into that detail and insight, businesses can better understand their consumers’ needs, which allows them to tailor their engagement and the experiences they offer. Whether you own that technology or buy it, traditional data models build an expected behavior pattern based on previously observed behavior. These models are used across many industries with extremely high accuracy for static applications, such as identifying a customer as part of a group and using that classification to offer added value. This approach to modeling customer preferences works well when a consumer never changes his/her behavior or if their behavior changes slowly over time and models are retrained occasionally to maintain performance. To combat this, customer profiling techniques have been developed to learn a customer’s preferences and maintain an up-to-date view to continue to make highly relevant decisions as to what to show to the user.
Enrich your data with other sources. It’s important to look at more than just the base data. Additional value can be extracted when a data set is enriched with other data sources. Covid has taught us that outside factors such as public holidays, national or international disasters, wars, and political events may impact behavior and need to be considered, and businesses need to be ready to cope with any eventuality. The pandemic caused huge behavior changes amongst consumers overnight, and there was no way businesses could tune and adjust their profiling models quickly enough to keep up. The products and services consumers wanted were not readily available in the ways they’d been used to, which meant they had to find suitable alternatives and change their expected behavior patterns. We didn’t stop buying food. We started buying it differently, whether we used a different payment type or bought our goods from a different shop.
As a result, Covid taught marketers that they need to include outside factors in any future behavior predictions. Retailers also need to know whether any change is temporary or whether it’s likely to be permanent, so they can adjust their strategy accordingly. Understanding the ‘why’ and not just the ‘what’ when it comes to customer behavior. Other data sources like global and local news, social media, public holidays, dark web forums, and weather can all help with that.
Automate. Automation is critical for ensuring any machine learning, behavioral and profiling models can remain up to date and continue to drive the highest value for users while enabling speed and predictability – two things consumers really like. When a consumer’s behavior changes as rapidly as it did with Covid, retailers can ensure they are in the position to react quickly. Traditional style models can still be used to great effect, but the process of bringing those models to production will most likely need to be re-engineered. That said, there are many approaches, and one may work better than another depending on the situation, but the key takeaway is the importance of ensuring the production model is always up to date. This can be achieved through online learning, where the same model is updated with the latest training data, or through continually training new methods and taking a challenger/champion approach. Either way, the process should be automated as far as possible to enable the models to keep up with the pace of changing data trends.
Fully automated approaches, such as customer profiling, which powers Google’s and Meta’s targeted advertising and Amazon’s purchase recommendation engine, are able to remain up to date with changing customer behavior and don’t require huge training resources like a traditional model does, as only the data for that customer and their recent activity is required to update the model, rather than a much larger dataset comprised of many customers recent activity. This technology combines recent and frequent behavior to make decisions and show relevant content; older behavior-based recommendations decrease in relevancy over time and are shown less and less until they become irrelevant.
Of course, failing to consider outside data (news, weather, etc.) makes these tools less effective – and unfortunately, very few businesses seem to have realized that. In the future, technology that can incorporate additional data sources, quickly and easily, as and when required, and helps to make sense of the world in which the customer lives will win out and provide the best possible recommendations throughout the lifetime of the event (such as Covid) without requiring time to adjust.
Make use of modern compute resources. The term ‘everything is in the cloud’ is becoming much more relevant as this technology has so many benefits when compared to the traditional server approach. Gartner research (2022 CIO Agenda) shows that in the UK, cloud and machine learning technology is within the top 10 investment areas for many businesses. Cloud technology enables profiling solutions to be more responsive and therefore is better placed to meet customer and business needs, due to the highly scalable and always available nature of properly designed, cloud applications.
Quantum computing is already being hailed as the next age in computing due to its incredible processing power. The technology inside the machine you are reading this article on can only go so far. Typically, any performance improvements are associated with an increase in the number of transistors, but the physical limits of PCs and mobile phones means we are close to reaching a hard stop on processing speed improvements. Quantum computing will enable much faster data processing speeds, meaning more data can be processed and more complex models can be developed. That processing power will make any model more accurate and useful in a real-world setting. Having large, enriched data sets readily available and easily accessible, as well as the proper tools to gain any insight, will be key to driving useful insights for retailers. In the future, this level of insight will arm retailers with the knowledge to engage in new, meaningful content and experiences with their users.