The more quickly customers can find the product they are seeking, the more likely they are to complete a transaction and to return to the site in the future. Personalizing offers and making well- targeted recommendations can bring customers and products together faster, and are effective ways to engage customers by creating a more positive customer experience. In order to do this, companies need to capture and use as much relevant information as possible. The more that is known about the customer, the more effectively the recommendation system works. Customers generate many signals through their online behavior, and those signals can also be used to understand their interests, purchasing patterns, and needs. Reading their digital body language accurately and creating a valid customer model is essential to anticipating and fulfilling those needs.
Personalization and recommendations have a lot in common, but they are not identical. Personalization means tailoring a communication in a way that is unique to that customer; for example, by using their first name, incorporating information about their location, or presenting content that they have previously indicated is of interest in response to a search.
In contrast, recommendations are explicitly stated, and are typically presented as suggestions: “You might also like…” or “often bought together…” based on product relationships or populations with similar characteristics. Not all forms of personalized interactions include a recommendation. However, they both rely on information about the customer, the products or services offered, and other contextual information.
Different types of information about the individual can be combined, and then used for either personalization or recommendations. For example, past purchases and the knowledge that the individual has a technical background can guide the selection of products to display or recommend. Information about similar customers who have made purchases can also be used to make offers that provide the best match to the customers’ needs.
Some forms of personalization are simple, while others are more complex, such as anticipating what a customer needs at a particular stage of the customer journey. In the latter case, it’s important to think carefully about objectives—why are you collecting the information, can you determine where the customer is in the decision making process, and do you know enough to achieve the objective?
Given the impossibility of collecting every potentially relevant data point, models are needed in order to help determine what data should be captured and how it should be organized. A customer looking for a set of products might think of them alphabetically; for example, the supplies needed to paint a room (masking tape, spackling, paint brushes, paint thinner). More likely, they will group them mentally by task, which is why many websites say “What are you trying to do?” and then create a list of products. Ideally, the website model will mirror the customer’s mental model, to make it easy for the customer to find what they want and complete a transaction.
Product data models provide details about each product, based on attributes and metadata. They show product categories and relationships between different data points and among different products. These models allow customers to navigate to or search for products with certain characteristics, such as brand, horsepower, and price.
Customer models and product models may change over time, as new insights are gained as to what data customers want. Each touchpoint and metric provides new information; for example, does it take longer for a customer to purchase their product if they use the navigation function or if they use the search function? Continuous testing allows for improvement in each type of model.
In general, the use of statistical data is the simplest form of personalization. An example of the use of statistical data for personalization is shopping basket analysis—you look at prior purchases, combinations of products that are purchased together or have things in common, or that are purchased by related types of populations. You can gain a finer degree of granularity by narrowing the population to engineers, if you have that descriptor in you attributes. Then you can start to segment the shopping basket analysis in those groupings according to other characteristics.
Purchases can be clustered in different ways, and aligned with other types of purchases. When there is a missing piece of information for a particular person that is needed to make a good recommendation, you can look at a population similar to the individual. If the data shows that typically people in this group purchase three related products, and the individual has only purchased two of them, you can then recommend that third product to the individuals.
Statistical data may also relate to customer characteristics. The more you know about the individual, the greater the likelihood that you can personalize results or recommend appropriate products. If you are willing and able to collect extensive customer data, it’s possible to get a delineation between one type of customer and another, and differentiate offers to each group.
Personalization can also be accomplished by using structural relationships among individuals or products. Sometimes characteristics across groups may not be readily discernible by a human, but a computer or a machine learning algorithm can find those relationships. These include latent attributes that can often be detected on social media, such as location, travel plans, or political leanings. This information is more subtle but provides clues about the customer’s data fingerprint. Information about connection to others can also be obtained from social media, and that information can be useful in adding attributes that provide a more detailed profile.
Developing product relationships requires a fair amount of involvement from product and marketing teams. Ultimately, establishing these relationships increases conversion and revenue across the website and other channels. It will do this by prompting more return visits and generally a higher level of customer satisfaction because of a better user experience. For example, building in good structural relationships allows the system to find replacement parts or make recommendations for substitutes. Better recommendations generally lead to higher satisfaction scores and NPS scores, since the overall discovery and navigation as you're moving customers through a product catalog is improved. Customers can be exposed to different parts of your product catalog to show a breadth and depth that they weren't previously aware of.
Real-time data allows personalization and recommendations in a much more dynamic way. The marketing stack, which is the marketing technology ecosystem, is what allows us to capture those real-time digital signals of interactions throughout the customer journey. This data can then be combined with the profile data, and can then provide better personalization and more precisely targeted recommendations. Compared with statistical or structural data, real-time data changes much more rapidly.
The signals come from different touchpoints in many different systems, to detect the digital body language. The question is then what can be done with this information so that they can be incorporated back into the customer model. Customer data models can be built just like product data models can be developed by understanding product relationships. The ultimate goal is to combine the two, so that the audience attributes can be matched with the product or content attributes. Therefore the metadata that describes the attributes becomes essential, as well as the ability to orchestrate the various types of data.
Location is one form of context, so a recommendation can be made about “something near you.” The individual’s industry is part of context, so products that are a match for that industry are candidates, or matched to the individual’s demographics. But things that are moving more quickly, like downloads and searches and conversions, or an abandoned shopping cart, provide instantaneously relevant information.
Then an offer can be tailored for that specific customer based on where they are in their journey. In some cases the response might be to offer a better product match, ask a question, or just to re-engage the customer who abandoned the shopping cart so they continue their search. What you are really trying to do is represent that user and that context and those preferences, all in terms of their data. Your systems should be able to interpret the data and respond to it.
Once the customer data model is built and tested with certain types of offers, performance metrics need to be captured. A metrics-driven decision-making model helps determine if the assumptions about customer and product models are working. Scenarios and various use cases should be designed to test out whether the predictions for what a customer wants or needs are valid. Multiple iterations will typically be needed in order to test changes that are implemented based on metrics. Typical performance metrics are NPS, customer satisfaction scores, reviews, conversion rates, and revenue per transaction.
The most important thing about metrics is that they should be acted on. There is no point in collecting data and then conducting an analysis that does not result in action. The action might be to change the product hierarchy to better match the customer’s method of searching, or to provide additional product details. The more these are aligned, the more effective product recommendations can be, and the better the outcome for customer and company alike.
If you need help setting up effective recommendation systems, contact us to set up some time to see how we can help.