Who’s buying which of your products? Which items are being combined in people’s shopping carts most often? How can you utilize recommendations — one of the main tools now being deployed to create the high-quality online experiences customers expect — to improve your sales?

Those are all great business-process questions. Fortunately, the answers to the first two can be wrangled with the help of entity relationship models (ERMs), also known as ER models, entity relationship diagrams (ERDs), and ER diagrams. These pictures, which look like flowcharts of different-shaped symbols connected by different styles of connecting lines, let you grasp the relationship types between your various data items, such as products, customers, and orders.

When you can do that effectively in database tables, you’re halfway there. Then you can then use an algorithm to provide accurate recommendations by predicting your customer preferences.

What is an entity relationship model?

In software engineering terms (thank you, Techopedia), an ERM is a “theoretical and conceptual way of showing data relationships in software development…a database modeling technique that generates an abstract diagram or visual representation of a system’s data that can be helpful in designing a relational database.”

An entity is a noun, a thing; for instance, a product. An entity-relationship diagram features category-associated symbols — rectangles, diamond shapes, and ovals, among others, plus lines that connect them (for example, indicating a many-to-many relationship), to illustrate the connections.

An ER diagram tool also often shows attributes for the entities, for example, age or ID number.

Identifying entities and their relationships, and knowing how and why they relate, is important for a variety of applications. For instance, if you have a retail website, you want to figure out how to show product recommendations that will help your shoppers get what they came for and turn into repeat customers.

The origin of the ERD diagram

Here’s a quick look at how entity relationship diagrams made their debut:

In the 1960s, American computer scientist and industrial researcher Charles Bachman came up with data-structure diagrams.

In the early 1970s, three types of data models for databases were being used, called network, relational, and entity set. The problem: none of them offered a complete view of a database. 

So in 1976, a computer scientist interested in improving existing database design, Peter Chen, came up with the concept of an ERD. He wanted to add clarity for the database terms being used to describe the entity relationships. His new model eliminated ambiguity and  unified the database modeling framework.

Chen’s model served as the foundation for Unified Modeling Language (UML).

In the 1980s, James Martin, also a computer scientist, refined Chen’s work and introduced IE notation, which uses crow’s-foot notation to show one-to-many relationships (cardinality).

More recently, entity relationship models have been utilized by AI pioneers. By mapping the relational data of entities, analysts and marketers can begin to see patterns and trends in user behavior that facilitate the prediction of user preferences and help produce the right recommendations.

Types of ERD models

There are three types of diagram when it comes to entity relationship models: conceptual, logical, and physical. Here’s how to tell the difference between these styles of data-flow diagrams:

• A conceptual data model is the most abstract; it’s a generalization of what’s going on. 
• A logical data model portrays data elements in detail and, through visuals, facilitates understanding of data relationships, entities, attributes, and keys.
• A physical data model is detailed, identifying the primary keys and foreign keys of entities. This style is used as a blueprint or working design.

How does an entity relationship model help power recommendations?

So now you know the basics of entity relationship models. What’s probably more important to you, though, is knowing that when you can collect enough customer data and put together an effective relational model, you can generate nuggets of highly useful information. You can clearly see patterns documented in black-and-white diagrams, and you don’t have to guess at things like which items, such as potential add-ons at checkout or also-interesting news stories, your customers might love to know about.

The first step, gathering all that detailed data, should be relatively straightforward. Why? According to Accenture, the majority of customers are willing to give you their data, including their online activity, purchase history, likes, and dislikes, if it leads to better experiences for them. Of course it’s all about them. But who cares? When you get your hands on this invaluable information, you can then get ready to build yourself one powerful recommendation engine. 

So how do you transform people’s online meanderings and website-poking-around preferences into intelligent, intuitive recommendations that will delight and entice them? In other words, how can you unleash spot-on suggestions to improve your customers’ online experiences, increase your conversion and average cart value, and yata yata?

How AI recommendations work

Want to know how artificial-intelligence-based recommendations work so your organization can figure out how to effectively deploy the right setup?

OK, let’s say you have a movie-subscription website. That means your recommendation engine must decide which flicks your viewers could most enthusiastically want to devour next on their watch list. There are two ways for a recommendations engine to discern this kind of information: 

Content-based filtering: just movies

This approach takes a customer’s preferences (likes, dislikes, user activity) and, based on their movies, endeavors to make helpful recommendations.

Well, not any kind of recommendations, needed though they might be. If one of your customers is watching The Emoji Movie five or six times a week, for instance, you might think the recommendation engine would go out on a limb and suggest a therapist. No such personal advice functionality exists there (yet), but based on tags such as genre, subgenre, director, actor, and movie length, the software would at least be able to recommend other animated family movies, which could potentially break the addictive viewing streak and restore some couch-potato normalization.

Collaborative filtering: movies plus other people

This second type of smart recommendation takes into account the preferences not just of the viewer in question but of other users on your site. By making intelligent connections between what different people rate highly, the engine can offer up accurate recommendations to seemingly similar movie buffs. The movies’ content is still key in driving this method, but there’s a lot more going on behind the (movie) scenes, joining up the dots between what similar users in the community choose to watch next, what they like, and what they dislike. So based on various data points, someone who’s been engrossed in The Emoji Movie might next be nudged toward Despicable Me or Toy Story 4. 

For either of these methods to work, someone or something needs to make some pretty sophisticated connections between the data points being captured in large databases. Fortunately, our AI robot friends can do this in their sleep (or at least while we sleep), that is, if they have the requisite data.

So here we are back at good old entity relationship models — those information-systems tools whose comprehensive views of data make educated-guess recommendations possible. Whether the content-based or collaborative-filtering approach is best for your business, ERD can help you map out relationships between entities and alight on just the right recommendations. All it takes is the right tools.

How to provide A-rated recommendations

Because websites like Amazon are continually upping the ante, a stellar customer experience matters now more than ever. That means adding highly personalized, hit-the-mark recommendations to your website or app would likely be a win.

For this reason, Algolia Recommend enables fast, scalable product discovery through an enterprise-grade engine. You can use our proven API to unlock the full potential of user recommendations, including for:

• Related products, which could be even better suited to what your customers wanted when they first landed on a page
• Frequently bought together items: learning that real-world shoppers have shelled out for an item can make people feel justified in adding it to their carts, too
• Related content they may find worth reading or engaging with, which keeps them on the site longer (always a good thing)

Want your programmers to provide the recommendation experiences your users would love? Learn more now about handling your data relationships and how our popular solution can deliver outstanding recommendations that drive higher conversion and revenue!