AI Product Recommendation Engine for eCommerce

Konstantin Sadekov
August 29th, 2021

Read more
product recommendation engine

As artificial intelligence is becoming more and more important across enterprises and industries, we’ve entered an era of intelligent automation. When we think of product recommendation engines in eCommerce, we might think of Amazon and Netflix; whereas considering that 35% of what consumers purchase on Amazon and 75% of what they watch on Netflix come from product recommendation engines, we have a good reason to do so.

But how to make a product recommendation engine for eCommerce that doesn’t fail in the first month? In this article, we are providing a comprehensive guide based on actual cases.

In this article we are going through the following topics

  • What is a product recommendation engine?

  • Why hard-coded engines are failing?

  • Types of recommendation engines.

  • Where recommendation engines can be used?

This problem might be a serious challenge in the world of high competition, as for instance even when consumers love a brand or product, 59% of them will stop using it after just two bad experiences.

Those questions and challenges could be handled by an AI-powered recommender engine. This type of engine can make product recommendations based on millions of data points – producing highly relevant results for the end-user and thus helping to sustain great user experience.

nlp applications

What is a product recommendation engine?

A product recommendation engine is a combination of machine learning models that are capable of matching the right services and products with the right clients at the right time, so as to increase sales, customer engagement and to improve customer experience.

 

How this can be implemented depends heavily on the type of business in question, but the general idea remains the same – provide your customers with the offers they are most likely to accept when they are most likely to accept them.

Not All Recommendation Engines Are “Intelligent” due to Cold Start Problem

Many current product recommendation engines are “out-of-the-box solutions” that essentially try fitting people into boxes based on past behavior. But these solutions are very seriously affected by the so-called cold start problem whereby recommender systems can not make the most engaging product recommendations to the new users simply because they do not have any previous interactions. Also, this problem occurs when brands launch new products: no conclusions can be drawn when no one has actually purchased the product yet. However, there are ways for eliminating the cold start problem.

More advanced product recommendation engines use not only historical data on user interactions, which is often known as collaborative filtering, but also factor in two additional variables:

  • The parameters of the items being recommended are size, shape, material, or whatever information is available.
  • All known information on the person a recommendation is made to, e.g. age, gender, location etc.

This means that even if you add a new product to your inventory, you can still offer it to the people most likely to buy it because you know the product and who are the people most likely to be interested in it.

It also allows you to provide people with novel content. It’s a common issue for recommendation engines to reinforce their own product recommendations, nudging people further and further into an echo chamber. Correlating relevant background information and item properties allows the recommendation engine to look beyond the products a person has already seen and present them with new experiences.

But let’s have a look at how “hard-coded” traditional models differ from AI-enabled product recommendation engines.

Hard-Coded vs AI-powered

 

In a traditional model, the system works on a trigger and rule basis – a user performs a predefined action and the system sends them an offer. This action could be anything from a product view or a click to adding something to a shopping cart – anything that could signify interest.

product recommendation engine

However, this way has many disadvantages. For instance, it does not take into account daytime, changes in consumers’ behaviour, or seasonal fluctuations and trends. All of this requires manual adjustments and is much less efficient when there are thousands of various products. It might be a good solution for a small E-commerce venture, but not for an organization having thousands of products in stock.

product recommendation engine

An AI-powered product recommendation engine is a step forward in both its capabilities and also the data that it can use for its benefit. Some of its benefits compared to traditional would be:

  • matching the right product with the right customer at a specific moment
  • correlations across the entire customer database impact a single sales attempt
  • understanding features & benefits, matching these with complex customer profiles

This allows companies to improve their sales offers and make product recommendations based on location, or past or current activity.

But to really call a recommendation engine intelligent, it should be able to:

  • Know your customer: customers are people and people are complex – useful product recommendation engines are able to take advantage of this personalization opportunity. It should understand what the customer needs at any given moment.
  • Adapt to the changing world: as consumer demands and product selection are in constant flux due to changing trends and preferences, the engine should be smart enough to adapt and ensure high performance.
  • Enable scenario testing: the ability to run simulations against your customer profiles can be immensely beneficial as it allows companies to minimize risks and deliver products that customers really need.
  • Serve many uses: a recommendation engine is not just a device for pushing inventory – it’s a useful internal tool and should enable your business to make better decisions, measure goals and understand your audience.

Types of Recommendation Engines

There are various types of recommendation engines used today. Most of us have probably interacted with one recently, or do so from time to time.

  1. Ecommerce recommendation engine – this is probably the most popular one that is used widely. It is most likely you have purchased something from Amazon, eBay, or an equivalent local platform. In fact, our studies show that on average an E-commerce recommendation engine can increase sales by around 10-20% compared to traditional models.
  2. A content recommendation engine suggests engaging content in various places on a webpage, a good example would be an online magazine or newspaper. It obviously is wiser to suggest different content depending on the age and gender of the person browsing, and even on the time of the browsing session.
  3. A movie or video recommendation engine is obviously one of the most efficient ones. Netflix and Youtube are great examples of how to increase the average spending time on a website.
  4. Music recommendation engine – well, you have probably heard about Spotify, haven’t you?

Even though it sounds simple, companies spend years on experiments until they get the optimal results, but it totally pays off. For instance, Netflix believes that their recommender system is worth 1 billion USD, as matching their customers with the right content helps to almost eliminate any subscription cancellations. Meaning that recommendation engines not only help to increase sales but also solve customer churn problems.

As such projects are complicated, companies often fail when they start building these internally, without a proper data science partner. MindTitan has built various product recommendation engines for telcos, retail, and pharmaceutical companies for their online and offline shops. Talk to our experts and we will help you to come up with the best approach to your problem.

Where product recommendation engines can be used

There are various ways to use product recommender engines

. The most popular ones are:

  1. Online stores. Home page “trending or most popular products”, cart page “people also bought”, product pages “these products are frequently bought together”.
  2. Bricks and mortar stores. Screen terminals at the entrance where a customer can scan their loyalty card and receive personalized recommendations on the terminal screen, or PoS screen, and the client screens at the checkout counter where a cashier either suggests buying a specific product or a client sees the recommendations.
  3. Telemarketing and call centers. The model can suggest which products to offer to which specific customers at what specific point of time. One of our clients managed to generate even as much as 150% growth in telemarketing sales efficiency.
  4. Marketing campaigns like email marketing and banner ads.

Conclusion

With all said, you probably now understand what a product recommendation engine means and how it is used in e-commerce. We have helped several big companies from Japan, Finland, and Estonia to successfully implement recommendation engines and are happy to share our experience. Book a free consultation, and let’s discuss your idea.

Free consultation

Free consultation

Konstantin Sadekov
HEAD OF GROWTH AND MARKETING

Konstantin has graduated from the Estonian Business School majoring in economics and finance, and acquired an MBA degree in the USA. Being strong in business processes and understanding the nature of data science capabilities, Konstantin has helped various companies implement the latest AI solutions.

Contact us

Free consultation

Free consultation
konstantin sadekov mindtitan