25 Use Cases for AI in eCommerce

It’s easy to tune out when you hear “AI.” 

The same goes for “machine learning,” “NLP,” and “personalization.” So many e-commerce companies today tout their ability to serve next-level experiences through artificial intelligence that we’ve begun to ignore the words completely. 

Ignoring artificial intelligence in eCommerce, however, might not be the best idea. 

While many tech vendors do use black-boxed “AI” to woo unknowing potential customers, the decision to integrate real artificial intelligence algorithms into your eCommerce site could play a huge part in your success in the coming years. 

Why eCommerce Businesses Are Integrating AI & Machine Learning

In a world where hundreds (or thousands) of similar eCommerce companies are competing against each other for the attention, loyalty, and sales of their target audience, everyone is searching for a way to stand out. 

They’re seeking a competitive advantage where it’s incredibly difficult to find one. And that’s where AI plays its part. 

AI Can Make Customer Experiences Seamless

Search and browse experiences are often some of the most under-looked aspects of eCommerce websites. 

Sometimes your users don’t know exactly what they’re looking for when they come to your site. Sometimes they misspell search queries and receive little to no results. And even if they don’t, many of the products that appear for a search query may not appeal to their specific wants and needs. 

Certain facets of AI — like natural language processing and machine learning-enhanced re-ranking — can dissolve these issues without any human interaction. 

(We’ll talk more about each of these facets shortly.) 


Want to learn more about serving users attractive search results and the limitations of optimizing search for “relevance?” Check out our recent webinar here.


AI Can Increase CTR and Conversions

According to Boston Consulting Group,

“Companies that use advanced personalization methods can realize an improvement of 20% or more in their net promoter scores. In addition, these retailers can see productivity gains of 6% to 10% and incremental revenue growth of 10% or more.” 

Furthermore, according to Accenture,

“AI has the potential to boost rates of profitability by an average of 38 percentage points and could lead to an economic boost of US$14 trillion in additional gross value added (GVA) by 2035.” 

AI Frees Up Time

Let’s use one of our examples from before: 

In any eCommerce site, there are hundreds, thousands, or hundreds of thousands of search queries that should present a significant number of results, but don’t — whether it’s due to a misspelling in the search query, the use of a synonym that isn’t detected by the search system, or something similar. 

In the past, merchandisers and engineers would have to manually set up search redirects for each of these queries. With constant new additions to product catalogs and user behavior, this became a never-ending process. 

AI has the power to do all of that for you. 

AI Allows You To Make Better Decisions, Faster

In much the same way merchandisers and engineers have to set up redirects for null search queries, they’re also tasked with knowing which products to merchandise at certain times in the year, which products to rank highest for any given query, and more. 

With algorithms that track how well certain products sell for any given query or for any given time period within the year, these decisions become much easier (and with the right algorithms, even automatic.

The Data AI Collects Can Be Used Everywhere

AI isn’t only used to improve on-site customer experiences. 

The data AI can gather about your customers can be used to serve relevant content and offers to them across multiple touch points — like email marketing, advertising, and more. 

Image Credit: Olark

Use Cases for AI in eCommerce

Search

NLP and Autocomplete

According to Nacho Analytics, autocomplete can boost conversions by up to 24%.

Good autocomplete, however, does more than estimate basic user queries. E-commerce giants are now combining Natural Language Processing (NLP) algorithms with autocomplete to present query results that approximate user intent. These algorithms are commonly used to:

  • Correct phonetic misspellings, keyboard proximity typos, punctuation nuances, and omitted character typos. Is it “blue ray”, “blue-ray” or “bluray?” It’s officially spelled “blu-ray”, but you shouldn’t require users to know this to get good results.
  • Detect synonyms to serve relevant results to users. If a user is searching for “suntan lotion,” good autocomplete will recommend “sunscreen.”
  • Identify word importance to better understand what results will most likely lead to a conversion.
  • Add support for multiple languages to give the best search experience to any visitor.

For performance reasons, autocompletes often use separate indexes or more advanced optimizations. For some helpful advice your engineering team can use, check out this article on how we’re continuously optimizing our trie-based spell corrections at Constructor.

Automatic Synonym Implementation

Imagine you have “sunscreen” in your product catalog. 

If your users look for “sunblocker,” there’s a good chance they aren’t going to find any results (especially with rudimentary search systems). Many of these users will then leave the site — but some may refine their search to test other queries. 

In the past, merchandisers or engineers would have to manually set up redirects for each of these potential synonyms. Good AI systems, however, have the ability to automatically detect when null result pages are found for a query and monitor what users refine their searches to (and what they convert on those refined searches). They can then set up redirects automatically (given the two queries are similar).

No manual work needed. 

Results Re-ranking (or “learning-to-rank”) 

Search systems backed by machine learning can automatically re-rank search results for products most likely to lead to a conversion.

A basic example of results re-ranking is as follows:

If 40% of users who search for “laptops” purchase the laptop in the 8th search position and 22% purchase the laptop in the 5th search position, it would make sense to re-rank the 8th laptop higher (perhaps in one of the first few positions) and the 5th laptop closely following. Other laptops that users don’t frequently purchase should be moved down in the search results.

Learn-to-rank systems are fantastic for ensuring your users not only see relevant results, but attractive results too. 


If you’re interested in learning more about attractive results and the limitations of optimizing search for “relevance,” check out our recent webinar. 


Great re-ranking systems, however, don’t only re-rank exact-match product results. They also re-rank attractive products.

Take a look at these results for “baby carrots” on Amazon for example:

“Hummus” ranked for the “baby carrots” search query.

Although “hummus” was not part of the search query, Amazon knows their users frequently purchase hummus along with baby carrots, so they’ve included it in the results. But that’s not all.

Amazon found it profitable to rank “hummus” for the “baby carrots” query not only because their users tend to buy hummus with baby carrots, but also because when users see the hummus, they can better picture themselves eating baby carrots with a tasty condiment — thus making the baby carrots more attractive.

Voice Search

With the growing popularity of devices like Alexa and Google Home, the importance of implementing natural language processing voice search is growing. Here’s why: 

Let’s say a user is trying to find “red eyeliner” on your site. If they’re searching from a computer, they’ll probably just search for “red eyeliner.” If they’re using voice search, however, their query may look closer to “show me red eyeliners.” 

Great natural language processing systems know that “show” and “me” are stop words that should not be included in the actual search query. Instead, they’ll focus on the actual product the user is trying to search for — red eyeliner.

Visual Search

Sometimes it’s hard to describe an item we’re looking to purchase. We know what it looks like, but we don’t know exactly what it’s called. This is sometimes the case for users who search on your site as well. 

Visual search systems allow users to simply upload a picture of the item they’re looking for to receive relevant results using image matching algorithms.

Shutterstock, a popular stock photography website, has a system for users who are looking for images similar to ones they’ve seen previously: 

Facet Re-ranking

With basic search systems, when a user searches for an item, the navigation facets available to them will stay in the same order — regardless of the query. 

Search systems that use AI, however, can make the item navigation process for users much simpler.

For instance, if a user searches for “organic milk” on a grocery website that uses AI, the facets that appear under the “Nutrition” filter can be automatically re-ranked to display facets the user would be more likely to use based on their query for organic foods: 

 

“Milk” search before facet re-ranking

Once the user tells our personalization systems they like organic milk, facets like “Organic” and “Low Fat” are automatically ranked higher in the navigation.

Smart Autocomplete Using Previous Search Data

What types of products do your users usually search for, and how do they search for them? How can you customize your user’s search experience from those habits?

Google does an amazing job at customizing search (and more importantly, autocomplete) based on past searches. For instance, if I search for “breaking bad poster” in Google, complete the search, then type “b” into the search bar, all the results shown are related to “Breaking Bad:”

Personalization

According to Forrester, “77% of consumers have chosen, recommended, or paid more for a brand that provides a personalized service or experience.”

Personalization should affect every facet of your search — from your auto-suggest results to the products you rank for any search query. In fact, it can influence all the previous examples we presented. 

If a user tends to search and purchase organic foods, your search should automatically re-rank organic foods to the top for every query. Your facets can be re-ranked as well to display organic (or other related facets) near the top of the filter selection. Your autocomplete should use user data to inform what search recommendations it shows to users. 

Moreover, great personalization is more than simple segmentation covering demographic data like gender or age.

Personalization is about learning your individual customers’ preferences based on their historical data and previous interactions with your site. With every search, click, bounce, add-to-cart, and purchase, your users are telling you what products they prefer to buy. It’s your job to show them more of those products.


Want to learn more about serving users attractive search results and the limitations of optimizing search for “relevance?” Check out our recent webinar here.


General Cases

Personalized Product Recommendations

The data you collect on your users’ click-stream and purchase data can be used for more than personalization within the context of search. 

Amazon, for example, serves personalized product recommendations on nearly every page on their site (home page, product pages, etc.): 

They also recommend complementary items to users who’ve viewed or purchased similar items:

Image Credit: Amazon.com

General Product Recommendations

While gathering extensive data on your users’ click-stream and purchase behavior is incredibly useful, there are also plenty of simple data points you can use to segment your site for new and returning users. 

Age

What types of products do users in different age groups tend to purchase? Your AI can use this information to customize the trending products you show on your home page, the categories you recommend, the products in those categories, and more.

Gender

If you know the gender of your user, your AI can also customize which products they see across your site.

Gender re-targeting can be used to redirect users to product pages they’re most interested in viewing, like in this example from ASOS:

When I visit asos.com, I’m automatically redirected to their “Mens” page.

Geography

Do users in different parts of the world purchase different types of products? Do they purchase different types of products at different times of the year? Can you recommend products with smaller shipping costs based on the locations of your users? Your AI can learn all these data points and optimize accordingly.

Predictive Pricing

Altering product prices based on customer demand or elasticity isn’t new (think airplane tickets). 

The data we gather from our users and our competition, however, can be used for much more than basic on-the-fly price altering. According to Ignite Outsourcing, 

“AI can automatically set the highest price a given customer is likely to accept and still buy the product. Prices presented to one customer may differ from what is shown to another, or from what is shown to the same customer at a different time.

Deep learning allows AI to make an informed hunch on what you are willing to pay for that hotel room or set of aluminum tumblers. And chances are it knows before you do.”

Below are some basic examples of how eCommerce companies today are using dynamic pricing: 

Predicting the Optimum Price for Each Product

In the same way AI systems can alter product rankings for search queries based on user reception, it can also alter the pricing of certain items based on how likely users are to purchase products they see at specific price points (as well as the highest price point they’re willing to purchase at). 

Predicting the Optimum Price Presented to Each Customer

If you know a user tends to purchase a specific product within a set price range, dynamic pricing AI can alter the prices of similar items to fit within that range in order to increase the likelihood of a conversion. 

Furthermore, if you know a user tends to purchase organic foods (which typically cost more than non-organic foods), dynamic pricing AI can continuously alter the prices of different organic foods to find the maximum price at which the customer is willing to pay for those items. 

In other words, dynamic pricing AI can serve personalized prices. 

Image Credit: Chicago Booth Review

Predicting the Optimum Price Based on Date or Time

Do your individual users tend to purchase certain items at certain times of the day or week? AI may be used to discount the prices of those items during times the user doesn’t typically purchase to get them back into the store and shopping. 

(We’ll talk more about using AI with marketing and sales shortly.) 

AI can also be used to decrease or raise the prices of items based on demand around holidays, birthdays, and more. 

Predicting the Optimum Price Based on Geographic Location

Are certain items priced differently in different areas of the world? Are there opportunities to change the price of items based on differing demands in different areas? 

AI can both gather data on what competitors in different areas charge for products as well as determine what items increase in demand from certain areas in the world at specific times to alter prices profitably. 


We don’t dive deep into the technical aspects of building predictive pricing algorithms; however, if you’re interested, feel free to check out these papers on the subject: 


Rating Reviews

Recently, some retailers have received false or dishonest reviews on products with the goal of artificially boosting or reducing the trust users have for products. Your users are intuitive — if they suspect a product has false reviews, they’re much less likely to trust it. 

Image Credit: Yahoo Finance, Amazon

AI can be used to detect and filter out dishonest reviews. According to ReadWrite, 

“Even Amazon has come up with a new machine learning algorithm that can rank the genuine reviews and ratings on the top of the list. Amazon identifies honest reviews by analyzing different factors such as upvotes, verification of users, and recency.” 

Product Grouping

AI can also be used to automatically detect which products users frequently purchase with other products and serve them as recommendations: 

Image Credit: ASOS


Want to learn more about serving users attractive search results and the limitations of optimizing search for “relevance?” Check out our recent webinar here.


Chat bots

According to Ignite Outsourcing, 

“Using deep learning and natural language processing (NLP), AI chat bots do more than give rote replies. They give thoughtful answers. Compared to a traditional chat bot, one powered by AI is a different animal.” 

One of the best examples of a human-like chat bot is from Ebay’s “ShopBot”:  

Image Credit: Ebay

Apart from serving more human-like conversations to users and helping them find products they’re looking for, AI chat bots can be used to collect data about what products your users tend to purchase. This data can be used to personalize their experience on your site and serve more relevant product offers. 

Product Descriptions

AI is great at saving time — and one of the most time-consuming processes merchandisers and copywriters go through when expanding product catalogs is writing product descriptions. 

More and more strides are being taken every day in the world of AI-generated, human-like text: 

“AI can do more than copy product information from the manufacturer’s site. It can scour the Internet to find the most recent and most relevant details about the products an ecommerce store offers. Since AI can see what details are most often included in keywords, AI content software knows just the right details to include to the hype that wi-fi-enabled food processor. The descriptions it produces contain the exact information buyers are looking for.”

While AI-generated product descriptions haven’t yet advanced to the point of industry-wide acceptance, be on the lookout over the coming years for advancements in this area. 

Other Use-cases

Email Product Recommendations

Like we said before, the data your AI collects on your users can be used for more than in-store product recommendations — it can also be tied with your marketing activities. 

One of the best ways to engage customers through email is with timely product recommendations that match their product preferences.

Image Credit: ClearVoice

Moreover, AI can help you: 

  • Send emails automatically at a time when the emails are most likely to be read by customers
  • Collect further data on customer behavior based on their responses to emails (did they open? Click? convert?)
  • Further build your profile around each individual customer

Product Recommendations in Push Notifications

In addition to email, personalized product recommendations using customer data can be sent through push notifications: 

Image Credit: WebEngage

Push notifications work great for reducing the time it takes for a customer to purchase products they’re looking for:

“The next time a customer is browsing iPhone cases on your website, they may receive a push notification on their mobile, informing them about your flash sale for iPhone cases. They directly make the purchase on their phone, saving a lot of steps for both parties.” 

Inventory Management and Sales Forecasting

“With the data availability of previous purchases and buying behavior of the users, analyzing future sales can be done more accurately and efficiently. You can also strategize the marketing plan for certain products during a specific period. 

Also, with a clear insight into the product demand, you’ll be able to manage the stocks accordingly. Hence, with demographic user behavior and weather, analytics can help to manage inventory better.”

In short, using AI for inventory management and sales forecasting can help you: 

  • Analyze previous, current, and projected sales throughput
  • Predict, report on, and quickly solve vendor issues
  • Predict changes in customer demand
  • Analyze market changes that could affect sales

Want to learn more about optimizing on-site search to drive real business metrics?

For years, retailers have optimized search solely on “relevance,” hoping the results that users want to see (and the results that drive important business metrics) appear at the top.

This is exactly wrong — and it’s not how companies like Google and Amazon optimize their search.

Learn more about the limitations of optimizing search results for “relevance,” and the 3 attributes retailers should focus on to drive real business results from search in our new webinar:

Previous

Ecommerce Search
E-commerce Merchandising: Everything You Need To Know

Next

Ecommerce Search
Searchandising: Everything You Need To Know To Succeed
Webinar: Constructor's November Product Updates Pre-Register Now