Using Artificial Intelligence to Infer Visitor Intent

For many years, user-level data has been the go-to solution employed by eCommerce sellers for inferring visitor intent. However, recent developments in data privacy regulation mean that this method is becoming untenable. Using AI alongside session-based intent analysis can provide a startlingly accurate picture of your visitors’ intent, permitting you to predict their actions and target them with tailored promotions and content. Let’s find out how this works, and how to make it work for you.

What is Visitor Intent?

Visitor intent refers to the actions a user is likely to take on your website. For example, in an online store, this could include aiming to buy a particular product immediately, performing research ahead of an undetermined future purchase, visiting a website for an update on delivery, or finding out more about a product they’ve already bought.

Session vs User-Based Intent 

Each visit to a website is typically referred to as a ‘session’. When we can identify multiple sessions by the same ‘entity’ (such as a specific email address or device), we would refer to this collection of actions as ‘user-level data’. When we have a lot of user-level data, it can help us infer their intent in a specific session (e.g., if they bought a tennis racket last week, they might be looking for tennis shoes this week).

While user-based inference used to be standard, the method is growing more difficult as visitors become increasingly anonymous. Ad blockers and the gradual abandonment of third-party cookies by major browsers mean that by 2025, this kind of intent analysis will only be possible for users who sign into profiles – and they are few and far between. According to various estimates, only 15% of users remain logged in throughout their online shopping sessions.

Without user-level data case, we are left with session-based intent – meaning we need to use the signals about the same browsing session or visit to make an educated guess as to who the visitor is, and what they intend to do. This means acting on less data and making the prediction in a shorter timeframe – according to Databox benchmarks, the median session duration is just 92 seconds for B2C websites.

How AI Helps Predict User Behavior Based on Session-Level Data

Fortunately, while user-based intent is becoming more difficult to predict, session-based methods are becoming more advanced. Within each session, it is now possible to collect hundreds of privacy-respecting data points that can segment users into diverse cohorts, after which AI can tie these points into an intent-related prediction.

For instance, in eCommerce, that might be an affinity to purchase. Data points such as whether a user uses incognito mode, their coupon extensions, or onsite behavior, are rapidly analyzed to determine this. If a potential customer has a high affinity to purchase, AI targets them with smaller, or delayed promotions to save on an already likely conversion.

In gaming, it might be the probability of an in-app purchase. If a customer is deemed unlikely to make an in-app purchase, they can be issued free rounds or discounts to keep them playing for longer, after which they may be more likely to purchase.

In hospitality or travel, it might be the likelihood to respond to an upsell, make a cancellation, or fail to show up. Airlines routinely overbook flights based on how many people they expect to fail to turn up on the day. AI predictions can help with this by showing whether a larger or smaller portion of passengers are likely to show up, potentially allowing them to sell more seats or save money on compensation by selling fewer.

How Does This Work?

The AI weighs browsing signals and attributes in real-time and compares them to an existing record of users, including data such as when they made purchases, subscribed to email lists, or abandoned carts. The most advanced programs make this kind of calculation in milliseconds, weighing thousands of variables in ways that might take a human analyst days or weeks to complete manually.

AI models are trained on millions of data points to create relevant associations between a visitor’s past actions and potential future actions. It then applies this model in real-time to the data it has about the new visitor, enabling it to make a prediction.

This can be compared to a person who relies on their past experience to infer the correct course of action regarding a new situation – only without the problem of the fallibility of memory. In fact, the AI remembers more and more each day, allowing it to make increasingly accurate predictions.

Why do these AI predictions matter?

Predictions facilitated by AI and based on first-party data shine a strong light on the vast majority of your users, whose intentions otherwise remain invisible. Of course, predictions are not valuable for their own sake, but for what they allow you to do; they enable a digital-native business to take action such as:

• Recommend content in a way that is usually only attainable with user profiles. This kind of personalization is highly valuable: research shows that 89% of marketers see increased ROI when using personalization-based campaigns, while 60% of consumers are willing to become repeat customers after a personalized UX shopping session.
• Show promotions that are tailored exactly to the customer, and appear at the moment when the user is most likely to purchase, or to abandon their cart, improving the odds of conversion and keeping users onsite for as long as it takes.
• Trigger personalized SMS or email that are suggested to users at the optimal moment to secure a sign-up. Email marketing is a tried and tested method that still offers the highest ROI of all marketing channels. Research shows that emails with personalized subject lines are opened 26% more frequently.

You Can Call Me AI

While user-based inference only reveals the potential actions of a dwindling proportion of visitors, session-based strategies address every user and are becoming more powerful every day. AI searches the thousands of data points left by every user to make predictions about both what potential customers will do and what they will respond to most favorably. This transition, from user-based to session-based inference, is speeding up as third-party cookies continue to lose traction. The readiness with which businesses embrace this change could mean the difference between success and failure in 2024.