Reading:
How UserReplay Uses AI To Predict A Visitor’s Likelihood To Convert With 95% Accuracy

How UserReplay Uses AI To Predict A Visitor’s Likelihood To Convert With 95% Accuracy

Ruth Peters
15th October 2020

Bounce rates, CTRs, cart abandonment rates and totals. These are all incredibly valuable insights, with the capacity to encourage more conversions and reduce lost sales.

But these metrics often lack the detail required to live up to their true potential: an explanation as to why and what that means for your website right now.

While it’s important to know that 15% of site visitors abandon their cart right before checkout — was that because they lost interest, or did they encounter a technical issue instead? Is this a one-off, or is it indicative of many other customer journeys, as well?

Because when you see these customer behaviors happening, you need to understand the wider implications — to explore new ways of setting it right, and securing the sales. 

Better still, you could have the technology in place to predict and pinpoint customers who might end their on-site journey before purchase. That way, you can proactively step in — to troubleshoot on the visitors behalf — and reap the commercial results.

And that’s what we’ve been working on, at UserReplay.

Using artificial intelligence to predict conversions — a UserReplay breakthrough

To create a system capable of predicting conversions (and therefore lost sales, too) we needed to look at which pages a user visited, the order in which they visited them, and then find a way to determine how likely that user was to finalize a purchase. 

To see this goal through, we built an AI inspired by the human brain; one that could view the sequence of page URLs visited by a shopper, and use what it saw to predict the probability of conversion at each stage of the customer journey. 

The AI was then trained — via machine learning analyzing over 100,000 journeys — to predict with 80% to 95% accuracy whether or not the observed sequence would result in a conversion.

While it may sound far-fetched, our team managed to put this piece of software together over the course of six months. 

We spent the first four months creating and refining the machine learning model, teaching our AI how to read the sequence and draw meaningful conclusions from it. Our team would create a model and then dedicate two days testing it. Afterward, we would tweak the model based on the results and run the cycle again. 

After four months of rinse and repeat, we started to see the accuracy we were hoping for. From there, UserReplay’s programmers wrote the backend of the program. Our package was written in Python through the TensorFlow framework — one of the leading platforms for machine learning. 

Two months later, we had a model that could consistently predict conversions of previously unseen user journeys.

Here’s how it works…

So far, we’ve touched on all of the tech that went into this project, but we haven’t had a close up look at what our AI looks like in action. To help you understand what a product like this could do for your ecommerce business, we’ve outlined two common customer journeys that the UserReplay AI analyzed. 

The “smooth” customer journey

The first is a “smooth” journey, so-called because it involves a straightforward path to conversion. In this scenario:

  • A customer lands on the homepage of an eCommerce shop 
  • They then click a link to view a page of categorized products (e.g., shoes)
  • Filter for a specific subset of products (e.g., running shoes)
  • Visit a product page
  • Add it to a cart
  • And then click a checkout button. 

At different stages along this journey, our AI could predict whether or not this customer is going to convert. As we follow this visitor’s journey through the website, we pick up on cues that indicate their future behavior and intent. A pretty clear picture is being painted.

But site visits aren’t always that straightforward, are they?

The “problematic” customer journey

In a slightly less smooth customer journey — one you might deem “problematic” — the customer is either faced with uncertainty or a frustrating interface. Here’s what that looks like to the lens of our AI. 

  • The customer lands on a homepage
  • Clicks a category of products
  • Filters to a specific product
  • Visits a product page
  • Adds the product to his cart
  • And then goes to the checkout page. 

So far, everything is the same as the first customer. 

However, it’s here when things start to break down. At this point:

  • The customer goes back to check their cart
  • Returns to the product page
  • Revisits their cart
  • Starts to checkout
  • Then returns to the homepage.

Oftentimes, customers along the problematic journey do not convert. Maybe they’re unsure about their purchase, can’t find what they’re looking for, or are simply confused by your website’s layout. They may start by exhibiting all the right behaviors, but for whatever reason they drop off.

The result is that Customer #2 leaves without making a sale, just as our AI predicted. 

How can AI help increase conversions?

We made this point at the start, but it bears repeating again: once you realize a customer isn’t going to convert, you can start to take a closer look at why. 

And once you know why a problem exists, you can start to fix it. 

Here are two scenarios we’ve identified where UserReplay’s AI can help you increase your conversion rate. 

Flag non-converting sessions

The simplest method is by flagging non-converting sessions, particularly ones that should have converted, like the one described in the problematic journey above. 

In these situations, where a customer was on the path to conversion, you can start to pinpoint where the stumbling blocks in your customer journey are — indicated by a drop in predicted probability to convert. And by knowing where those stumbling blocks are in your on-site experience, you can begin to remove them.

Identify struggle patterns in your customer journey

Ecommerce websites should be on-going works in progress. You may uncover an issue at checkout, and resolve it. Or you may find that customers are hopping back and forth from one page to another (product page and delivery T&Cs, for example), and work to provide all the information required in one place.

By highlighting these previously unseen patterns of struggle — of shared experiences, many site visitors all go through — you can see the commercial benefit of resolving them. Further to that, you could also start clustering journeys based on a set sequence or pathway through the site — why do some people all exhibit the same struggles, what does that mean, and how can you, as a business, respond? 

That’s something we’re really excited about exploring next. Until then, check out the rest of UserReplay’s features and services by clicking here. And watch this space for more AI-powered insights making their way onto our platform very soon.


0 Comments

Leave a Reply

Are you an ecommerce professional?

Join 5,000+ ecommerce professionals who subscribe to our weekly checkout audits.

Highlighting the smartest ideas from leading ecommerce companies.

Unsubscribe any time with one click.