Product recommendation engines in online retail: is your organisation taking full advantage of them?
Product recommendation engines are a must-have for retailers. If you want to offer tailor-made experiences at-scale, they’re a fundamental building block for your online store.
Use it well, and it can help you beef up average order value (AOV) and increase customer stickiness. Manage it poorly, and it can put your customer satisfaction and brand reputation at stake, like generating irrelevant or inappropriate recommendations. But we’ve identified a new problem: while all retailers have product recommendation engines, few are using them to their full advantage.
Put simply, a majority of retailers aren’t utilising all of their available features, and are failing to flex their true capabilities. Most notably, they can’t experiment with different filters and strategies to offer their customers the best shopping experience possible, nor validate whether their engine is performing optimally.
While your product recommendation engine could still be keeping customers happy, the issue is they could be happier. They could be more loyal to your brand, or they could be forking out for bigger orders. And your ability to compete is at-risk, too, if you can’t improve the online shopping experience and match your rivals. But what makes this problem so critical is that it’s invisible; if you don’t know how to maximise the full value of your recommendation engine, you won’t know how to assess how much value it’s generating, either. This paradox comes down to the nature of the technology.
Since product recommendation engines are highly complex, they can only really be built, customised and improved by specialist data scientists. So, most retailers turn to off-the-shelf engines. While there’s nothing wrong with buying your AI, vendors typically leave you with simple, surface-level tutorials for this highly-complex technology. Your team, therefore, is stuck with a limited understanding of a critical tool.
The solution sounds obvious: ditch the off-the-shelf product for an in-house build. However, only an army of specialist data scientists could handle such a task. Few organisations sport an in-house product recommendation team, like Spotify. And their dedicated R&D teams are backed by millions of pounds and years of research to deliver their disruptive recommendation engine. Most retailers simply don’t measure up, both in terms of expertise and resources.
What you should be doing, instead, is building out your in-house data literacy and enhancing your team’s understanding of your recommendation engine. This can enable you to finally go deep into the technology, uncover features concealed by 30-minute tutorials and tap into its full capabilities. Let’s explore how this can tackle some of the most common problems associated with product recommendation engines, and turn that paradox into an opportunity:
1. Your team can’t set up experiments with your product recommendation engine
Most recommendation engines have a wide range of filters and strategies you can adjust to see how your customers respond to them. But most users don’t know these different options are there, let alone that they can change them. So, they stick with the default settings provided on setup. Settings which might be failing to capture value. For instance: a fashion retailer might flag what’s “most popular” for their online visitors (based on purchases) on their homepage without knowing that a large proportion of their customers really want to see what’s “most viewed”.
If they can’t test different strategies and compare their individual impacts, they can’t determine what the optimal combinations are for your customers. Users need to be fully aware of all the different filters and strategies available to them, as well as how to proficiently carry out experiments.
2. Your team can’t assess whether your product recommendation engine is performing optimally
Good metrics are critical to validate whether your engine is generating good recommendations. However, the default metrics provided on setup rarely meet the standard. Take one of the most common metrics: click-through-rate (CTR). CTR can’t tell you if users buy more, if they return to make new purchases, nor if visitors are even converting. But since users aren’t made aware of how relevant these metrics are to the engine, as well as the best ones to leverage, they only analyse information that offers little insight.
Knowing the right metrics to assess – like purchase rate, conversion rate and AOV – and how to assess them is essential. Not simply to gauge if the recommendations need to be improved, but why and how.
Ready to put your product recommendation engine to work?
Those that can refine their personalisation will be in a strong position to challenge retailers lumped with default settings and limited capabilities. In most cases, this is all they can do in lieu of a Netflix-worthy team of experts. To do that, you’ll need to explore the different ways of building out your in-house data science literacy and capability. And we can help.
To find out more about how Faculty can help you enhance in-house capability and improve your product recommendation engine, get in touch with our team.