We all know that a generic site experience doesn’t cut it any longer. A lot goes into creating a great user experience that’s tailored to different individual interests and needs — and it’s very easy to get it wrong. Poorly executed attempts at  personalization can lead to lost revenue, lost trust, and lost customers.

In this blog post, we’ll show you how to think about and implement e-commerce personalization in a way that holds people’s attention longer, drives engagement, and builds customer loyalty.

Implicit versus explicit intent in personalization

One very common approach to personalization is tracking behavioral data points and building profiles and segmentations. Most personalization solutions do this by focusing on implicit personalization: responding to a user’s particular behaviour while they are navigating. They typically apply one of these techniques:

1. Tailored page layout and design based on the user profile (changing certain elements of the page)
2. Targeted overlays & notifications (e.g., triggering messages in app or via or seeing a discount as you are about to leave the website without buying)
3. Recommendations (e.g., when you click on a video or a product you are interested in, you get a sidebar of similar or complementary items)

The risk with implicit personalization is that it’s highly prescriptive, making it fairly easy to misjudge the visitor’s goals. Without asking directly and getting an explicit answer, you’re simply inferring what the user’s after and responding with a content that you hope is relevant.

Let’s say I bought a gardening tool set as a birthday gift for my aunt last week. If this week Amazon recommends a collection of gardening tools, it’s no longer relevant to me. Ultimately context matters, which makes personalization more about understanding the individual in real time as opposed to solely relying on segmentation and profiling.

With 84% of customers saying being treated like a person, not a number, is very important to winning their business, it’s more important than ever to handle both implicit and explicit intent, and achieve the right balance of both.

Search: the gateway to explicit intent

Users expect their questions to be answered from the interface they use on the site — and that interface is the search bar. The search bar is where users can freely type in what they’re looking for and express that explicit intent.

When we think about user intent in search, it’s all about understanding relevance.

Let’s look at a personalized online shopping experience. If a user types in: “evening dress for a cocktail party”, we need to calculate relevance that incorporates the context, then combine that with the implicit intent that we derive from the behavioral patterns we’re seeing: for example, clicks, the way they add things to a wishlist, and so on. Mixing in these personalization signals in the ranking strategy, and combining them with business metrics such as sales rank or popularity is the key to successful personalization.

If you are implementing e-commerce personalization, and within it, a search merchandising (searchandising) strategy — say if you’re promoting items per keyword — we need to make sure that that also still holds while you blend in the personalization results. In other words, you need to balance what the results that users want to see and the result that you as a business want to show. And all of that needs to be layered on top of the natural language processing capabilities like typo tolerance and synonyms.

For example, if the user types in “sports” (plural), the search engine should understand the plural and return results for “sport” (singular). Once we have handled the textual relevance portion, it could be about ranking based on popularity metrics. Five star reviewed items would be boosted to show up first. If the user is a movie fan, a highly ranked documentary will show up first. And if you’re a sneaker addict, then we want to show you some of the highly ranked shoes, like shoe A. And then maybe merchandise shoe B in a second position before showing other sneakers that you might be interested in.

Side note: Algolia lets you test these strategies in our personalization simulator before implementing them.

E-commerce personalization: from understanding to implementing

How do you go from understanding it and appreciating the problem as a concept, to preparing yourself for a personalization solution to implement it correctly without doing damage, knowing that getting it wrong is worse than not doing it at all.

Preparation is about understanding your users, understanding your business, and acknowledging the fact that we’re at a stage in the evolution of algorithms and personalization in general where we’re working with an incomplete picture.

Here are some things to consider before you jump on the personalization bandwagon.

Is it right for your (sub)industry?

While personalization has a ton of potential benefits, it’s not for everyone. The technology behind website personalization, as it exists today, doesn’t lend itself to every use case.

For example, personalization is an expected necessity for a video streaming site. And, of course, personalized shopping experiences have become the norm for e-commerce sites. In other industries, the use case isn’t as straightforward.

Is it the right time?

When trying to determine the impact that personalization will have on your site, ask yourself:

1. How much do you know about your users? Will you be able to infer some context about them? Do they have strong, consistent preferences over time?
2. Is personalization going to move the metrics that matter to you? If you want to increase cart size or time on site, then personalization can be very helpful. If your goal is to push users toward a specific conversion point, then it might not be the right strategy for your needs.

Are you able to paint the right picture of your user(s)?

What personalization is doing at its core is taking behavioral signals and inferring intent, then using the picture it’s created to make predictions. But your recommendations are only as good as the picture you paint of that user. The problem is that the paint that’s available can only do the job for a very small subset of who that person actually is.

Purchasing preferences stem from the user’s external environment (situational factors) and internal environment (personal preferences, beliefs, and affinities). Someone’s internal environment will always change based on their income bracket changing, their age changing, and so on. You can still track some of these signals over time.

What’s a lot harder to do is get signals from the external environment. For example, if a customer is looking for wedding shoes, it isn’t useful to continue showing them wedding shoes after they’ve made a purchase.

In order to solve for this, your best bet is to try and identify affinities for the user to your different kinds of products or offerings that aren’t affected as much by the external environment. Take a look inside your business and look for parts of your offering where users can gain a consistent preference over time. In the example of our wedding shopper, while it’s not useful to continue showing them wedding shoes, it is useful to show them shoes from brands they indicated a preference for.

Can you meet the demand for trust and transparency?

Consumers are becoming more and more attuned to how companies collect and use their data online. You want a personalization solution that builds user trust rather than undermine it. You can do this through transparency — show the user what you’re collecting, and why. When users see the data you’ve collected and why you’re recommending certain options, not only do they feel more secure, but they can help you refine your search with their feedback. Giving them the chance to opt out of personalization in their user profiles ensures that those who remain have truly bought into the concept.

You are what you measure

Adding personalization to your site can have an impact on sales, customer satisfaction, and other important metrics. It can disrupt your business if done wrong, so institute personalization in a controlled environment. You want to start simple, slowly iterate, refine, and get better.

To better direct and track this process, pick a North Star metric that’s close to your bottom line and see how the needle changes over time. For a media business, that primary success metric could be view time or bounce rate. With e-commerce personalization, it could be checkouts, active carts, or conversion rates.

You should A/B test your personalization strategies, and create feedback loops to learn from and constantly iterate on. You can’t ever simply trust the algorithm: you need to have some form of human intervention, and the ability to intervene if necessary. It’s critical to create a structured testing program so that if things go South you can always revert.

In summary…

Personalization is hard. Whether you have already chosen your e-commerce personalization tool, or are just thinking about it, it is important that you have the right foundation and way of thinking about it.

Meanwhile, for some examples of e-commerce personalization done right, check out Shopify’s 4 examples of great hyper-personalization, Sailthru’s retail personalization index, and, for a unique use case and a bunch of interesting metrics to track, how Decathlon Singapore increased conversion rates by 50% with personalization.