← Blog

How to retain customers with personalization

There’s a lot of competition for your customers’ attention. Their phones are full of apps built by teams whose whole job is to keep them engaged, and those apps set the bar for what your customers expect from you.

I first wrote about this in 2019 and 2020 across several posts. This brings them together, updated for how we work today.

What the best apps get right

The one that gets me is LinkedIn. Just after I accept a connection request, it shows a grid of people I’m not yet connected with. I can’t help but scroll through a few screens. Sometimes I hold my hand over the screen when I know I don’t have time.

That pull comes from two things working together.

A good experience. The grid itself, with its spacing, images and the amount of information on each card, looks the same for everyone.

One-to-one personalization. The faces in the grid are different for every person, chosen from everything LinkedIn knows about you and about people like you.

Each one is much weaker without the other. A beautiful page showing the wrong things doesn’t hold anyone, and good recommendations on a clumsy page get ignored.

Personalization is a conversation

I think of personalization as a conversation between a customer and your product. When you open a new Netflix account, it asks you to rate a few shows. That’s your half of the conversation. Netflix answers by showing you shows it thinks you’ll like, based on people with similar tastes, and it keeps updating as you watch.

Now imagine Netflix asked you to rate shows and then ignored your answers. You’d close it and probably never come back. Plenty of companies do exactly that:

  • A marketing email from a company that knows your purchase history and still gets the basics wrong. Often the product team personalizes the site and the marketing team never gets that data.
  • A “most popular” row that shows the same items to everyone, when different groups of customers have very different ideas of what’s popular.
  • A news site you visit every day that greets you as a stranger every time.
  • An onboarding form that asks your preferences and then ignores them. A meal delivery service should never suggest steak to someone who told it they’re vegetarian.

When a customer gives you information, act on it.

Where to start with the data you have

You don’t need to be LinkedIn. Say you run an online store with 50 products and a few hundred thousand customers. Here are three things that tend to pay off:

  1. Reach past customers with offers based on what they bought. One email per customer, built from their own history, beats one email with your best sellers.
  2. Suggest add-ons in the cart based on what other customers bought together.
  3. Predict which customers are about to leave, so your retention team can reach them first.

If you’re not ready for any of that, start smaller. Show what’s popular within each group of customers, by region or age range, and count only recent sales so last season’s favorites don’t crowd out this season’s. I go into this in how popularity can kill personalization.

Simple approaches like these do more than you’d expect. They also force you to build the pipeline that moves data from your systems to the model and back to the customer, and once that exists, swapping in a smarter model is a small job.

Scale doesn’t have to mean complexity

YouTube has an enormous library and new videos arrive every minute, and no single recommendation model could score all of them for every viewer in real time. YouTube’s engineers published their approach in 2016, and it works in two steps. First, one model narrows the full library to a short list of candidates for each viewer. Then a second, more accurate model ranks only that short list.

We approach large data the same way. We look for ways to make the problem smaller: training on a recent window of data, splitting customers into groups that behave differently, and dropping fields that don’t add anything. The data is often big when we start, and it rarely needs to stay that way. I explain the main types of these systems in the guide to recommendation systems.

Measure retention where the business reads it

Engagement only matters if it shows up in the metrics leadership already watches: retention, lifetime value and the cost of acquiring each customer.

That’s what we built for a Fortune 250 consumer goods company. Its leadership now sees lifetime value, acquisition cost and retention for seven brands on one set of definitions, so it can tell which brands lead or lag and where acquisition cost is rising. You can read how in the analytics platform we run for seven consumer brands.

Whatever you build, measure it against where retention stood before you started. Our Discovery step does exactly that before anyone builds anything, as you can see in how we work with clients.

Talk to us

If you want to know what your customer data could do for retention, the first call with one of our senior engineers is free and takes 30 minutes.

Let's talk about what you want to improve

Talk to an engineer