Martech, explained · Content and channels · 2:15
Personalization engines, explained
What decides which banner, product or offer one person sees, in the time the page takes to load
The short answer
A personalization engine decides what one person sees right now, on your website, in your app or in an email: which banner, product or offer. It decides by rules, tests or models, using this session and what you already know. That second kind can be out of date, and only a control group shows what it is worth.
Questions this answers
What does a personalization engine decide?
What one person sees right now: which banner, which product, which offer. On your website, in your app, and sometimes inside an email at the moment it is opened. Two people open the same homepage at the same moment and see different things; something chose each in the time the page took to load.
How does a personalization engine decide?
In one of three ways. Rules: if she came from a running ad, show running. Tests: show two versions and keep the one that wins. Models: predict which product this person is most likely to want next. Most engines do all three.
What does it decide on, and when is that out of date?
Two kinds of facts: what this person is doing in this session (the page, the click, the cart) and what you already know about them, like a segment or a score sent in from your CDP or warehouse. The second kind can be stale: if she bought the jacket this morning and her score refreshes tonight, she sees the jacket all day.
Do you need a separate personalization tool?
Maybe not. Of the 44 personalization tools in the Martech Stack Builder catalog, only 8 do nothing else; 16 also run A/B tests and 6 are part of a CDP or a CEP. Check what you already own before you buy one.
Why do teams run out of content, and how do you prove it works?
An engine can only choose between things that exist: five segments need five versions of the banner, and many teams run out of content before they run out of segments. To know it works, keep a control group, a slice of people who see the default page; the difference between the two is what personalization is worth.
Transcript
Every chapter as text. Click a time to play the film from there.
0:00 Two people, one homepage
Two people open the same homepage at the same moment. One sees running shoes. The other sees a rain jacket.
Something chose each of those in the time it took the page to load. That's a personalization engine.
0:16 What a personalization engine decides
Personalization engines decide what one person sees, right now: which banner, which product, which offer. On your website, in your app, and sometimes inside an email, at the moment it's opened.
0:30 Rules, tests and models
It decides in one of three ways.
Rules: if she came from a running ad, show running.
Tests: show two versions, and keep the one that wins.
And models: predict which product this person is most likely to want next. Most engines do all three.
0:49 What it decides on
It decides on two kinds of facts. What this person is doing in this session: the page, the click, the cart.
And what you already know about them, like a segment or a score, sent in from your CDP or the warehouse.
1:05 When the facts are out of date
The second kind can be out of date. If she bought the jacket this morning, and her score refreshes tonight, she'll see the jacket all day.
1:15 You may already own one
Of the forty-four personalization tools in our catalog, only eight do nothing else. Sixteen also run A/B tests. Six are part of a CDP or a CEP.
So check what you already own before you buy one.
1:31 Content for every version
And an engine can only choose between things that exist. Five segments need five versions of the banner. Many teams run out of content before they run out of segments.
1:43 Proving it works
To know it works, keep a control group: a slice of people who see the default page. The difference between the two is what personalization is worth.
1:53 Three questions to ask
So before you choose one, ask three things.
Does it decide in the session, or from yesterday's data?
Which of your tools already does this?
And who makes the content for every version?
Learn more at Martech Stack Builder, or build your own.