Behavioral Segmentation: Targeting Based on Actions
The actions worth segmenting on, the ones that are noise, and how far back the window should reach.
On this page
The short version
- Actions predict better than attributes. What someone did last week tells you more than anything they declared at signup.
- Most tracked events are noise. The ones worth segmenting on are those that change what you would send next.
- The lookback window is the setting nobody sets deliberately, and it decides whether a segment describes current behaviour or ancient history.
Behavioural segmentation groups people by what they have done rather than what they are. It outperforms attribute-based splits consistently, for an obvious reason: an action is evidence, and a declared attribute is a claim made once at signup and never revisited.
The difficulty is that modern tooling records far more events than are useful, and the temptation is to segment on all of them.
Which actions carry signal
The test is the same one that applies to every segment: would this change what you send next? An action that would not is worth recording and not worth segmenting on.
Actions with strong signal share a property — they cost the person something. Time, attention, a decision. Reading a long guide, returning three times in a week, starting a trial, requesting something specific. These are expensive to fake and rare enough to mean something.
Actions with weak signal are the ones that happen incidentally. A page view, an email open, a click on a navigation link. They are common enough that segmenting on them produces groups that are almost everyone.
Signal by action
| Action | Signal | Worth a segment? |
|---|---|---|
| Started a trial or a tool | Very strong | Yes |
| Visited pricing more than once | Strong | Yes |
| Replied to an email | Strong | Yes |
| Read a long guide to the end | Moderate | In a cluster, yes |
| Returned three times in a week | Moderate | Yes |
| Clicked a link in an email | Weak alone | Only as part of engagement tiers |
| Opened an email | Very weak now | No |
| Viewed any page | None | No |
Opens sit here rather than higher because privacy pre-fetching means an open is partly a machine event. Segmenting on opens alone builds groups containing people who have never seen your email.
Clusters beat single events
One action is weak evidence; a pattern is strong. Someone who read three deliverability guides in a fortnight has told you something no single page view could.
That argues for segmenting on grouped behaviour rather than individual events — a topic cluster rather than a URL, a count of visits rather than a visit. It is also more robust: a URL changes when you restructure the site, and a cluster survives it.
The practical version is to tag content by subject and segment on the tag. Then a reader who has engaged with three pieces in one subject is identifiable regardless of which pieces they were.
Absence of action is also a signal
The most valuable behavioural segment on most lists is defined by what someone has not done: no opens, no clicks, no visits in a defined window.
That is engagement segmentation, and it is behavioural. It is worth naming as such because it is often treated as a hygiene task rather than as the most reliable behavioural split available — which it is, since inaction over a long window is far harder to misread than a single action.
Behavioural segment check
- The action costs the person something
- A lookback window is set explicitly, not left at all time
- The window matches your actual cycle length
- Segments key on tagged clusters, not individual URLs
- Opens are not the basis of any segment
- You have written the email this segment would receive
- Someone reviews the rules when the site or product changes
Frequently asked questions
How much event tracking do I need?
Less than most setups collect. Half a dozen well-chosen events with clean naming beat forty events nobody can interpret, and the forty are usually why the data is unusable.
Should behavioural segments be automatic?
Yes, where the platform supports it — a rule that evaluates continuously stays current. Manual segments become historical snapshots the moment the person maintaining them moves on.
What if someone matches several behavioural segments?
Write down the precedence rule before that happens. Without one, which email they receive depends on the order the sends happen to run in, which is not a decision anyone made.