Behavioral Targeting

Behaviour-based Targeting · Behavioural Targeting

Definition

Behavioral targeting is the selection of an audience and of message content from recorded user actions — views, searches, add-to-cart events, exits, campaign reactions — rather than from demographic characteristics.

What behavioral targeting is

Behavioral targeting selects the audience and the content of a communication from what a person
did, not from who they are. Instead of women aged 25–34 in large cities, the rule reads:
entered the Coffee machines category twice in the last five days, applied a price filter, did not
add to cart.

The difference is fundamental. A demographic trait is static and barely connected to the shopper’s
current task. Behaviour reflects intent here and now, and therefore offers an entry point for a
specific action: help the choice, remove an objection, return them to an abandoned cart.

Technically, behavioral targeting is a layer over an event stream. First
event tracking records the actions, then they compose into an audience
with a recomputation rule, and that audience becomes the display condition for a scenario.

Which signals are used

Not all actions are equally informative. Below is a working set of signals and what each turns into.

Signal	Scenario	Mechanic
Two or more product views in one category	The shopper is choosing	A Similar items block plus a selection within the viewed price band
A search with no click into a card	They did not find it	A query refinement hint, a banner for a relevant subcategory
Add to cart with no order	Abandoned cart	An on-site reminder on the next visit, a You did not finish block
Exit from a product card within 10 seconds	An irrelevant entry	Do not target — the signal is noise; use it as an exclusion
Six or more pages with no cart	A drawn-out decision	Decision help: comparison, a quiz, a top-three selection
Three or more visits in a week	High engagement, doubt about price	Show delivery terms, availability, warranty — not a new offer
A view of the delivery or payment page	Checking terms before buying	Remove the objection on the product card itself
A click on a past campaign	Confirmed interest in the topic	Continue the sequence rather than repeat the same banner
No click across three impressions	The scenario does not work for this person	Exclude them and free the slot for another offer

A separate class of signals is product properties rather than products. A rule on viewed this SKU
covers a narrow audience; a rule on viewed items from brand X in a given price band covers an order
of magnitude more people at the same precision. Which is why the quality of behavioral targeting is
capped by the quality of the product feed.

Recency windows: every signal has a shelf life

A signal with no expiry is the main cause of inappropriate offers. Interest decays at different
speeds depending on the decision cycle in the category.

Category type	Indicative window	Logic
FMCG, groceries, consumables	1–3 days	The decision is fast and the interest does not persist
Apparel, footwear, beauty	5–14 days	There is a consideration cycle, but seasonality devalues the signal
Electronics, appliances	14–30 days	A long selection, comparison, waiting for a promotion
Furniture, DIY, large purchases	30–60 days	A long cycle, often with an offline stage
Gift and event-driven queries	Until the event date	After the date the signal is useless and harmful

⚠

The second critical timing is the switch-off. Once an item is purchased, interest signals for it
must immediately exclude the shopper from the matching scenarios. Chasing someone with an offer for
something they already bought is the most visible and the cheapest-to-fix error in behavioral
targeting.

Onsite and advertising targeting rest on different foundations

One term covers two technologically distinct practices.

Onsite targeting runs on first-party data: events are collected
on the retailer’s own site and app, the profile stays inside their perimeter, and the rule applies
to their own traffic. Browser restrictions on cross-site tracking barely touch this.

Advertising behavioural targeting historically relied on data about behaviour on other sites,
collected through third-party cookies and network pixels. That layer is degrading: identifier
lifetimes are shrinking, the share of browsers blocking by default is rising, and audiences go stale.

Property	Onsite targeting	Advertising (external data)
Signal source	Own site and app	Networks, partner pixels, DMP segments
Identification	Own ID, sign-in, first-party cookie	Third-party cookies, advertising IDs
Durability	High	Falling with browser restrictions
Interpretive precision	The context of every action is known	The fact is known, the context is not
What can be changed	Content, product order, blocks, popups	Creative and bid only

The practical consequence: scenarios previously covered by retargeting in external networks are
moving into the owned perimeter — the site, the app and owned communication channels.

Common mistakes

Targeting a single action. One touch is not intent. A rule on viewed a card collects almost all
category traffic and devalues the scenario. Demand repetition (two or three touches) or strength
(cart, wishlist, comparison).

No frequency limits. Without frequency capping one scenario
competes with itself and with everything else on the page. Set limits at three levels: impressions
per visit, impressions per period, cooldown after a dismissal.

Chasing a purchased product. This needs both an exclusion on the order and category awareness:
after someone buys a fridge, other fridges are pointless while accessories are not.

Mixing signals of different ages in one rule. A view from three months ago and an add to cart
yesterday should not carry the same weight in the same audience.

No contextual exclusions. Someone in the returns or order status section is solving a service
problem; selling scenarios do damage at that moment.

A rule with no verification. Behavioral targeting is easy to justify logically and hard to
justify empirically. Every material rule is checked with an
A/B test against the untargeted scenario: some clever rules perform worse
than one general offer to everyone.

A launch checklist for a rule

The signal and its threshold — which action, how many times, over what period.
The recency window — when the signal stops being current.
Exclusion conditions — purchase, service context, no reaction across N impressions.
Frequency limits — per visit, per period, cooldown after a dismissal.
Priority against other scenarios — what is shown when someone qualifies for two audiences.
The success metric — not banner click-through, but conversion and revenue per visitor in the
segment.
The verification plan — a control group and a test period fixed before launch.

Frequently asked questions

How does behavioral targeting differ from demographic targeting?

Demographic targeting answers who this person is — gender, age, city, income. Behavioural targeting answers what they are doing right now. For e-commerce the second is usually more precise: a 35-year-old man in a large city is too broad a frame, while viewed running shoes in a given price band three times in the last three days is a ready scenario. In practice they combine: behaviour sets the offer, demographics set the tone and the creative.

What data is needed to start?

At minimum, an event stream from the site or app: product view, category view, search, add to cart, order placed. That covers 80% of scenarios. Next come product attributes from the feed (brand, price, size) so a rule can work on a property rather than one SKU, and reactions to past campaigns so the same thing is not shown twice.

Does behavioral targeting die with the third-party cookie?

Only its advertising variety does — the kind built on third-party data about behaviour on other websites. Onsite targeting runs on the retailer's own data inside the first-party perimeter, so browser restrictions barely touch it. That is precisely why budget keeps shifting from external behavioural targeting into owned surfaces.

How often can a behavioural offer be shown?

A frequency cap is part of the rule, not an option. A reasonable starting point for onsite mechanics: no more than one or two impressions of a scenario per visit and no more than three to five per week, with a cooldown after a dismissal. Without that, a scenario that works for 5% of the audience irritates the other 95% and drags the whole page down.

Why is a single product view not enough to target on?

One view is noise: a stray click, a wrong turn from search, someone else's interest on a shared device. A signal becomes meaningful through repetition (two or three touches of a category or brand) or through strength (add to cart, wishlist, a return visit to the card). Rules built on a single view produce a wide, imprecise audience and ruin the economics of the impression.

Gravity Field