B2B Retargeting

Advertising to people who have already been to your site

Retargeting shows advertisements to people who have already interacted with you — visited the site, watched a video, opened a form, attended an event. Google calls the same practice remarketing; in B2B the words are used interchangeably. The audience is defined by past behaviour rather than by attributes, which is the distinction from every other kind of targeting.

It gets confused with three neighbouring things. It is not account-based advertising: targeting a named list of companies reaches people who have never heard of you, which is prospecting. It is not email remarketing, where the identifier is an address you already hold. And it is not brand advertising, though it is often reframed that way once the conversion numbers stop looking good.

Retargeting ads are cheap per impression, quick to set up, and almost always the best-performing line in the reporting. They are also the line where the gap between reported performance and incremental effect is widest.

The mechanics, stated plainly

Every retargeting system works the same way in outline. A tag on your site records an identifier for the browser. That identifier joins an audience list with a membership duration. When someone on the list appears on inventory the platform can sell, an ad is served. Everything that determines whether the channel works is a property of that pool: how it was built, how large it is, how long people stay, and who has been excluded.

Three delivery paths differ in their exposure to browser privacy controls.

• On-platform engagement audiences, built from actions inside a single platform — a video view, a form open, a page visit there. Nothing cross-site is involved, so browser cookie policy is largely irrelevant.
• Site-tag audiences delivered on the same platform, where a tag tells the platform which of its members visited you.
• Open display retargeting, the classic case, and the one most dependent on cross-site identifiers surviving in the browser.

Every platform enforces a minimum audience size before delivery starts, and on list-based audiences that floor counts matched members, not rows in your file.

Third-party cookies: what is actually true in 2026

Most published B2B advice on this subject is wrong, and wrong in the same direction. Chrome is not deprecating third-party cookies. Google reversed that plan and reaffirmed the reversal: in April 2025 it said it "will not be rolling out a new standalone prompt for third-party cookies" and would "maintain our current approach to offering users third-party cookie choice in Chrome," and it repeated that position in October 2025. Cookie-based retargeting still functions on Chrome traffic.

What Google did retire was most of the replacement machinery. The Privacy Sandbox advertising APIs marked for deprecation and removal include Topics, Protected Audience, Attribution Reporting, Private Aggregation and Shared Storage. What continues is privacy plumbing — CHIPS, FedCM, Storage Access, Private State Tokens — none of it a targeting replacement.

The other two browsers are where the loss already happened, and the difference between them is routinely mangled. Safari blocks cross-site cookies by default, and has since Safari 13.1 and iOS 13.4, announced in March 2020; WebKit's wording was "cookies for cross-site resources are now blocked by default across the board." Firefox partitions them: Total Cookie Protection is on by default in standard mode and, in Mozilla's description, "builds a fence around cookies, limiting them to the site you're on so third parties can't use those same tracking beacons to follow you from one site to the next." The effect on cross-site tracking is similar; the mechanism is not, and writing that Firefox blocks third-party cookies is imprecise.

Why the reporting flatters it, and how a holdout settles the argument

Retargeting reports well because of who is in the audience, not because of what the ads did. The pool consists entirely of people who already showed interest and who, in many cases, were coming back anyway. Serve them an ad and any conversion inside the attribution window is credited to it. Last-click models are especially generous here, because a retargeting ad is by definition late in the sequence.

There is one way to find out what the channel contributes: withhold it. Randomly split the eligible pool, suppress ads from one group, and compare outcomes. The difference is the incremental effect. Everything else — reported conversions, view-through conversions, platform cost per acquisition — measures correlation with an audience selected for converting anyway.

Two caveats. A holdout needs enough volume in each arm to detect a difference, and many B2B programmes do not have it; if you cannot power the test, say so rather than believing an underpowered result. And measure the right outcome over the right period: accounts reaching opportunity stage, over a window at least as long as your cycle, not form fills over thirty days.

Frequency, burnout, and the ninety-day chase

A ninety-day membership window with no frequency cap means one person who read one blog post sees your advertisement several times a day for three months. This is the most common configuration in B2B accounts and the most reliably damaging. Impressions past the point of usefulness cost money, and they cost goodwill: the response to being followed by a vendor you glanced at once is irritation, not consideration. In a market of a few thousand relevant people, irritating them at scale is a strategic error.

• Cap frequency deliberately per person per week, and be willing to set it low.
• Match window length to the behaviour. A pricing page visit justifies a longer window than a blog visit; a careers page visit justifies none.
• Suppress the people who should not see it: customers, open opportunities, everyone who already converted, and job applicants.
• Rotate creative before it goes stale, and sequence rather than repeat. If someone is worth ninety days of attention, they are worth something new in week six.

Retargeting a named account list versus retargeting everyone

An untargeted retargeting pool in B2B is mostly not buyers. Traffic on a typical business site contains job seekers reading the careers page, customers looking for documentation, competitors checking positioning, students, suppliers, and a substantial volume of bots and AI crawlers. Building one audience from all of it advertises to people who will never purchase anything.

Two refinements do most of the work. Segment by page and depth: a visitor who read three technical pages and a case study is a different proposition from one who bounced off a blog post. Then intersect the pool with the target account list, so spend concentrates on visitors from companies sales has agreed to pursue — on LinkedIn, company list targeting combined with website retargeting.

Account-level identification has real limits. Reverse-IP resolution identifies the organisation assigned an IP range, not a person and not necessarily an employer. Remote workers appear on residential ISP ranges, mobile traffic behind carrier NAT is unidentifiable at company level, VPNs relocate the apparent address, and shared offices resolve to the landlord. No vendor publishes an independently audited match rate.

When a B2B company should not run retargeting at all

This is one of two channels routinely bought by companies far too small to benefit. The other is LinkedIn advertising, and the reason is identical: the platform minimum is low enough to feel like a safe experiment, while the volume the mechanism needs is far higher.

• Traffic is too low to build a pool that clears the platform minimum. Stretching the window to inflate the list makes the audience staler, not better.
• The traffic you have is mostly not prospects. A site whose visitors are predominantly customers looking for support content has no audience worth buying, only a support audience to annoy.
• You cannot measure at account or pipeline level. With no CRM link, retargeting reports the best cost per conversion in the account forever and you never learn whether it earned anything.
• The site does not answer the questions a returning visitor has. Paying to bring someone back to a page that failed to persuade them is paying twice for the same failure.

Fix the pages, get outcomes flowing from the CRM, build enough traffic to have a real audience, then buy retargeting. In that order it is a sensible, modest line item. First, it produces a report that means nothing.

Frequently Asked Questions