A B2B SaaS retargeting audience should be a time-bounded eligibility decision, not a permanent list of everyone who visited the website. A person belongs in an audience only while the entry event, commercial fit, consent state, CRM state, message, and time window still justify the next ad. Suppression should override inclusion. Eligibility should expire unless new evidence refreshes it.
The advertising platform can remember a browser. It cannot decide whether that person is a buyer, an active customer, an open opportunity, a job candidate, or someone who solved the original problem three months ago. That is a business-policy problem.
I co-founded Stackmatix, which manages paid growth programs. That commercial relationship informs the editorial position here. The framework and worked numbers below are hypothetical; they are not client results or performance benchmarks.
The platform setting is not the audience strategy
Google Ads describes data segments as rules that add site visitors or app users, paired with a membership duration that determines how long people remain. It also notes that a new visit can update the timestamp and that one person can belong to multiple segments. Google Ads: how your data segments work
Those mechanics answer whether a platform can include someone. They do not answer whether the business should.
Google also supports audience exclusions, and a segment can start empty or be pre-filled with people who matched during the previous 30 days. Those choices affect who can receive an ad from the first day of a campaign. Google Ads: audience exclusions and Google Ads: create a website visitor segment
Treat platform defaults as implementation details. The policy needs to exist before those settings are chosen.
Privacy and consent requirements also vary by jurisdiction, technology, purpose, and data relationship. As one concrete example, the UK Information Commissioner's Office says online advertising that uses storage or access technologies generally requires consent in the circumstances covered by its guidance, and it emphasizes meaningful user control and consent withdrawal across the processing chain. ICO: online advertising guidance This article is an operating framework, not legal advice. Have qualified counsel review the actual jurisdictions, products, vendors, and data flows. Platform and regulatory sources were checked on October 22, 2026.
Write seven rules before building the segment
A useful retargeting policy should make seven decisions explicit.
1. What evidence creates eligibility?
Name the event. “Visited the site” is usually too broad. A pricing-page view, product-documentation sequence, webinar registration, trial action, or demo abandonment represents a different job and a different level of evidence.
Record the event name, timestamp, source URL or object, identity state, and qualifying conditions. A campaign built around high-intent evaluation should not quietly absorb every blog reader because both groups received the same tracking tag.
The high-intent keyword diagnostic applies the same discipline before the click: a phrase that sounds commercial does not prove the buyer, use case, timing, or economics fit.
2. Who is commercially eligible?
Define fit at the person and account level. The relevant fields might include geography, company type, company size, role, supported use case, product compatibility, or account tier. Use only data that the company is allowed to collect and use for this purpose.
Unknown is not the same as eligible. If the message depends on enterprise readiness, do not assume every unidentified visitor is an enterprise buyer. Route unknowns to a broader educational experience or leave them out until better evidence arrives.
3. What consent and policy state is required?
Document the required consent signal, privacy-policy version, regional treatment, vendor permissions, and withdrawal behavior. If consent is withdrawn, the audience and downstream vendors need an operational path to honor it. The ICO's practical guidance specifically addresses recording consent, refreshing it when purposes change, maintaining preferences, and enabling withdrawal. ICO: managing consent in practice
Do not let an ad-platform match become evidence of permission. Matchability and permission are different states.
4. Which commercial state controls the message?
CRM and product state should override a stale page view. At minimum, distinguish:
- anonymous or known prospect with no active opportunity;
- marketing-qualified or sales-accepted account;
- open opportunity;
- closed-lost account, with a documented reason and cooling period;
- trial or active product user;
- current customer;
- former customer;
- disqualified, competitor, employee, applicant, vendor, partner, or internal tester.
An open opportunity may still receive advertising, but it should not automatically receive the same acquisition offer as a new visitor. A current customer should not be shown “start your first trial” simply because someone at the account read a comparison page.
5. What job should the next ad perform?
Every inclusion needs a defined next decision. Examples include explaining a technical requirement, helping a buyer compare evaluation paths, returning a trial user to an incomplete setup step, or presenting relevant proof to an account already evaluating the product.
“Stay top of mind” is not specific enough to review. Name the buyer question, approved message, destination, and action. Confirm that the destination is actually ready using the paid-traffic landing-page review.
6. When does eligibility expire?
Choose a duration based on how quickly the evidence loses meaning, not the longest window the platform offers. A pricing-page visit may decay differently from a completed event registration or an abandoned product setup.
Expiration should be tied to the original event and refreshed only by an approved new event. A generic return to the homepage should not necessarily restart a high-intent clock. Store eligible_from, eligible_until, the event that opened the window, and the event that refreshed it.
The example durations later in this article are editorial inputs for a test, not universal recommendations. Use sales-cycle evidence, sample size, consent requirements, and buying behavior to set the real windows.
7. Which suppression wins when rules conflict?
Use a deterministic order:
- consent withdrawal or legal/policy restriction;
- disqualified or prohibited identity state;
- current commercial or product state;
- campaign-specific suppression;
- approved inclusion rule;
- default exclusion when evidence is incomplete.
This order prevents a recent page view from overriding a more important fact. It also makes debugging possible: each excluded person should have a reason code, rule version, and effective timestamp rather than disappearing into a black box.
Use a state table instead of one giant audience
The table below is an example policy, not a benchmark. Replace its windows, messages, and owners with decisions supported by the actual business.
| Current evidence and state | Example treatment | Exit or refresh rule |
|---|---|---|
| Pricing or comparison visit; eligible account; no open deal | Product-evaluation message for 21 days | Exit at day 21; refresh only on another approved evaluation event |
| Two qualifying educational visits; fit still unknown | Educational message for 14 days | Exit when window closes or fit is resolved |
| Trial started but setup incomplete | Suppress generic acquisition; route to approved lifecycle or product motion | Exit on activation, abandonment rule, or trial end |
| Open opportunity | Suppress generic offer; use only sales-coordinated account messaging | Exit when opportunity state changes |
| Active customer | Suppress acquisition; include only in a defined expansion motion with separate eligibility | Exit on product or account rule |
| Closed-lost for timing | Suppress during an approved cooling period | Re-enter only on a new qualifying event after the cooling period |
| Disqualified, employee, applicant, vendor, competitor, or unsupported region | Suppress | Re-enter only after the controlling state is explicitly corrected |
| Unclassified record with conflicting evidence | Hold out by default | Re-evaluate when identity or state is resolved |
This is not just campaign hygiene. It is message integrity. The company should be able to explain why a person received an ad that matches their current relationship with the business.
Work through a hypothetical audience
Suppose a fictional B2B SaaS company begins with 2,000 matched people who meet the technical and consent conditions for advertising. Its old campaign includes all 2,000 because each person visited at least one page during a broad membership window.
The team applies the policy and classifies the current state:
| State | People | Decision |
|---|---|---|
| Recent evaluation event; eligible; no open deal | 550 | Include in evaluation audience |
| Qualifying education event; eligible; no open deal | 500 | Include in education audience |
| Open opportunity | 300 | Suppress from generic acquisition |
| Active customer | 250 | Suppress from acquisition |
| Disqualified or unsupported fit | 200 | Suppress |
| Employee, applicant, vendor, or internal user | 120 | Suppress |
| Conflicting or incomplete state | 80 | Hold out pending resolution |
| Total | 2,000 | 1,050 included; 950 suppressed or held |
The policy reduces the generic acquisition pool by 47.5%. That is arithmetic, not proof that performance will improve. A smaller audience may cost more, reach fewer people, or lose conversions. The gain is that the remaining audience has an explainable commercial reason to receive a specific message.
The company can now test a decision. For example, randomly hold out an eligible subset and compare mature account outcomes over an appropriate observation period. Do not declare success from a lower platform cost per click or a high view-through conversion count. The attribution-versus-incrementality guide explains why assigned credit and causal lift are different questions.
Measure the policy, not just the ads
Review the eligibility system and the campaign separately. A useful weekly record includes:
- people or accounts entering each audience and the qualifying event;
- current eligible, reachable, served, suppressed, expired, and unresolved counts;
- median and upper-percentile frequency by segment;
- leakage: customers, open opportunities, disqualified records, or withdrawn-consent records served incorrectly;
- returns to the intended product or commercial step;
- accepted opportunities and value under a fixed marketing-sourced pipeline definition;
- cost per accepted opportunity, with cohort maturity stated;
- holdout outcome and confidence limits when an experiment is feasible;
- rule version, data-through time, source freshness, and failed syncs.
Do not divide ad spend by every opportunity that happened to contain an exposed account and call the result incremental. Preserve the audience, delivery, CRM, and outcome records needed to distinguish association from effect.
Diagnose the system when performance looks wrong
| Symptom | Likely policy failure | First check |
|---|---|---|
| Audience grows but rarely shrinks | Windows refresh on weak events or never expire | Inspect refresh event and eligible_until |
| Frequency rises while qualified progress is flat | Pool is too small, window too long, or message has finished its job | Review reach, frequency distribution, and last qualifying event |
| Customers see first-purchase offers | CRM or product suppression is late or incomplete | Sample served identities against current state |
| Sellers report conflicting messages | Open-opportunity treatment is undefined | Compare campaign eligibility with opportunity stage and owner |
| Cheap clicks produce weak opportunities | Inclusion proves activity, not fit | Audit account, role, use case, and source query |
| Reported conversions stay high after tighter rules | Credit window or view-through reporting masks the change | Compare attribution with a holdout or other causal design |
| Exclusion volume suddenly drops to zero | Sync failed or reason codes were overwritten | Check source freshness, row counts, and last successful job |
The first debugging move should be a record trace, not a creative rewrite. Pick several served and suppressed people, reconstruct the rule inputs as of decision time, and verify that the same policy produces the same result.
Copy this policy brief
Before launch, complete this template:
| Field | Decision to document |
|---|---|
| Policy owner | Person responsible for the eligibility definition |
| Audience purpose | Buyer question and next decision the ads support |
| Entry event | Exact event, timestamp, source, and minimum evidence |
| Identity and fit | Allowed person/account match and fit requirements |
| Consent requirement | Required signal, jurisdictional treatment, and withdrawal path |
| Commercial-state source | Authoritative CRM or product fields and freshness standard |
| Inclusion window | Start, end, timezone, and approved refresh events |
| Suppressions | Consent, state, identity, account, campaign, and internal exclusions |
| Conflict order | Deterministic precedence when rules disagree |
| Message and destination | Approved promise, proof, call to action, and page |
| Frequency controls | Limit, measurement grain, and escalation threshold |
| Outcome | Commercial or product progression and maturity period |
| Experiment | Holdout, randomization unit, contamination checks, and analysis plan |
| QA | Record samples, audience counts, sync checks, leakage threshold, and owner |
| Versioning | Policy version, approval date, change log, and rollback condition |
Run the policy before the campaign. If the team cannot state why someone should enter, what should happen next, and what fact forces exit, the audience is not ready to scale. A pixel can collect a visit. It cannot supply the commercial judgment the campaign still needs.
Co-founder and CEO of Stackmatix, startup advisor, and former Head of Sales at MightyHive. Connect with Matt on LinkedIn ↗ · More about Matt →