How to use Data Cloud segments for targeted Google Ads campaigns

A segment of people goes to an advertising destination on a Friday. Two systems disagreed about consent for a slice of that list. The CRM said yes. The preference centre said no. The segment owner took the CRM value because the count looked healthier. The match that came back on Monday was disappointing, and the first proposal in the room was to drop the refusal list so the next send would be larger. The withdrawal that should have stopped Friday's send was still sitting in a nightly file.

#The name on the page

People still search for Data Cloud segments. The product is Data 360. The rebrand date is 14 October 2025. The behaviour is unchanged, including the job the help page already stated: ingest, harmonise, unify, and analyse streaming and batch data. Salesforce's own line is worth keeping beside the old name: "While the name is new, the functionality and content remains unchanged." A segment sits on top of that unified data. It does not launder the consent, the identity, or the fields you chose to include.

Treat the segment as a list of people you are prepared to explain. If you cannot say why each person is on it, and which consent flag allowed the use, the segment is not ready for an advertising destination. Admiring the count is not a legal review. A large count with a blank consent attribute is a larger argument, not a better campaign.

#Consent before the destination

Consent is an attribute you carry, the same way you carry an email or an account id. It has to be on the profile before anyone publishes a segment to ads. A flag that lives in a consent tool and never lands in Data 360 cannot govern the activation. A flag that lands after the send cannot govern the people already offered.

Give the attribute three parts, not one. A value for the purpose you will actually enforce, the system that asserted it, and when. A bare yes with no source cannot be defended when the other system disagrees, and it cannot be repaired. Advertising use requires a specific consent value, not a blank, and not a conflict. Write that rule down before the destination is switched on. A rule that exists only in a workshop transcript will be rewritten by whoever needs the count to look right.

If two systems disagree, do not average them and do not let the segment win. Hold the person out until a person, not a batch job, resolves the conflict. The stricter flag stands in the meantime. Stricter means the person is not sent. Newest-stamp-wins is a default people reach for because it feels mechanical. It is wrong when the newer stamp is a CRM import that never asked the person, and the older stamp is the preference centre they used last week.

Write the refusal list next to the rule. Typical refusals: no email where consent is missing, no phone where the flag is advertising-off, no precise location, no seed audience built from people who opted out. The list will be shorter or longer in your organisation. It has to exist as a list someone can read, not as a vibe. The longer treatment of how a conflict becomes a queue, rather than a footnote, is in consent management across Data 360.

Identity resolution will not settle the disagreement for you. It builds unified profiles from source profiles. Those profiles are keys. They are not a golden record. The engine does not pick a winning value and does not override source values, including two consent assertions that contradict each other. If you unify first and hope the profile "has consent," you have hidden the conflict inside a key ring. The activation rule has to see both assertions, or it has to see a purpose-built attribute that already applied stricter-wins and says so in one sentence. Hidden logic is how a yes leaks through.

#Activation is a destination, not a promise

Choosing an ad channel as a place a segment can go is a destination decision. It is not a statement about match rate, cost per acquisition, or incremental revenue. Those numbers, if they come, come back from the ad platform after the send. Put them in a measurement note. Do not put them in the design as a commitment Data 360 has made. It has not. This page will not invent a match rate to fill the gap. A borrowed percentage would be a fiction about your identifiers.

Keep the activation narrow. Send the identifier the destination needs and the consent that justified the send. Do not attach the rest of the profile because the mapping screen has empty slots. Extra attributes are extra disclosure. They rarely improve a match, and they expand the argument you must have if someone asks what left the org. A phone number added "in case matching is thin" is a new purpose, not a tuning knob. If consent does not cover that identifier, it stays off the send.

This guide does not walk a connector wizard. Destination setup changes, and a stale click-path is worse than none. Confirm the current activation target in your org, with an admin who can see it, before you promise a field list to marketing. Do not name a connector you have not opened. The durable work is the same either way: the consent rule, the refusal list, and the measurement after the send.

Zero copy does not change that order. Zero copy is federation. The help page says you can connect data using zero copy data federation, which leaves bytes in an external store. It is not a latency promise for an activation window, and it is not a consent decision. If the identifier you hoped to send still lives only in a lake, federating the lake does not make the send lawful, and it does not tell you how long the destination will take to accept the row. Measure that after the connection exists. Do not brief it before.

#What a segment is allowed to mean

A segment definition is a filter you can read aloud to the person who will be asked why a customer was included. "Open opportunity, advertising consent yes, country in this list" can be read aloud. "People who look like buyers" cannot. If the filter depends on a score nobody can explain, take the score out. The destination will not explain it for you.

Calculated flags you derived yourself are allowed when the sentence is honest. "Consent yes and not suppressed" is a formula. Call it a formula. Do not brief it as a propensity, and do not let the segment name say "likely" when the logic is a filter. The person who inherits the campaign in six months will trust the name. The name is part of the control.

People get added between the moment the segment is counted and the moment the destination accepts the file. That gap is where a blank consent sneaks in, and where a withdrawal that landed after the count is ignored. Check the flag at activation as well as inside the segment, if the destination can receive a broader set than the segment intended. One filter is a hope that nothing changed. Two checks are a design.

The same rule applies when the audience is described as a lookalike, a predictive slice, or an account plan. The technique does not create a purpose. If the person has not consented to advertising use, they stay out. A buying group does not consent on behalf of a colleague unless you have a lawful basis you have written down for that specific case. Do not imply one by putting every employee of a target account into the activation. The seller can still work the account. The ad destination receives only the people whose flag allows it.

#What you measure, and what you stop

After the first send, record three facts: how many people you offered, how many the destination accepted, and how many you refused under the consent rule. The acceptance figure is your match outcome. It is allowed to be disappointing. A low match is a reason to check identifiers and consent coverage. It is not a reason to drop the refusal list, and it is not a reason to backfill a yes on the blanks so the next count looks complete. Backfilling a yes is an incident. Treat the proposal as one the moment it appears.

Stop the activation when consent changes. A withdrawal has to reach the segment rules on a path you have tested, including the streamed path if that is how the withdrawal arrives. A batch-only consent load will not outrun a send that already left. If ads are live, the withdrawal path has to be at least as fast as the spend. Test it with a named person. Change the flag in the source. Watch the segment, then the destination. Write down the minutes. If you cannot run that test, you do not have a stop. You have a policy.

Do not move this measurement into a new chart and call it done. Tableau Next does not replace Tableau. The help page lets you expand Data 360 using either. A workbook can show the three counts. It cannot invent the fourth, which is the match rate you wished you had been promised. Put "offered, accepted, refused" where the operator already looks. If that place is Tableau, stay there. Changing tools does not make a thin match into a strategy.

#The argument you will have to repeat

Three proposals come back every time the accepted count is smaller than the room expected.

Loosen consent to grow the list. The refusal list is doing what you wrote it to do. Growing the list by sending blanks is not an optimisation of match. It is a decision to treat missing as yes. If someone wants that decision, they can write it down with their name on it. The segment definition should not smuggle it.

Add more attributes so the platform can match. More attributes are more disclosure. Add one only when you know the destination uses it and consent covers it. "The mapping screen had a slot" is not a use. Confirm the slot with the admin who can see the current target. Then decide.

Unify harder. A looser match rule will make the profile look richer and can attach one person's consent to another person's identifier. Under-matching leaves a duplicate you can see. Over-matching is quiet. The merged profile has more fields filled in, and the segment count goes up. That is not a better audience. It is a wrong person with a lawful-looking flag. Start strict. The identity decisions that are cheap before the first activation and expensive after it are set out in getting Data 360 right the first time.

#A first send you can defend

One purpose. One destination, confirmed in the org. One segment whose filter you can read aloud. Consent populated, conflicts held out, blanks held out. A refusal count you record beside the offered count. A withdrawal test with a time written down. Then stop.

Do not add a second destination in the same week. The second destination has different identifiers, a different disclosure, and a different temptation to fill empty mapping slots. The first send is finished when you can explain, without a slide, who was offered, who was refused, and who can turn it off. Anything you cannot explain that way is not a campaign yet. It is a list with a logo on it.