Consent-gated audience syncs to ad platforms via reverse ETL

The first audience I sync is almost never an acquisition list. It is suppression: active customers, open claims, cancelled subscribers, employees, anyone mid-complaint. On the accounts I have picked up, prospecting campaigns were spending 20 to 30 percent of impressions on people the business already had. Suppression is the sync that pays for the engagement and it is usually live inside two weeks. How the pipes run: Hightouch or Census on top of your warehouse, models in dbt, syncs defined in code and reviewed like code. Destinations: Meta Conversions API and customer lists, Google Ads Customer Match with Enhanced Conversions for Leads, offline conversion import for pipeline that closes on the phone, LinkedIn Matched Audiences, TikTok Events API, DV360 or a trade desk where an agency needs the seed. Identifiers are normalised before hashing, lowercased, trimmed, phone numbers in E.164, because unnormalised SHA-256 is the most common cause of a match rate that looks like a targeting failure and is actually a string failure. The consent gate: Every sync reads lawful basis and purpose from the profile before a row leaves the warehouse. No recorded basis, no send. No special category inferred from behaviour and then shipped to a platform, which for an insurer is an enforcement risk and a reputational one at the same time. For family products, an age flag under the market threshold blocks the row outright, and no child's data seeds a lookalike. These are gates in the model rather than lines in a policy document, so a campaign manager under pressure cannot switch them off in a platform UI. Match rate is a diagnostic, not a target: A 72 percent match on 40,000 known subscribers tells you something is wrong with your email hygiene. A 30 percent match on cold data tells you the data was cold. I report match rate beside list decay and the share of the audience genuinely reachable, and I will not raise the headline number by loosening the identifier rules, which is the usual method. These fail quietly, so they get monitored: Ad platform syncs fail partially and without complaint: a rejected batch, a deprecated field, an expired token, and nobody notices until a quarterly review. Every sync ships with row-level error alerting, a freshness test, and a weekly reconciliation between what the warehouse sent and what the platform reports as ingested. That reconciliation is the deliverable most teams have never had, and it ends the argument about whether the audience or the media was at fault. Not in scope: Campaign strategy, creative, bidding, budget allocation. Your media team gets audiences they can trust and no instruction on how to buy. Also excluded: building the underlying customer model when none exists yet, CDP implementation, and clean room work with a retail or platform partner. Wrong fit if: Your audiences live in spreadsheets emailed to an agency and nobody wants that to change, because tooling is not the obstacle in that situation. If the plan is a lookalike from every behavioural segment somebody can name, on the reasoning that the platform will sort out the rest. Or if the intent is to activate data collected for a purpose other than advertising, in which case the conversation ends the engagement rather than starting it. The monthly arrangement: Continuous operation suits teams running more than a handful of audiences. New syncs, deprecation handling as platforms change their APIs two or three times a year, quarterly pruning, and one session where we switch off the syncs nobody used. That session normally removes about a third of them, and the removal is the part that keeps the rest trustworthy.