Direct answer: do not connect more AI to campaigns while every channel interprets permission differently. Build a decision layer between identity and execution. For each action, it should receive subject, contact point, purpose, channel, market, record source, policy version, validity, and revocation; return allow, suppress, or review; record the reason; and propagate the result downstream. AI may organize legacy fields and flag conflicts. It should not invent consent, choose a lawful basis, or override a revocation.
The problem is rarely a total absence of data. It is fragmentation. A form stores a checkbox, commerce keeps its own preferences, CRM has a Boolean field, the email platform knows subscriptions, the app records notifications, and media receives consent signals. When a customer changes channel, country, or purpose, each system applies a different memory. Commercial consequences include unexplained audience shrinkage, improper messages, incomplete personalization, incidents that cannot be reconstructed, and AI projects that stall in legal review.
Write the Permission Decision Contract
- 1. Subject and identity — reconciled person, account, and identifiers, with confidence and merge rules.
- 2. Purpose — why data will be used: service, measurement, personalization, marketing, or another approved purpose.
- 3. Channel and contact point — specific email, phone, SMS, messaging, push, cookie, device, or address.
- 4. Provenance — source, capture event, timestamp, displayed notice, version, and accountable system.
- 5. Market and policy — operating jurisdiction, brand, business unit, age or restriction, and rule version.
- 6. State — allowed, denied, revoked, expired, unknown, or conflicting, with effective dates.
- 7. Decision — allow, suppress, or review, with reason code, confidence, owner, and SLA.
- 8. Downstream evidence — action, destination, policy used, propagated update, and outcome.
Salesforce documents consent tied to purpose, contact point, where, when, how, and for how long it was granted. HubSpot distinguishes the lawful basis for processing data from subscription types used for communication. Those models show why one ‘consented’ field loses essential meaning. Your contract should preserve the granularity your operating case needs without blindly copying one vendor's data model.
Permission precedes pressure, deliverability, and autonomy
Four decisions look similar but are not interchangeable. Permission asks whether this purpose may use this contact point now. Contact pressure decides when and how often to communicate. Deliverability estimates whether a message will reach an inbox. An agent action contract defines what automation may do. A person may be eligible for transactional email, suppressed for marketing, below a frequency cap, and still have a technically valid address. Execute these decisions in order and preserve each reason.
Use AI to reconcile evidence, not create authority
- Use deterministic rules for revocation, blocks, purpose, channel, market, validity, precedence, and propagation.
- Use AI to map legacy field names, classify authorized free text, cluster conflict patterns, summarize a trail, and suggest remediation.
- Require human review for ambiguous bases, policy changes, vulnerable audiences, merged identities, campaign exceptions, and novel use.
- Prohibit the model from turning unknown into allowed, treating engagement as consent, or prioritizing revenue over suppression.
- Log model and rule versions, minimum inputs, output, confidence, override, and accountable owner.
Google explains that Consent Mode changes tag and SDK behavior based on choices about storage and use; in basic mode, a tag can be blocked until permission is granted. That solves a particular destination, not the entire enterprise contract. A central layer must distribute decisions to analytics, media, CRM, marketing automation, service, and product—and accept revocations back without creating loops or competing states.
Pilot one purpose, two channels, and one market
- Week 1 — choose one marketing purpose, email plus a second channel, one market, and a high-volume action; freeze the current rule as baseline.
- Week 2 — inventory fields, notices, sources, APIs, delays, owners, and 200 anonymized records, including conflicts and revocations.
- Weeks 3–4 — build the canonical view, reason codes, precedence, decision endpoint, propagation, and test cases.
- Week 5 — run in shadow mode; compare the central decision, each channel's decision, and independent review.
- Weeks 6–7 — release by cohort with a kill switch, review queue, and daily reconciliation.
- Week 8 — assess error, experience, effort, risk, and operating value; expand, repair, or remain manual.
The deliverable is not merely a preference center. It includes a purpose dictionary, data map, identity model, API contract, precedence matrix, reason-code catalog, decision ledger, revocation propagation, monitoring, tests, runbook, RACI, and retention policy. Test an opt-out during a campaign, merged identity, country change, late offline capture, shared account, hard bounce, new purpose, failed destination, and replayed event.
Measure decisions, propagation, and commercial effect
- Coverage — subjects, contact points, purposes, and destinations with a known state and usable provenance.
- Consistency — cross-system conflicts, divergent decisions, and records without a policy version.
- Quality — false allow, false suppress, human correction, reopened decisions, and identity precision.
- Operations — decision latency, revocation time to the last destination, backlog, incidents, and audit time.
- Experience — repeated opt-out, complaint, respected preference, recovery after correction, and unexpected contacts.
- Business — eligible audience, delivery, conversion, and revenue by cohort, without assigning causality to the layer alone.
Do not call audience growth a success if unknown states were promoted to allowed. Do not call audience decline a failure if the layer removed duplicates and inconsistent decisions. Compare the old and new policies, retain a holdout when practical, and monitor countermetrics. The NIST Privacy Framework is voluntary and risk-oriented; use it to structure governance, not as a compliance seal.
Scope investment and accountability
Cost rises with systems, channels, purposes, brands, markets, local rules, identity quality, volume, real-time needs, history without provenance, brittle APIs, and retention requirements. Batch may be enough for nightly reconciliation; revocation may require far faster propagation. Legal and privacy interpret limits; marketing owns purposes and experience; data owns identity and ledger; engineering integrates; RevOps governs CRM; security controls access; channel owners execute; audit validates evidence. A partner implements and transfers operations without replacing those owners.
Connect the layer to the existing stack
After permission, apply frequency limits in https://makinai.co/insights/en/ai-cross-channel-contact-pressure-crm-frequency and protect inbox delivery in https://makinai.co/insights/en/ai-email-deliverability-crm-personalization-inbox. Bound agent autonomy in https://makinai.co/insights/en/ai-agent-crm-action-contract-permissions, test journeys in https://makinai.co/insights/en/ai-test-broken-crm-customer-journeys, and resolve identity before propagating decisions in https://makinai.co/insights/en/ai-customer-identity-resolution-crm-duplicates. This order reduces patches and makes every suppression explainable.
Bring the system map, 200 anonymized consent and preference records, current rules, and 30 days of action logs. Through https://makinai.co/services/en/crm-ecommerce-commerce-transformation, MAKINAI can scope a pilot for one purpose, two channels, and one market—connecting CRM, data, and execution without requiring a new platform before the operating decision is proven.
The sources describe product models and controls plus a voluntary risk framework; they do not provide legal advice or prove revenue lift. The contract, architecture, pilot, metrics, and accountability model are MAKINAI editorial recommendations. Validate lawful basis, notices, markets, time limits, and rights with accountable professionals before production.