For CDP architects + customer-data leadership + identity engineers

LiveRamp resolves identity across the broader publisher ecosystem. The first-party algorithm running across your POS, your booking system, and your loyalty platform is your wiring.

Third-party cookies are deprecated. iOS 14 ATT limited cross-app tracking. The probabilistic device- fingerprinting that worked five years ago has degraded. The first-party deterministic-plus-probabilistic resolution algorithm running on operator-owned signals with confidence-weighted merge decisioning + identity- conflict reconciliation + per-jurisdiction PII gating is what now produces the unified customer record at multi-location scale.

By Jay ChristopherMay 29, 202611 min read

What this gets you

Two-layer resolution algorithm— deterministic on explicit identifiers (email + phone + loyalty ID + payment instrument + hashed identifier with matching salt) at near-100 percent confidence; probabilistic on inferred signals (device + IP + behavior + geo + temporal) with confidence-varying scoring.
Confidence-weighted merge decisioning— high-confidence matches auto-merge; borderline matches route to analyst review with signal evidence surfaced; low-confidence matches stay separate. Reviewer decisions feed confidence- threshold tuning per signal type per cycle.
Identity-conflict reconciliation— when two confident matches disagree (signal stream A says identity-1 matches identity-2; stream B says identity-1 matches identity-3), the conflict- resolution layer evaluates strength and produces a winning merge with the losing match audit-logged.
Post-cookie + iOS 14 ATT first-party-data strategy — increased weight on deterministic identifiers, expanded first-party-data collection via offline POS plus call-tracking plus walk-in foot-traffic plus loyalty, probabilistic resolution focused on intra-session and intra-device signals where deprecation has less impact.
Per-jurisdiction PII policy gating + per-vertical reconciliation policies — GDPR consent state + CCPA opt-out + HIPAA BAA coverage + cannabis-state PHI restrictions gate which identifiers participate in matching; healthcare conflicts route more conservatively; financial services require explicit reviewer approval for multi-PII-anchor merges.

The third-party identity graph stopped working three years ago

A 200-location operator wired LiveRamp into the CDP in 2019 to resolve identity across publishers and ad platforms. The third-party identity graph linked the operator first-party email list to the third-party cross-publisher identifier set. The operator could target paid-media audiences against the resolved identity. The probabilistic device-fingerprinting layer ran underneath LiveRamp and produced ~40-50 percent additional cross-device matches against the pure-deterministic baseline.

Three years later the world changed. Third-party cookies are deprecated. iOS 14 App Tracking Transparency limited cross-app tracking. The cross- publisher third-party graph that LiveRamp maintains has thinned because the upstream identity signals fed into the graph have thinned. Probabilistic device- fingerprinting that worked at 40-50 percent additional match rate has dropped to 10-15 percent. The operator addressable audience has shrunk. The cross-channel attribution that depended on the resolved identity has degraded.

The new strategy is first-party-first. Increase the weight on deterministic identifiers — encourage more frequent login in the customer experience, push loyalty signup at every touchpoint, capture phone and email at booking, link loyalty to payment- instrument fingerprint at POS. Expand the deterministic anchor set across offline channels. Re-tune probabilistic resolution to focus on intra- session and intra-device signals where cookie deprecation has less impact. Run the first-party identity-resolution algorithm on the operator-owned substrate the customer-graph agent maintains.

The Resolve axis on the customer-graph agent runs the algorithm. Deterministic layer first — match on email + phone + loyalty ID + payment-instrument fingerprint + hashed identifier with matching salt at near-100 percent confidence. Probabilistic layer second — match on device + IP + behavior + geo + temporal with confidence-varying scoring. Confidence- weighted merge decisioning produces the canonical customer record across every device, every channel, every location.

What is in market — and what each category leaves to you

The third-party identity-graph layer is thinning. The first-party operator-owned resolution algorithm at multi-location scale is architecture.

Third-party identity graphs — LiveRamp, Acxiom, Throtle, Tapad (Experian), Neustar

Excellent at cross-publisher identity linking across the broader consumer ecosystem. Match rate has thinned with cookie deprecation and ATT. The third-party graph remains valuable for paid-media audience targeting and look-alike modeling; it is no longer sufficient on its own for resolving customer identity across the operator-owned data estate.

CDP identity layers — Adobe Identity, Treasure Data, Tealium AudienceStream, Salesforce CDP, mParticle, Segment, BlueConic

Strong at resolving identity within the operator data estate using the platform identity-graph primitive. Deterministic-plus-probabilistic algorithm tuning per identifier mix, confidence- weighted merge decisioning, identity-conflict reconciliation, and per-vertical regulatory reconciliation policies are operator-side architecture above the CDP identity primitive.

First-party identity specialists — Habu, Hightouch

Strong at first-party resolution and reverse-ETL identity-graph projection to downstream destinations. The customer-graph-integrated algorithm running on the operator substrate with the per-vertical regulatory reconciliation policies sits at the customer-graph layer.

Open-source identity-resolution — PostHog (group-aware identity), Snowplow, RudderStack

Capable foundations for operators with in-house data-engineering capacity. The two-layer deterministic-plus-probabilistic algorithm implementation, the confidence-weighted merge decisioning, the identity-conflict reconciliation, and the per-jurisdiction PII policy engine are the operator-side build above the open-source layer.

The CRM-deduplication script the data team runs every Saturday

The status quo at most multi-location operators below the enterprise-CDP tier. The data team runs a SQL script every Saturday that merges CRM contacts matching on email or phone exactly. The script handles deterministic matches. Probabilistic matches stay as separate contacts. Cross-device identity stays unresolved. The data team treats it as good enough until the next attribution audit surfaces the gap.

The pipeline, end to end

Position in the 7-axis customer-graph pipeline. Ingest (behavioral-signal-ingestion) + Resolve (this skill) + Version (versioned-customer-history) + Compute-LTV (ltv-math-primitives) + Cohort (behavioral-cohort-computation) + Subscription-Ingest + Emit-Change (customer-change-event-emission). The Resolve axis reads from Ingest and writes to Version-and-downstream.
Deterministic resolution layer. Match on explicit shared identifiers — email + phone + loyalty ID + payment-instrument fingerprint + hashed identifier with matching salt + government ID hash. Match confidence near-100 percent when identifiers match. Deterministic layer handles high-confidence cases first to constrain the probabilistic search space.
Probabilistic resolution layer. Match on inferred signals — device fingerprint (canvas + audio + WebGL + user-agent + IP), behavioral pattern (page-sequence + dwell-time + interaction-pattern), location proximity (geo + time-window), temporal sequence (events fitting plausible customer-journey timing), household-level aggregation (shared IP plus shared device plus shared address signal).
Confidence-weighted merge decisioning. Every candidate match carries a confidence score. High-confidence (deterministic match plus consistent probabilistic signals) auto-merge. Borderline (deterministic match but conflicting probabilistic signals, or strong probabilistic match without deterministic anchor) route to analyst-review queue with signal evidence surfaced. Low-confidence stay separate until additional signal arrives.
Identity-conflict reconciliation. When two confident matches disagree (signal stream A says identity-1 matches identity-2; stream B says identity-1 matches identity-3), conflict-resolution layer evaluates strength (signal count + signal recency + signal type weight) and produces a winning merge with the losing match logged for audit. Persistent conflicts route to analyst review.
Per-jurisdiction PII policy gating. GDPR consent state + CCPA opt-out + HIPAA BAA coverage + cannabis-state PHI restrictions gate which identifiers participate in matching. PHI under HIPAA cannot match cross-channel without explicit BAA-covered consent. The policy engine shares substrate with the marketing-compliance cross-agent overlay.
Per-vertical reconciliation policies. Healthcare conflicts under HIPAA route more conservatively than non-regulated conflicts. Financial-services under FINRA require explicit reviewer approval for any merge involving multiple PII anchors. Cannabis state-by-state policies vary per state. Per-vertical policies tune the threshold and the escalation requirement per cycle.
Post-cookie + iOS 14 ATT first-party strategy. Deterministic identifier weights increase. First- party-data collection expands through offline POS + call-tracking + walk-in foot-traffic + booking events + loyalty interactions. Probabilistic resolution focuses on intra-session and intra-device signals where cookie deprecation has less impact. The post-cookie architecture shift reshapes the algorithm rather than tunes its parameters.
Reviewer decision feedback loop. Analyst-review queue decisions feed the confidence- threshold tuning per signal type per cycle. Decisions that consistently approve borderline matches loosen the auto-merge threshold; decisions that consistently reject borderline matches tighten it. The system self-improves at where the threshold should sit for each operator data mix.
Match-rate, precision, and recall measurement. Match rate (events resolving to a known identity vs total events). Precision (merges that prove correct against periodic ground-truth audit). Recall (true matches the pipeline found vs estimated total true matches). Per-vertical metrics tracked separately for regulated-vertical reconciliation tuning.
Right-to-deletion propagation. Deletion events propagate through the customer- data-orchestration emit-change pipeline. Affected identities tombstone in the version store. Subsequent resolution skips the tombstoned identity. The tombstone propagates to downstream consumers so they stop targeting or referencing the deleted identity.
Cross-arc integration. The Resolve axis feeds the Version axis (versioned- history substrate), the Compute-LTV axis (LTV computed on resolved customer record), the Cohort axis (behavioral cohorts computed on resolved customer record), and the Emit-Change axis (changes to the resolved record broadcast to downstream consumers). The seven-axis customer-graph pipeline operates on the resolved-identity substrate the Resolve axis produces.
ROI measurement. Match-rate trend per cycle. Cross-device attribution accuracy pre vs post deployment. CRM-duplicate rate pre vs post (target dropping). Marketing attribution accuracy on multi-touch journeys against ground- truth audit. Per-vertical attribution-completeness metrics. Signal feeds threshold-and-policy tuning per cycle.

Frequently asked

What is identity resolution software?

Identity resolution software links the multiple identifiers a single customer touches (anonymous device cookies, login emails, hashed phone numbers, loyalty IDs, payment-instrument fingerprints, household-level aggregations) into one canonical customer record. The category covers third-party-graph providers (LiveRamp, Acxiom, Throtle, Tapad now Experian, Neustar) that maintain cross-publisher identity graphs of the broader consumer ecosystem, enterprise CDPs with identity layers (Adobe Identity, Treasure Data, Tealium AudienceStream, Salesforce CDP, mParticle, Segment, BlueConic), and identity-resolution specialists with first-party data emphasis (Habu, Hightouch). The operator-owned first-party resolution algorithm that runs deterministic-plus-probabilistic matching with confidence-weighted merge decisioning on the customer-graph substrate at multi-location scale is operator-side architecture.

What is the difference between deterministic and probabilistic identity resolution?

Deterministic resolution matches on explicit shared identifiers — same email address, same phone number, same loyalty ID, same hashed identifier with matching salt, same payment-instrument fingerprint. Match confidence runs near-100 percent when identifiers match because the match relies on an unambiguous shared key. Probabilistic resolution matches on inferred signals — device fingerprint (canvas plus audio plus WebGL plus user-agent plus IP), behavioral pattern (page-sequence plus dwell-time plus interaction-pattern), location proximity (geo plus time-window), temporal sequence (events that fit a plausible customer-journey timing), household-level aggregation. Match confidence varies per signal mix. Real-world multi-location identity resolution needs both layers running in sequence.

How is this different from LiveRamp, Acxiom, Throtle, Adobe Identity, Treasure Data, or Tealium AudienceStream?

LiveRamp, Acxiom, Throtle, Tapad/Experian, Neustar ship third-party identity graphs that link identifiers across the broader consumer-data ecosystem (publishers, broadcasters, retailers, ad platforms). They are excellent at cross-publisher resolution. Adobe Identity, Treasure Data, Tealium AudienceStream, Salesforce CDP, mParticle, Segment, BlueConic ship CDP identity layers that resolve within the operator data estate. The operator-owned first-party resolution algorithm that runs deterministic-plus-probabilistic matching tuned per identifier mix, the confidence-weighted merge decisioning that handles borderline matches with analyst-review escalation, the identity-conflict reconciliation when two confident matches disagree, the per-jurisdiction PII policy that gates which identifiers can be used in matching under HIPAA-plus-cannabis-state-plus-GDPR, and the integration with the broader 7-axis customer-graph pipeline are operator-side architecture on top of the platform layer.

How does confidence-weighted merge decisioning work?

Every candidate match between two identity records carries a confidence score from the resolution algorithm. High-confidence matches (deterministic identifier match plus consistent probabilistic signals) auto-merge into the canonical customer record. Borderline matches (deterministic match but conflicting probabilistic signals, or strong probabilistic match without deterministic anchor) route to an analyst-review queue with the candidate match plus the signal evidence surfaced. Low-confidence matches stay separate until additional signal arrives. Reviewer decisions feed the confidence-threshold tuning per signal type per cycle. The system self-improves at where the auto-merge threshold should sit for each operator data mix.

How do you handle the post-cookie + iOS 14 ATT first-party-data strategy?

Third-party cookies are deprecated and iOS 14 App Tracking Transparency limits cross-app tracking. Probabilistic resolution that relied heavily on device fingerprinting has degraded coverage. The strategy compensates three ways. First, increase the weight on deterministic identifiers (logged-in email plus phone plus loyalty ID) by encouraging more frequent login in the customer experience. Second, expand first-party-data collection (POS receipts plus call-tracking plus walk-in foot-traffic plus booking events plus loyalty interactions) to provide deterministic anchors in offline channels. Third, focus probabilistic resolution on intra-session and intra-device signals where cookie deprecation has less impact. The post-cookie strategy is a substantive architecture shift, not a configuration tweak.

How do you handle identity-conflict reconciliation when two confident matches disagree?

A single customer can produce conflicting identity matches when one signal stream says identity-A matches identity-B and another stream says identity-A matches identity-C. The conflict-resolution layer evaluates the strength of each match (signal count + signal recency + signal type weight) and produces a winning merge with the losing match logged for audit. Persistent conflicts that resist the algorithm route to analyst review with the conflict evidence surfaced. Per-vertical reconciliation policies tune the weights — healthcare conflicts under HIPAA route more conservatively than non-regulated conflicts; financial-services under FINRA require explicit reviewer approval for any merge involving multiple PII anchors.

Hire the agent that runs the Resolve axis

The customer-data-graph agent owns the 7-axis pipeline — Ingest + Resolve + Version + Compute-LTV + Cohort + Subscription-Ingest + Emit-Change — sitting on top of whichever third-party identity graph (LiveRamp, Acxiom, Throtle, Tapad, Neustar), enterprise CDP (Adobe Identity, Treasure Data, Tealium AudienceStream, Salesforce CDP, mParticle, Segment, BlueConic), or open-source pipeline (PostHog, Snowplow, RudderStack) you license downstream. Two-layer deterministic-plus-probabilistic algorithm, confidence- weighted merge decisioning, identity-conflict reconciliation, per-jurisdiction PII policy gating, per-vertical reconciliation policies, post-cookie first-party-data strategy.

Hire the customer-graph agent

We scope on the call and send a private checkout link after.