Customer Identity Resolution: Methods & CDP Use

Customer identity resolution is a data management process that matches and unifies customer identifiers — such as email addresses, device IDs, and cookies — across multiple touchpoints into a single persistent profile.

In one line: identity resolution answers “are these records the same person?” and then keeps that answer current. It decides matches two ways — deterministic matching (exact equality on a shared key such as a hashed email or loyalty ID) and probabilistic matching (a statistical confidence score from signals such as IP address, device, and behavior) — and assigns every matched record one persistent customer ID. A customer data platform (CDP) typically automates the whole cycle, increasingly in real time, so AI agents and personalization engines act on a complete profile within the same session.

Enterprises rarely pick a single method. They make three decisions: which matching rules apply where (deterministic first, probabilistic only where reach matters more than certainty), where the identity graph lives (inside the CDP, inside the data warehouse, or in a standalone tool), and how much of the tuning and monitoring they hand to machine learning and AI agents. The approaches table below compares those options side by side.

A concrete case: a shopper browses anonymously on her phone (cookie ID), logs in on her laptop with an email address, and scans a loyalty card in store. Without identity resolution, those are three customers in three systems. With it, the login ties the cookie to the email, the loyalty account carries the same email, and all three collapse into one profile. The stakes are commercial: McKinsey found that fast-growing companies drive 40% more of their revenue from personalization than slower-growing peers (McKinsey, 2021), and personalization is only as good as the identity behind it.

Multi-brand companies feel the gap first. Customers interact with each brand in isolation, so one person receives redundant messages from several brands, wasting spend and eroding trust. Resolving IDs across brands turns those fragments into one customer that every brand team can coordinate around.

Why Identity Resolution Matters

Consumers interact with brands across a multitude of devices and platforms. A customer might see a mobile ad in the morning, browse a website on a tablet during her commute, and open an email on her laptop at work. Without cross-device identity resolution, these three touchpoints appear to come from three different people. With identity resolution, they are stitched into a single customer profile, enabling consistent engagement and smarter next-best-action decisions.

The challenge extends beyond devices. Data captured about customers is often trapped inside platform-specific silos. A web analytics system that identifies users by cookie ID has no connection to email addresses captured in a marketing automation platform. These data silos prevent businesses from creating unified profiles and result in inconsistent messaging across channels.

Duplicate records also cost money directly. In a Forrester Consulting Total Economic Impact study commissioned by Oracle (March 2023), the modeled composite organization deduplicated 20% of its 15 million customer and prospect profiles, and Forrester valued the resulting cut in redundant marketing spend at $4 million in present value over three years. The figure is vendor-commissioned and modeled, not a benchmark, but the mechanism is general: every unresolved duplicate is a person you pay to reach twice.

How Identity Resolution Works, Step by Step

Every identity resolution system, whether a CDP feature or a warehouse job, runs the same five-step cycle. The steps matter because most accuracy problems trace back to one of them, not to the matching algorithm.

1. Collect identifiers. Ingest every identifier each source carries — cookie and device IDs from web and app, emails from forms and email platforms, phone numbers from support, loyalty and order IDs from commerce and point-of-sale systems.
2. Standardize. Lowercase and trim emails, format phone numbers to E.164, normalize addresses, and hash PII consistently. Skipping this step creates duplicates that are costly to reconcile after the fact, because “USER@Example.com” and “user@example.com” never match exactly.
3. Match. Apply deterministic rules on shared keys first, then probabilistic scoring where no shared key exists. Each match records its evidence and confidence.
4. Merge and assign a persistent ID. Matched records join one cluster in the identity graph and receive a stable customer ID that survives cookie expiry, new devices, and email changes. Merge rules decide which value wins when sources disagree — the basis of a golden record.
5. Maintain. Re-evaluate clusters as new events arrive: split over-merged profiles, retire stale identifiers, and propagate consent changes across every linked ID.

Steps 1 to 4 build the profile; step 5 is what keeps it trustworthy a year later, and it is the step most often under-resourced.

How Enterprises Approach Customer Identity Resolution

Every enterprise program combines several approaches, because no single one covers every source and use case. Matching methods decide whether two records are the same person; the identity graph stores those decisions; the deployment choice decides who runs and owns it. The sections below cover each in depth — this table is the map.

Most enterprises land on a hybrid: deterministic rules as the backbone, probabilistic matching scoped to use cases that tolerate error, a graph with merge lineage, and resolution running close to where activation happens, with the graph exportable to analytics. If most of your revenue comes from logged-in or loyalty customers, deterministic matching inside your activation platform covers most of the value; add probabilistic matching and a warehouse or standalone graph only when a named use case needs it.

Types of Identity Resolution

Two primary methods decide whether two records belong to the same customer — deterministic and probabilistic matching. A further technique, transitive matching, is a graph operation that extends those matches rather than a separate way to decide them, and adaptive matching chooses between the two methods by use case.

Deterministic Matching

Deterministic matching connects records by searching for exact equality across identifiers such as email, phone number, or login credentials. This approach delivers the highest accuracy when the shared key, such as a personal email, belongs to exactly one person, and works best when first-party data is readily available. Common deterministic keys include email addresses, loyalty program IDs, and authenticated session tokens.

Probabilistic Matching

Probabilistic matching estimates the likelihood that two records belong to the same customer using signals like IP address, device type, and behavioral patterns. While less certain than deterministic matching, it extends reach to anonymous visitors and cross-device identity stitching. Marketers must define confidence thresholds — typically 70-95% depending on use case — to determine what constitutes a positive match. Note that browser-based signals like fingerprinting have become less reliable as Safari, Firefox, and Chrome have progressively restricted or blocked these techniques.

Transitive Matching

Transitive matching discovers hidden connections between records that share no direct identifier. It is not a separate way to decide that two records match, but a graph operation that chains together links already established by deterministic or probabilistic matching. If record A matches record B via email, and record B matches record C via phone number, transitive matching infers that A and C belong to the same customer — even though they share no common key. This technique surfaces relationships that deterministic and probabilistic methods miss individually, particularly across fragmented offline and online data sources. Transitive matching requires careful threshold management to prevent over-merging, where unrelated profiles collapse into a single identity.

Adaptive Matching: Choosing by Use Case

Leading CDPs no longer force a single matching strategy across all use cases. Adaptive matching lets teams toggle between deterministic and probabilistic methods depending on the context. A transactional email campaign requires near-perfect accuracy and uses deterministic matching; an ad-targeting audience benefits from broader reach and uses probabilistic matching. This use-case-driven flexibility replaces the one-size-fits-all approach that dominated earlier CDP generations.

Cross-Device Identity Resolution

Cross-device resolution is where the two methods meet. The reliable bridge is authentication: when the same customer logs in on a phone app and a laptop browser, the login ID links both device identifiers deterministically, and every anonymous event either device recorded before that login can be attached to the profile. Without a login, linking devices falls back to probabilistic signals such as a shared home IP address and overlapping browsing patterns — useful for ad reach and frequency capping, but too weak to drive an email, an on-site greeting, or any message that shows a customer’s name, order history, or account details. Two platform limits shape the design. Safari’s Intelligent Tracking Prevention deletes cookies written by client-side script after seven days without user interaction on the site (WebKit), so a returning web visitor can look new, and App Tracking Transparency requires apps on iOS 14.5 and later to get the user’s permission before tracking them across other companies’ apps or reading the advertising identifier (Apple). In practice, every reason to log in — a loyalty program, saved carts, member pricing — is also an identity investment. Durable cross-device graphs rest on first-party logins, identifiers set by the site’s own servers (since Safari 16.4, cookies set by servers on third-party infrastructure are also capped at seven days), and reasons for customers to sign in, not on device fingerprints.

Identity Resolution and Predictive Modeling

An additional benefit of identity resolution is enabling more accurate predictive modeling. Resolved profiles produce the training data necessary to identify lookalike audiences within other customer sets. With automated predictive modeling built into an enterprise-grade CDP, the model-building engine correlates hundreds of profile attributes to surface the most meaningful features. To build a reliable predictive model, you first need a large set of known customers as training data — which is why identity resolution is a prerequisite for effective AI.

Identity Resolution Architectures

How identity resolution runs depends on where customer data lives. Two architectural models dominate the market, each with distinct trade-offs for data ownership, latency, and governance.

CDP-Native Identity Resolution

In a CDP-native architecture, customer data is ingested into the CDP platform, and identity resolution runs inside the vendor’s infrastructure. The advantage is real-time event stitching — incoming events are matched to profiles as they arrive, enabling in-session personalization and instant audience segmentation. Full graph reconciliation (merging split profiles, replaying history) still runs in batch even in CDP-native systems, but the event-level latency is sub-second. The trade-off is that the identity graph lives inside the vendor’s system, which means PII is stored outside the organization’s own data infrastructure and subject to the vendor’s security perimeter.

Warehouse-Native Identity Resolution

Warehouse-native architectures run identity resolution directly inside the organization’s existing data warehouse or data cloud. The CDP reads from and writes back to the warehouse — customer data typically stays inside the organization’s infrastructure, though the exact data flow depends on the implementation. This model appeals to enterprises with strict data governance requirements or existing warehouse investments. Historically, warehouse-native resolution ran only in batch cycles, but modern implementations increasingly use change-data-capture (CDC) and streaming pipelines to narrow the latency gap with CDP-native systems.

Real-Time vs Batch Resolution

The dimension buyers ask about most is speed: does the CDP resolve identity in real time or in batch? Real-time resolution matches incoming events to profiles in milliseconds, so an AI agent or personalization engine can act on a complete profile within the same session. Batch resolution reprocesses the identity graph on a schedule — hourly or nightly — which is acceptable for campaign segmentation and churn scoring but too slow for in-session decisioning. Most production CDPs do both: real-time event stitching for immediate recognition, and periodic full-graph reconciliation to correct merges and replay history. As agentic use cases grow, real-time resolution is shifting from a premium option to the baseline expectation.

Beyond People: Household and B2B Account Identity Resolution

Traditional identity resolution focuses on resolving individual people — matching a cookie ID to an email to a loyalty number. But modern business requirements extend beyond individual profiles.

Household graphs group family members who share a physical address, enabling suppression logic (don’t mail two offers to the same household) and household-level spend analysis. Account graphs connect individuals to business entities for B2B use cases, mapping multiple contacts to a single company and tracking account-level engagement. Custom entity graphs model domain-specific relationships — insurance policies linked to policyholders, pets linked to pet owners, vehicles linked to drivers.

B2B identity resolution has its own failure points. The same buyer often appears under a personal email at a webinar and a work email in the CRM, so person-level matching needs both. Contacts map to accounts by email domain, but domains break on subsidiaries, regional entities, and free-mail addresses, so account matching usually adds firmographic keys such as company name, address, and registry IDs. And when someone changes jobs, their work email should leave the old account rather than drag their history into the new one.

Supporting multiple identity graphs within a single platform eliminates the need to maintain separate systems for B2B and B2C audiences. Organizations with complex entity relationships — multi-brand retailers, insurance companies, healthcare systems — should evaluate whether a CDP supports graph types beyond individual people before committing to a platform.

Transparency and Governance in Identity Resolution

As identity graphs grow in complexity, the ability to explain why two records were merged becomes as important as the merge itself. Early identity resolution systems operated as black boxes — profiles were unified, but the logic behind each merge decision was opaque. This created two problems: compliance teams couldn’t audit merge decisions for privacy regulation adherence, and data teams couldn’t diagnose accuracy issues when profiles were incorrectly merged or left fragmented.

Modern identity resolution demands five governance capabilities:

1. Merge lineage — a full audit trail showing which identifiers triggered each merge, when it happened, and what produced it: a deterministic or probabilistic match, or the transitive inference path that chained them.
2. Confidence scoring — every match should carry a confidence score that downstream systems can filter on. A loyalty email requires near-certain identity; a prospecting audience can tolerate lower confidence.
3. Self-healing — automated detection and correction of over-merges (unrelated profiles collapsed together) and under-merges (fragments of the same person left separate). Self-healing systems continuously monitor profile stability and flag anomalies rather than waiting for manual data hygiene.
4. Consent-aware merging — identity resolution must respect consent signals. If a customer withdraws consent in one channel, the merged profile must propagate that withdrawal across all linked identifiers to remain compliant with GDPR, CCPA, and other privacy regulations.
5. Cross-border data residency controls — for global organizations, identity graphs must enforce data residency rules, ensuring that profiles are resolved and stored within jurisdictional boundaries when regulations require it.

Measuring Identity Resolution Quality

Merge errors throw no exceptions, so identity resolution has to be monitored with metrics rather than trusted by default. Four numbers catch most problems, and an automated monitor should alert on sudden shifts in any of them:

• Profile compression ratio — source records divided by resolved profiles. A sharp rise after a new source is connected usually signals over-merging; a ratio near 1.0 means matching is barely happening.
• Oversized clusters — profiles carrying an implausible number of identifiers (dozens of emails or devices). These are almost always hub identifiers such as a shared office IP or a store’s checkout terminal chaining strangers together.
• Match rate by source — the share of each source’s records that join an existing profile. A source that suddenly drops is usually a formatting or hashing change upstream, not a change in customers.
• Profile stability — how often profiles merge and split week over week. Constant churn means thresholds are set near the edge of the score distribution.

Sampling is the check that numbers cannot replace: have a data steward review a few dozen random merges each month, including the confidence evidence behind each one.

Validating and Load-Testing Before Production

Monitoring catches drift after launch; testing decides whether a configuration should launch at all. Validation starts with a labeled sample: a few hundred record pairs that stewards have marked as same-person or different-person, sampled across match-score bands rather than at random and weighted toward the borderline zone and the hard cases (shared family emails, typos in names, recycled phone numbers). A random sample is almost all obvious non-matches and flatters the results. Running the candidate rules against that set gives precision (the share of proposed merges that were correct) and recall (the share of true matches the rules found); because recall only covers pairs the candidate step surfaced, also seed known true matches that share no obvious key. The set then becomes a regression test that every rule or threshold change must pass.

Load testing answers a different question: does resolution hold up at peak volume? Replay a recorded peak day, such as a holiday sale, at one to several times its event rate and measure four things — profile lookup latency at the 95th and 99th percentiles, merge throughput, how far behind real time the graph falls under load, and how long a full re-resolution of history takes. Watch for concurrency faults too: two events for the same new customer arriving milliseconds apart can create duplicate profiles if merges are not handled atomically. Any one of these breaking at peak means agents and campaigns read stale or split profiles exactly when traffic matters most. In a vendor evaluation, make both tests part of the proof of concept: give every shortlisted vendor the same labeled sample and the same peak-day replay, and compare precision, recall, and p99 lookup latency side by side instead of accepting a demo dataset.

Identity Resolution vs Identity Graph vs Entity Resolution

Three terms get used interchangeably and should not be. Identity resolution is the process; the identity graph is the data structure that process produces; entity resolution is the broader discipline that applies the same matching to any entity, not only customers.

In practice, a CDP’s identity resolution writes to an identity graph, and the unified profile is what downstream tools read.

Identity Resolution Challenges by Funnel Stage

Top of Funnel: Acquiring Unknown Prospects

At the top of the funnel, the challenge is identifying prospects who have never visited your properties. Although Google reversed its plan to deprecate third-party cookies in Chrome, Safari and Firefox have blocked them for years, and privacy regulations continue to limit cross-site tracking. Solutions include contextual advertising (showing ads based on page content rather than browsing history), lookalike modeling from first-party data, and alternative ID solutions like UID2.0, ID5, and RampID. Data clean rooms also enable privacy-safe matching of anonymized first-party data with partners.

Middle of Funnel: Converting Anonymous to Known

The middle of the funnel is where prospects have shown interest but remain anonymous. Communication relies on less personal identifiers like cookies and device IDs, making personalization difficult. Key challenges include cookie-based tracking limitations under privacy regulations and converting anonymous visitors into known leads without creating friction. Strategies like progressive profiling, gated content, and Conversion APIs (server-side event tracking that bypasses browser-level restrictions) help bridge this gap.

Bottom of Funnel: Deepening Loyalty

At the bottom of the funnel, recognized identities enable targeted loyalty programs, personalized offers based on purchase history, and consistent omnichannel experiences. The challenge here shifts to managing ID hierarchies — ensuring a promotional offer for a parent isn’t sent to their child, or consolidating multiple accounts associated with a single household.

Identity Resolution Solutions

Effective identity strategies combine multiple approaches depending on funnel stage:

• Contextual advertising for privacy-safe top-of-funnel reach (Google’s Topics API, once pitched for this role, is being retired along with most of the Privacy Sandbox — Google, October 2025)
• Conversion APIs and server-side tracking for accurate mid-funnel attribution without browser cookies
• Deterministic matching on first-party identifiers for bottom-of-funnel personalization
• Data clean rooms for second-party data partnerships that expand identity coverage without exposing raw PII

Built-In vs Standalone Identity Resolution

Identity resolution was once a specialized capability that justified a separate vendor. In the early CDP era (2016–2020), many organizations purchased standalone tools because their marketing platforms lacked native matching.

That market has fundamentally changed. Amperity built its reputation as an identity-resolution-first CDP during that early era, when specialized matching was a genuine differentiator, and has since expanded into activation as well. Across the market more broadly, that specialized capability is no longer scarce: major Agentic CDP platforms now include AI-assisted identity resolution as a built-in feature — deterministic and probabilistic matching, with machine learning that vendors credit for accuracy gains as data accumulates — closing much of the accuracy gap that once justified a standalone vendor for common use cases.

For most organizations, a separate identity resolution vendor is no longer necessary. Modern Agentic CDPs embed identity resolution into the same platform that handles segmentation, AI decisioning, and data activation — meaning unified profiles are immediately actionable without copying data to a separate system.

Standalone identity resolution tools may still add value in specific scenarios: organizations with hundreds of data sources and billions of records, multi-brand portfolios where a shared identity graph feeds analytics pipelines, ML feature stores, and BI tools beyond marketing activation, or enterprises that intentionally decouple identity infrastructure from activation for architectural flexibility. But for the majority of CDP buyers, identity resolution has become table stakes — a necessary built-in capability, not a differentiator worth a separate vendor contract.

Identity Resolution and AI Agents

In the agentic era, identity resolution shifts from a background data process to a real-time capability that AI agents depend on continuously. An AI agent selecting the next best action for a customer needs a fully resolved profile — not fragments scattered across three databases.

Identity resolution powers the UNIFY stage of the Customer Intelligence Loop. Without it, the loop breaks at its second step: agents cannot understand, decide, or engage a customer they cannot identify. For in-session personalization and real-time decisioning, Agentic CDPs perform identity resolution continuously rather than in batch — agents need current, resolved profiles at API speed. Batch-oriented use cases like churn prediction and email campaigns can tolerate hourly updates, but the trend toward agentic automation is shifting the baseline expectation toward real-time resolution.

The requirement for real-time resolution creates challenges for identity-only vendors. An identity platform that produces unified profiles but cannot act on them forces a handoff to external activation systems. That handoff can introduce delayed feedback loops, PII duplication across vendor boundaries, and latency — though the severity depends on the specific architecture and integration patterns used.

How AI Is Used in Customer Identity Resolution Today

Statistical matching in identity resolution is older than the AI label. Probabilistic matching rests on the Fellegi–Sunter record linkage model (Journal of the American Statistical Association, 1969), which weighs how strongly each field’s agreement or disagreement indicates a match. Current systems add three things. Trained classifiers learn match weights from steward-labeled pairs instead of hand-set rules. Similarity models compare names and addresses that are spelled differently but mean the same thing. And language models can clean messy free-text fields before matching, which calls for a data-processing and residency review wherever raw PII goes to a hosted model. Each improves recall, and each needs the same safeguards as rule-based matching: confidence scores, merge lineage, and a labeled test set to prove the model is right. When a vendor calls its identity resolution “AI-powered,” ask which of these it means, what the model was trained on, and whether every model-made merge carries the same lineage and confidence score as a rule-based one.

Agentic Identity Resolution (Emerging)

An emerging evolution moves AI agents from consuming resolved identities to performing identity resolution itself. Rather than relying on manually configured YAML files, SQL rules, or visual rule builders, agentic identity resolution uses AI agents to automatically discover identifier relationships, recommend matching thresholds, detect over-merges, and continuously optimize graph quality. This approach addresses a long-standing operational bottleneck: traditional identity resolution configuration requires specialized data engineering skills, and maintaining accuracy as data sources change demands ongoing manual tuning. While still early — most production implementations today combine rule-based systems with ML-assisted matching rather than fully autonomous agents — the trajectory points toward AI agents that monitor identity graph health, flag anomalies, and adjust matching logic with decreasing human oversight.

Identity Without Activation Is Incomplete

Creating unified customer profiles is only the first step. The business value of identity resolution is realized when those profiles are activated — when an AI agent reads a unified profile, decides on the optimal action, sends a message, and learns from the outcome. Platforms like Treasure AI (formerly Treasure Data) are designed with identity resolution and activation in the same platform, eliminating the handoff between unifying customer data and acting on it.

When identity resolution and activation live in separate systems, organizations face additional integration challenges:

• Feedback loop latency — The identity platform may not know what happened after the profile was sent downstream. Outcome data must flow back through separate data pipelines before models can learn from it. The severity depends on the integration architecture — reverse ETL and event-streaming approaches have significantly reduced this gap.
• PII duplication — Activation syncs can copy personally identifiable information to downstream vendors, adding compliance obligations under GDPR and CCPA. Warehouse-native approaches mitigate this by querying data in place rather than copying it.
• Latency — The handoff from identity platform to activation platform ranges from near-real-time (streaming integrations) to hours (batch-based activation). Use cases that require agentic marketing closed loops need the former; batch-oriented campaigns can tolerate the latter.

Organizations evaluating identity resolution should consider how tightly it integrates with their activation stack — a CDP RFP template is a practical way to put that integration question to every vendor on the same terms. Bundled platforms like Treasure AI (formerly Treasure Data) keep identity resolution, AI decisioning, and multi-channel activation within a single platform boundary — eliminating the handoffs that cause PII duplication and feedback loop latency. Best-of-breed approaches offer architectural choice but require careful orchestration to close feedback loops.

FAQ

What is the difference between deterministic and probabilistic identity resolution?

Deterministic identity resolution matches records by exact equality across identifiers like email, phone number, or login credentials. It delivers high accuracy when the shared keys are unique to one person and first-party data is available. Probabilistic identity resolution estimates the likelihood that two records belong to the same customer using signals like IP address, device type, or behavioral patterns. It extends reach to anonymous visitors but requires confidence thresholds to determine positive matches.

How does identity resolution work without third-party cookies?

Identity resolution without third-party cookies relies on first-party data, server-side tracking, and alternative ID solutions. Although Chrome reversed its cookie deprecation plan, Safari and Firefox have blocked third-party cookies for years. Businesses use Conversion APIs for server-side event tracking, alternative identifiers like UID2.0, ID5, and RampID, and data clean rooms for privacy-safe partner matching.

Why is identity resolution important?

Identity resolution is the foundational capability that makes customer data platforms and cross-device identity stitching useful. Without it, customer touchpoints remain fragmented across systems, preventing personalization and wasting marketing resources on redundant messaging. Identity resolution creates the unified profiles that power segmentation, predictive modeling, AI decisioning, and omnichannel activation.

What are the benefits of identity resolution?

Identity resolution unifies fragmented customer data into a single profile, enabling accurate personalization, consistent cross-channel messaging, and reliable measurement. It cuts wasted spend on duplicate outreach — a Forrester Consulting study commissioned by Oracle (2023) modeled $4 million in three-year savings from deduplicating 20% of 15 million profiles. Resolved profiles also supply the complete training data that predictive models and AI agents depend on.

How does identity resolution work for B2B?

B2B identity resolution links individual contacts to the business accounts they belong to, resolving identity at both the person and company level. Beyond matching a person’s emails and devices, B2B requires account graphs that map the multiple contacts in a buying committee to a single company node — keeping each person’s identity distinct while tracking engagement at the account level. This supports account-based marketing without collapsing separate individuals into one profile.

Do I need a separate identity resolution vendor?

For most organizations, no. Modern CDPs include deterministic and probabilistic identity resolution as a core feature, and the accuracy gap between standalone vendors and built-in matching has narrowed significantly. A separate vendor may still add value for organizations with billions of records, hundreds of data sources, or multi-brand identity graphs that feed systems beyond marketing activation.

Which CDP platforms combine identity resolution with native activation?

Look for a platform where identity resolution, AI decisioning, and omni-channel activation share one runtime, so the vendor handoff between unification and action disappears. Treasure AI (formerly Treasure Data) is an Agentic CDP built that way. This closed-loop architecture reduces PII duplication risks and ensures AI agents can learn from every customer interaction in real time. Other Agentic CDPs share this architectural approach, though implementation details vary by vendor.

What is warehouse-native identity resolution?

Warehouse-native identity resolution runs matching logic directly inside an organization’s existing data warehouse rather than inside a CDP vendor’s infrastructure. Customer data generally remains inside the organization’s own environment, which simplifies data governance and eliminating PII duplication. Modern CDC-based approaches are narrowing the latency gap with CDP-native systems.

What is an identity graph?

An identity graph is a data structure that maps relationships between customer identifiers — linking emails, device IDs, phone numbers, and cookies to a single unified profile. Modern identity graphs extend beyond individuals to model households, B2B accounts, and custom entities like insurance policies or vehicles, enabling cross-device identity resolution and household-level suppression.

How can you monitor identity resolution accuracy automatically?

Track a small set of graph-health metrics and alert when they shift: profile compression ratio, oversized clusters, match rate by source, and merge/split churn. A sudden jump in any of them usually means an upstream formatting change, a hub identifier chaining strangers together, or a threshold set too loose. AI-assisted tools increasingly run these checks continuously, but a monthly human review of sampled merges remains the ground truth.

What is agentic identity resolution?

Agentic identity resolution is an emerging approach that uses AI agents to automatically build, optimize, and maintain identity graphs. Instead of manually configuring matching rules and thresholds, AI agents discover identifier relationships, detect merge anomalies, and continuously tune graph quality — reducing the specialized data engineering effort traditionally required.

Related Terms

• Customer Data Platform (CDP) — The primary platform that performs identity resolution at scale
• Agentic CDP — AI-first CDP with built-in identity resolution and real-time matching
• Customer Intelligence Loop — The five-stage cycle where identity resolution powers the UNIFY stage
• AI Decisioning — What happens after identity is resolved
• Data Activation — Making unified profiles actionable across channels
• Single Customer View (SCV) — The output of identity resolution
• Personalization — Enabled by resolved customer identities

Further Reading: Identity Resolution Is Table Stakes: What CDPs Actually Need in the AI Era

This article is also available in: 名寄せとは？CDPにおける仕組みと種類を解説 · Identity resolution: o que é, tipos e uso na CDP · Identity resolution : définition, méthodes et CDP