The Definition Problem
Why Identity Resolution Is Hard to Define
Marketing has no shortage of jargon. The most dangerous terms are not the ones nobody understands. They are the ones everybody thinks they understand.
Identity resolution is one of those terms.
I have built a company around identity resolution. I consider myself an expert in the field. Yet when someone tells me they need identity resolution, I still have to stop and ask: What exactly do you need it to do?
Are they trying to connect an ad click to a purchase that happens five days later? Recognize an otherwise unknown website visitor? Match a personal email address to a work email? Enrich loyalty records with household data? Make CRM contacts reachable on social media? Identify target accounts visiting a website? Improve measurement accuracy?
All of these are routinely described as identity-resolution problems. But they depend on different data, require different capabilities, create value through different channels, and should be measured differently.
The definition problem creates two kinds of uncertainty.
The Need Is Clear. The Solution Is Not.
Marketers can see fragmented journeys, anonymous traffic, or incomplete records—but may not know which identity capability, provider, or success metric fits.
The Opportunity Itself Is Hidden.
Marketers may accept weak recognition, suppression, attribution, and lifecycle continuity without realizing that identity can raise the performance ceiling.
Sometimes the business need is clear, but the solution is not. A marketer may know that customer journeys are fragmented, anonymous traffic is going unrecognized, or customer records are incomplete—yet still struggle to determine which type of identity capability is required, which providers should be considered, and how success should be measured.
Other times, the opportunity itself is unclear. Marketers may not realize what stronger identity resolution could improve across audience acquisition, conversion, retention, and measurement. They may accept weak suppression, disconnected journeys, incomplete attribution, or inconsistent customer recognition as limitations of the channel rather than problems a better identity foundation could address.
This is not simply a buyer-education problem. Identity resolution sits at the intersection of advertising technology, customer data, analytics, privacy, and marketing operations. The market is highly technical and fragmented, and most providers define the category through the part of the problem their product solves.
A visitor-identification company may focus on anonymous traffic. A CDP may focus on unified profiles. An enrichment provider may focus on appending customer or household data. A B2B platform may focus on identifying companies and connecting people to accounts. Each definition can be valid. None is complete.
The Marketing Question
Identity Is Marketing’s Answer to “Who?”
Marketing has always had two halves.
The art is what you say, how you say it, and what emotion you reach for. The art is where a Volkswagen ad can put a tiny Beetle in a sea of white space under the words “Think Small” and change how a market thinks about a product.
The science is who you say it to, when, on what channel, at what cost, and whether any of it worked. The science is where the old advertising maxim often attributed to John Wanamaker lives: “Half the money I spend on advertising is wasted; the trouble is I don’t know which half.”
Identity sits at the center of that science.
Who are you trying to reach? Who saw the ad? Who visited the site? Is this a new visitor or someone who came last week? Is this person already a customer? Are these three CRM records the same person? Do these five people belong to the same target account? Did the person who clicked the email later buy through another channel?
Without a credible answer to “who,” targeting gets weaker, measurement becomes fragmented, and personalization rests on partial information.
Identity resolution is marketing’s answer to “who?”
In the simplest possible terms, identity resolution is the work of figuring out who you are dealing with.
In the physical world, recognition happens naturally. You walk into your neighborhood Starbucks. The barista sees you, recognizes you, knows your name, and starts making your drink before you finish walking to the counter. They did not run a query. They did not ask for your email. They recognized you, and the interaction became faster, warmer, and more personal as a result.
Online, the principle is similar. The mechanics are harder because there is no face to recognize. Instead, marketers work with signals: emails, cookies, devices, transactions, CRM records, IP addresses, ad clicks, logins, mobile IDs, phone numbers, and offline records. The goal is to understand which signals belong to the same person, household, company, or customer journey.
When that works, the experience starts to feel more coherent. When it fails, people get irrelevant ads, see promotions for products they bought yesterday, fill out the same form three times, or get treated as strangers by companies they have spent thousands of dollars with.
Working Definition
A Practical Definition of Identity Resolution
Identity resolution is the process of determining when different identifiers, devices, sessions, records, and interactions belong to the same person, household, or company.
The result is recognition.
“Visitor recognition” is a useful phrase when the use case involves a website visitor, but it is too narrow to replace identity resolution as the category term. Identity resolution also happens across customer databases, households, business accounts, mobile devices, advertising platforms, retail environments, and offline records.
But before choosing an identity solution, the business problem needs to be clear.
A Better Starting Point
Start With the Growth Motion
The right identity capability depends on where the business needs leverage and what the marketer is trying to make possible across Attract, Convert, Retain, and Measure.
At a broad level, identity-resolution use cases show up across four growth motions:
Attract
Define and reach the right audience before it enters the funnel.
Convert
Recognize and reengage people or accounts already in the funnel.
Retain
Maintain customer continuity and improve lifecycle intelligence.
Measure
Connect spend, behavior, identity, and outcomes across the journey.
Attract: bring the right people, households, or accounts into an observable environment.
Convert: maximize conversion of the people or accounts that have entered the funnel.
Retain: maximize revenue, engagement, renewal, or repeat purchase from already converted customers.
Measure: connect spend, identity, behavior, and outcomes well enough to make better decisions.
In this framework, Attract does not mean acquiring a customer. It means acquiring attention, traffic, visits, app users, store visitors, or target-account engagement at the top of the funnel.
Once a person, household, or account has shown up, the identity problem shifts from attraction to conversion. Retargeting, abandonment recovery, lead recovery, and sales follow-up are usually conversion problems, not traffic-acquisition problems. They act on people or accounts that have already entered the funnel.
That distinction matters because marketers often talk about identity resolution as if it were one capability. It is not. The identity problem in Attract is different from the identity problem in Convert. Retention has a different identity need than measurement. A CPG brand building audiences for retail media does not have the same identity problem as a Shopify brand trying to connect email clicks, sessions, carts, and purchases. A B2B SaaS company trying to recognize target-account engagement does not have the same problem as a high-traffic lead-generation business trying to recover anonymous visitors who abandoned a quote form.
“Which identity solution should we buy?”
“For this business model and this growth motion, what identity problem are we trying to solve?”
Once the growth opportunity and identity problem are clear, the type of capability required becomes much easier to identify.
| Growth motion | Core business question | Identity question | What identity must do |
|---|---|---|---|
| Attract | How do we get the right people, households, or accounts to show up? | Can we define, find, expand, or activate the right audience before they enter the funnel? | Improve audience definition, expand reach, make known audiences more activatable, or use external graphs to find the right people or accounts. |
| Convert | How do we get more of the acquired audience to take the desired action? | Can we recognize visitors, leads, carts, form starts, accounts, or store visitors and reengage them appropriately? | Connect fragmented first-party signals, recognize returning visitors, recover abandoners, or use external recognition when the brand cannot identify the visitor or account itself. |
| Retain | How do we increase value from people or accounts that have already converted? | Can we recognize customers across channels, understand lifecycle stage, and improve segmentation? | Connect customer behavior over time, maintain continuity across channels, enrich customer or account records, and trigger better lifecycle actions. |
| Measure | What worked, what did not, and where should we invest next? | Can we connect identity, spend, touchpoints, behavior, and outcomes credibly? | Resolve journeys, deduplicate people or accounts, connect touchpoints to outcomes, improve platform signals, and make performance metrics more accurate. |
A simple shorthand: Attract is pre-funnel audience creation. Convert is post-arrival recognition. Retain is post-customer intelligence. Measure is cross-funnel identity continuity.
This is where the definition problem starts to resolve. Across these growth motions, the identity work keeps falling into three recurring patterns.
Sometimes the brand already has the signals, but those signals are fragmented across sessions, devices, platforms, channels, and systems. The work is to connect what the brand should already be able to know.
Sometimes the brand cannot recognize the person, household, device, or account on its own. The work is to use an external graph or platform to make a connection the brand cannot independently make.
Sometimes the brand already has a person, household, company, account, or customer record. The work is to make that known identity cleaner, richer, more complete, more reachable, or more useful.
The Framework
The Three Modes of Identity Resolution
Those three patterns are the three modes of identity resolution:
Direct Signals Exist
Connect first-party interactions into a resolved journey.
The brand should be able to know.Recognition Is Missing
Use an external graph to identify what the brand cannot.
The bridge comes from outside.A Known Record Exists
Clean, complete, connect, or expand an existing identity.
The identity becomes more useful.Stitch: Recognize people and reconstruct journeys using signals created through your own customer relationships.
Borrow: Use an external provider’s identity graph to recognize people, households, devices, or companies you cannot identify independently.
Enrich: Clean, complete, connect, and expand records for customers or prospects you already know.
These modes can work together, but they solve different problems. The starting point should be the growth opportunity and identity problem—not the product category a provider happens to sell.
Mode 1
Stitch
Recognize people using signals created through your own customer relationships
Stitch is first-party identity resolution. It recognizes visitors, customers, users, leads, accounts, and journeys using signals generated through a brand’s own direct interactions.
Those signals may include browser or device identifiers, email- and SMS-click identifiers, logins, customer IDs, form submissions, phone numbers, purchases, subscription IDs, CRM records, ad click IDs, mobile-app identifiers, call-center interactions, offline transactions, and server-side events.
This is not static database cleanup. It happens continuously as people click, browse, identify themselves, change devices, return through different channels, submit forms, log in, purchase, subscribe, call, or engage offline.
Stitch matters most when the brand already has useful signals but those signals are fragmented. That fragmentation usually shows up in Convert, Retain, and Measure.
Consider a shopper who clicks a promotional email on her phone, views three products, leaves, returns two days later through a Google ad on a laptop, and eventually purchases after seeing a Meta ad. Most marketing systems will record several fragments: an email click in the email platform, an anonymous mobile session in analytics, a desktop search visit, and a purchase that Meta, Google, and the email provider may each try to claim.
A strong first-party identity layer attempts to stitch those fragments into one journey using evidence the brand collected directly. The first page view may technically begin as anonymous, but if the email link contains an identifier that connects the visit to the recipient, the brand can recognize the person using its own signal. The same can happen through an SMS link, login, form submission, checkout, loyalty interaction, subscription, or later purchase.
This distinction is central:
Stitch can recognize an anonymous visitor. It does so using evidence generated through the brand’s own relationship with that person.
Stitch can include both deterministic and probabilistic methods. “First-party” describes the source of the evidence; it does not mean every connection must be deterministic. A strong probabilistic model can sometimes connect a seemingly anonymous interaction to a known profile with high confidence while still relying on data the brand collected through its own interactions.
Many companies overestimate how well they do this. They assume that having a data warehouse, CDP, CRM, email platform, GA4 implementation, customer table, or server-side tracking means identity resolution is already in place.
Those systems can support identity resolution. Their presence does not prove that identities and journeys are actually being resolved.
A more useful test is operational:
- Do identifiable email and SMS clicks resolve to the correct customer journey in your analytics system, CRM, or data warehouse?
- When someone later purchases or submits a form, is their earlier anonymous activity attached to the resulting profile?
- Can the company reliably distinguish a new visitor from a returning visitor and a new customer from an existing customer?
- Can recent purchasers be suppressed consistently across paid media and owned channels?
- Can the business reconstruct a credible journey across paid media, email, SMS, organic, direct, CRM, and transaction systems?
In my experience, few brands do all of this well, regardless of how much they have invested in CDPs, warehouses, and integrations.
When implemented properly, Stitch can improve attribution, journey analysis, triggered email and SMS, retargeting, paid-media suppression, new-versus-returning customer measurement, personalization, incrementality analysis, and AI-driven decisioning.
Stitch also powers better signal quality for platforms like Meta and Google. Meta’s Conversions API sends event data directly from a business’s server to Meta, and customer information parameters can improve event matching. Google Enhanced Conversions uses hashed first-party customer data to improve conversion measurement by matching conversions to signed-in Google accounts.
Meta and Google are two of the largest identity graphs in the world. But as privacy regulations and browser restrictions make third-party signals harder to rely on, these platforms increasingly depend on the quality of the first-party signals brands send them.
Mode 2
Borrow
Use an external identity graph to recognize what you cannot recognize independently
Borrow is recognition too, but the bridge comes from outside the brand.
In Stitch, the brand recognizes someone using evidence generated through its own interactions. In Borrow, the brand asks an external graph, platform, or provider to make a connection it cannot make from its own data alone.
Borrow often shows up in Attract and Convert. It can help a brand find or reach the right audience before the funnel, and it can help recognize or reengage people, households, devices, or accounts that have entered the funnel but remain unknown to the brand.
The word Borrow is intentional. The recognition capability belongs to the provider, not the brand. It may be valid only for a session, a campaign, a destination, a platform, or as long as the vendor’s graph continues to support the match. The same graph and identity relationships are often made available to many customers. The brand is gaining access to someone else’s recognition capability rather than building that relationship entirely from its own signals.
Borrow therefore carries different questions about provenance, control, portability, permitted use, accuracy, durability, and privacy compliance.
An external provider may rely on some combination of hashed emails, publisher or login networks, device identifiers, mobile-app data, IP addresses, household graphs, advertising identifiers, proprietary platform activity, partner data, or licensed datasets. The output may identify a person, browser, device, household, company, or provider-specific token.
Those are not interchangeable.
For a consumer marketer, a provider may associate a website visit with an email, phone number, postal address, household, device, or proprietary identity. Depending on the provider, permissions, and activation rules, that recognition may support email retargeting, display media, CTV, direct mail, suppression, lead recovery, or measurement.
This can be particularly valuable for high-traffic lead-generation businesses. Suppose a company pays to drive 100,000 visitors to a quote form, application, quiz, calculator, or advertorial funnel, and only 3,000 submit their information. Stitch can reconstruct the journeys of visitors who eventually identify themselves. It cannot necessarily tell the company who the remaining 97,000 people are. Borrow may allow an external graph to recognize a portion of that otherwise unknown demand.
The B2B version is different. A provider may use IP, network, device, publisher, or proprietary account data to infer which company is visiting a website. For a low-traffic enterprise business, recognizing that several visits came from a target account may be more valuable than obtaining a handful of person-level identities. That signal can support account scoring, sales alerts, advertising, buying-group analysis, and outbound prioritization.
However, knowing the company is not the same as knowing the person. A match to Acme Corporation does not establish that the visitor is Jane Smith, VP of Procurement. Account-level and person-level recognition should be evaluated separately.
Borrowed recognition is frequently less durable than first-party recognition. The match may depend on a browser, IP address, device, campaign window, publisher network, platform environment, or provider-specific identifier. Some connections persist. Others decay quickly or disappear when the context changes.
That does not make Borrow inherently weak. It means durability must match the use case. An identity that remains useful for a 14-day retargeting campaign may be sufficient even if it cannot support a five-year customer history. Conversely, a temporary household-level signal may be unsuitable for personalized lifecycle communication.
This is why match rate is not enough.
A 40% match rate could mean 40% of browsers, devices, households, companies, people, or simply traffic for which the provider returned some identifier. It does not tell the marketer whether the match was accurate, incremental, usable, compliant, or profitable.
Value of Borrow =
incremental recognition × accuracy × activation × economics
This is not a literal accounting formula. It is a way to expose weak propositions. If recognition is not incremental, the brand is paying for identities it already had. If accuracy is poor, the provider creates waste and risk. If the result cannot be activated, it creates a report rather than a marketing capability. If activation costs more than the value it produces, the economics fail.
Match rate is an input. It is not the outcome.
Mode 3
Enrich
Improve and expand records for identities you already possess
Enrich begins where Borrow does not.
Borrow starts with an interaction the brand cannot recognize and asks an external graph, “Who is this?”
Enrich starts with a person, household, company, account, or customer record the brand already has and asks, “What else can we reliably learn, connect, correct, or activate?”
Enrich often shows up in Attract and Retain. It can help a brand build better audiences before the funnel, and it can help improve segmentation, personalization, householding, account mapping, and activation after a person or account is already known.
The company may already have customer emails, phone numbers, postal addresses, CRM contacts, loyalty records, product registrations, subscriber profiles, donor records, business accounts, or transaction histories. Those records may be incomplete, duplicated, outdated, inconsistent, or difficult to activate.
An enrichment provider may help by deduplicating identities, standardizing names and addresses, matching personal and work emails, appending phone numbers or postal addresses, connecting people into households, connecting contacts to companies, adding demographic or firmographic attributes, supporting audience onboarding and clean-room matching, or creating modeled audience extensions.
This is often called offline identity resolution, although much of the source data, processing, delivery, and activation is now digital.
The CPG use case makes the distinction clear. A CPG brand may have millions of consumer records from loyalty programs, coupons, sweepstakes, product registrations, recipe subscriptions, promotions, or customer support. Its primary problem may not be recognizing an anonymous website visitor. It may need to clean and deduplicate the database, connect people into households, append useful attributes, match consumers into retail-media networks, activate audiences in Meta or CTV, build suppression lists, measure overlap with retailer transaction data, or create modeled audiences that resemble high-value customers.
That is identity resolution and enrichment, but it is not the same problem as connecting one shopper’s email click to a later purchase.
One distinction is worth keeping clear:
Lookalike modeling is not identity resolution in the strict sense.
Identity resolution determines whether multiple signals or records represent the same entity. Lookalike modeling identifies additional entities that resemble a source audience. It is often enabled by identity data, but it is a downstream audience-expansion capability.
Decision Guide
Stitch, Borrow, and Enrich at a Glance
| Mode | Starting condition | Source of the identity bridge | Typical outcome | Common use cases |
|---|---|---|---|---|
| Stitch | The brand has direct signals but has not connected them reliably | First-party customer and prospect interactions | A resolved visitor, customer, user, lead, or account journey | Attribution, lifecycle marketing, conversion APIs, suppression, personalization |
| Borrow | The brand cannot recognize the interaction independently | An external person, device, household, or account graph | Recognition beyond the brand’s own signals | Anonymous visitor recovery, B2B account recognition, CTV, direct mail, external retargeting |
| Enrich | The brand already has a person, household, company, account, or customer record | Existing records combined with external enrichment data | A cleaner, richer, more activatable known audience | CRM hygiene, householding, data append, retail-media matching, audience modeling |
Borrow begins with an interaction you cannot recognize. Enrich begins with an identity you already have.
Sometimes Borrow and Enrich appear together. A provider may recognize a visitor and also return attributes such as interests, household data, company information, or audience segments. That does not erase the distinction. It means the provider is performing more than one identity function.
Most mature marketing organizations will eventually use more than one mode. The mistake is assuming they need all three for the same reason, from the same provider, or at the same time.
Success Criteria
Measure What Changes in the Business
Identity resolution should not be measured by the mechanism alone. Match rate, IDs created, fields appended, and audience size are operating metrics, not business outcomes.
The scorecard should match the growth motion being improved.
| Growth motion | What identity should improve | Primary modes | Example metrics |
|---|---|---|---|
| Attract | Bring the right people, households, or accounts into the funnel. Identity should improve audience quality, reach, relevance, and cost efficiency before the first visit or interaction. | Mostly Borrow and Enrich | Qualified visit rate, target-account engagement, audience match rate, incremental reach, cost per qualified visit/account/store visit, downstream conversion quality |
| Convert | Maximize conversion of people or accounts that have already entered the funnel. Identity should help recognize, reengage, and recover visitors, carts, forms, leads, accounts, or store visitors. | Mostly Stitch and Borrow | Resolved journey rate, known-ID rate at each funnel step, abandonment recovery, reengagement lift, conversion-rate lift, recovered revenue or qualified leads |
| Retain | Increase value from customers, users, members, subscribers, or accounts that have already converted. Identity should improve recognition, lifecycle segmentation, personalization, suppression, renewal, expansion, and winback. | Mostly Stitch and Enrich | Returning-customer recognition, lifecycle-segment lift, repeat purchase, renewal rate, expansion rate, winback lift, churn reduction, suppression accuracy |
| Measure | Connect spend, touchpoints, behavior, identity, and outcomes well enough to make better decisions. Identity should improve the credibility of attribution, incrementality, platform signals, and business reporting. | Mostly Stitch, supported by Borrow and Enrich where relevant | New vs returning clarity, touchpoint-to-outcome continuity, platform signal quality, deduplication accuracy, incrementality, CAC, ROAS, LTV, pipeline, revenue, or margin lift |
The key is to measure the business motion, not just the identity mechanism.
The commercial question should always be:What changed in Attract, Convert, Retain, or Measure because identity was resolved?
References
Source Notes
- This paper uses “first-party” to refer to signals created through direct brand interactions with customers and prospects.
- Meta for Developers, “Conversions API.”
- Google Ads Help, “About enhanced conversions.”
- IAB Tech Lab, “Identity Solutions Guidance.”
- MDN Web Docs, “Third-party cookies.”
- Smithsonian National Museum of American History, Volkswagen “Think Small” advertising material.
- Quote Investigator, research on the “half the money” advertising maxim. The familiar attribution to John Wanamaker remains uncertain.