B2B Cross-Device Identity Resolution: Workflow and Examples

B2B teams should use cross-device identity resolution when linking records across devices will improve a specific, permitted decision. Start with evidence you can explain, distinguish person links from company links and leave uncertain cases unresolved. More linked records do not automatically mean better data.

A buyer might use a phone and a work laptop during research. Connecting those sessions can help analysis, but a shared office network can also connect unrelated people. Define the entity and the decision before merging records or changing a workflow.

What does cross-device identity resolution mean?

It is the process of determining whether activity or identifiers from different devices refer to the same entity. That entity might be a person, an account or a household. Those are different conclusions and should remain distinct in the data model.

The FTC’s January 2017 cross-device tracking report describes deterministic and probabilistic approaches. The report is a historical explanation of the technology and its privacy concerns, not a current product-performance benchmark.

• Deterministic: A direct identifier, such as the same authenticated account, supports a link.
• Probabilistic: A model infers a likely association from available evidence.
• Unresolved: Evidence does not support the needed conclusion.

Even a direct identifier needs context. An account can be shared, employment can change and a system can map an identifier incorrectly. Keep the source and reason available for review.

First ask what the link will change

A reporting task may need company-level continuity. A contact-specific follow-up needs evidence about the person and the appropriate channel. Account access or entitlements require authenticated identity controls rather than an inferred marketing association.

Write down the action, minimum evidence, conflicting fields and correction process. If company-level information answers the business question, do not turn it into a stronger person-level claim.

Example: the shared office problem

This is an original hypothetical example. Two employees at the same company use the same office network. One researches payroll software; the other visits a general business article. A network association may support company context, but it does not establish that both sessions belong to one person.

Keep the company association separate from person identity. Do not append the first employee’s research history to the second employee’s CRM record based only on that network evidence.

Example: a known account on two devices

A user signs in to the same business account on a phone and laptop. Within the appropriate first-party context, the account identifier can support continuity between those authenticated sessions.

It still does not reveal every device the person uses, every offsite action or another employee’s behavior. Use the link within the documented purpose and keep shared-account or reassignment exceptions visible.

A practical review workflow

1. Inventory sources. Record the identifiers, collection context and permitted purpose.
2. Define entity levels. Separate device, person and company records.
3. Propose links. Keep the evidence and date rather than silently merging everything.
4. Check conflicts. Review shared identifiers, differing employment and other contradictory context.
5. Test accepted links. Use known examples and reviewed exceptions.
6. Allow abstention. Hold uncertain cases instead of requiring an answer.
7. Correct and expire. Propagate corrections and remove links that no longer have support.

These are team design steps, not a claim that every software product implements them. Test the exact system and output you plan to use.

Measure link quality separately from coverage

A system that returns more matches may also make more false associations. Review confirmed links and unresolved cases, not just the headline match rate.

This is an original hypothetical calculation. A review contains 100 candidate pairs. The system accepts 40 links, and a labeled review confirms 36 of those. Accepted-link precision is 36 / 40, or 90%, for that reviewed set.

If the set contains 60 true links in total, recall is 36 / 60, or 60%. Without that known total, you cannot calculate recall. Neither number establishes performance for an untested traffic source or region.

Also measure false merges, missed links, unresolved cases and correction time. Compare the same sample and definition before changing rules. A threshold that is useful for company reporting may be insufficient for a person-specific action.

Where BrandWell products fit

MarketPulse provides topic-research audiences and ICP filtering. TrafficID helps surface website audiences. They provide different business context; neither should be presented here as an unrestricted general-purpose cross-device identity graph.

In MarketPulse, search available topics, apply ICP filters and review matching people and companies. A pull retrieves available data from the previous 24 hours for one topic. Counts, coverage and contact fields vary, and audiences can overlap.

Growth and Scale include AIMEE for the initial user to research and prepare next steps through configured tools. Keep the source context in the task and distinguish observed facts from proposed associations.

Current MarketPulse standard plans

• Starter: One topic, weekly refreshes and 25,000 monthly contact exports.
• Growth: Three topics, daily refreshes and 100,000 monthly contact exports.
• Scale: Ten topics, daily refreshes and unlimited contact exports.

Daily use across all included topics provides 90 Growth pulls or 300 Scale pulls in a 30-day month. Weekly Starter pulls vary by calendar month. Refreshes stop at the plan limit without overage charges. Connected products, channel usage and fulfillment may cost extra; eligible early-access discounts are separate time-limited offers.

Permission does not transfer through a match

A link between records does not create permission for every destination or channel. Keep collection context, permitted use and suppression decisions separate from the matching decision. An enriched mobile number is not an SMS opt-in.

For client work, restrict access to the appropriate client records and keep sources explainable. A branded report should not conceal an uncertain association or turn company evidence into a named person’s browsing history.

Frequently asked questions

Does the same IP address prove the same person?

No. Shared networks are a reason to distinguish company or network context from person identity.

Is a high match rate enough?

No. Evaluate correctness, unresolved cases and the decisions enabled. More matches can be less useful if they introduce false merges.

Should every uncertain match go to manual review?

Prioritize review based on the value and consequence of the action. Some cases can remain unresolved or be used only at a less specific entity level.

Does resolution prove marketing impact?

No. Record quality, workflow usefulness and business outcomes are separate measures. A linked journey alone does not establish that a campaign caused a purchase.

See MarketPulse’s current demo and plans to evaluate research audiences while keeping identity evidence and activation decisions clear.

Reviewed and updated October 3, 2026.

MarketPulse plans are starting at $500/mo. Review current package scope, usage and any separate product or service costs before purchase.

Written by

Farnaz Kia