When to Split, Merge, or Retire an ICP

Who this framework is for

This framework is for teams managing several ICPs across products, applications, relationship types, or regions. It helps when one definition has accumulated exceptions, two segments behave identically, or an older ICP remains active after strategy or market conditions changed. It governs account definitions, not the detailed people map inside each account.

Contents

1 Define an ICP lifecycle event 2 Diagnose the reason for change 3 Apply the split tests 4 Apply the merge tests 5 Apply the retirement tests 6 Preserve history and migrate accounts 7 Use the lifecycle decision record 8 Illustrative example 9 Common mistakes 10 What this does not mean 11 Common questions 12 Sources and methodology

1. Define an ICP lifecycle event

An ICP lifecycle event changes the definition that determines which accounts are researched, eligible, prioritized, or excluded. It is more consequential than correcting a typo or refreshing one account's evidence. Use four states: State Meaning New account use Active Approved for current research and commercial use Allowed Under review Evidence suggests a material change may be needed Controlled use only Superseded Replaced by one or more newer versions Not allowed for new decisions Retired Core hypothesis is no longer approved Not allowed Never delete the old definition from historical records. A report created under version 1.2 should remain attributable to version 1.2 after version 2.0 takes effect. Classification systems themselves evolve. The Census Bureau provides current and historical NAICS reference files and information about revisions. U.S. Census Bureau NAICS This does not mean every NAICS change requires a new ICP, but classification revisions can break a universe query or move activities between codes.

2. Diagnose the reason for change

Classify the proposed change before selecting an action:

• Offer change: product, configuration, service, price corridor, or delivery model changed.

• Application discovery: two use cases previously combined require different operating conditions.

• Relationship discovery: OEM, distributor, end-user, or contractor economics and roles diverge.

• Access change: qualification, partner, geography, service, or procurement conditions changed.

• Evidence change: previous criteria were unobservable, ambiguous, or systematically wrong.

• Population change: entry, exit, ownership, facility, or structural market patterns changed.

• Classification change: source codes, definitions, or versions changed.

• Governance failure: duplicate ICPs or uncontrolled local variants appeared. Statistics Canada's notice for NAICS Canada 2022 identifies real changes including new groupings, splits, and transfers, as well as virtual changes to codes, titles, or definitions without scope change. Statistics Canada NAICS 2022 release notice This is a useful distinction: a label change is not automatically a commercial segment change. Business populations also change. Census Business Dynamics Statistics reports annual measures such as establishment births and deaths and firm startups and shutdowns at aggregated levels. Census Business Dynamics Statistics These data provide context; they do not identify which named account now meets an ICP.

3. Apply the split tests

Split an ICP only when a proposed child segment passes all five tests: 1 Distinct decision rule: at least one material inclusion, exclusion, or gate differs. 2 Distinct evidence rule: researchers need different sources, units, or verification. 3 Distinct action: the commercial motion, proposition, service, partner route, or review cadence differs. 4 Reproducible classification: independent reviewers can assign accounts consistently. 5 Operational value: the split improves a decision enough to justify added complexity. Use the split worksheet: Dimension Current ICP Proposed child A Proposed child B Does decision change? Application [Rule] [Rule] [Rule] Yes/No Relationship [Rule] [Rule] [Rule] Yes/No Gates [Rule] [Rule] [Rule] Yes/No Evidence [Source/unit] [Source/unit] [Source/unit] Yes/No Service [Model] [Model] [Model] Yes/No Action [Motion] [Motion] [Motion] Yes/No A performance difference alone is insufficient. First test whether exposure, data quality, treatment, timing, or account mix explains it.

4. Apply the merge tests

Merge two ICPs when their separation no longer changes material decisions. Require:

• compatible account units and application definitions;

• the same relationship type or a justified common motion;

• substantially aligned gates and exclusions;

• the same evidence states and minimum proof;

• compatible service and access model;

• no material loss of reporting or ownership clarity. Run a blind reclassification on a mixed account sample. If reviewers cannot reproduce the boundary, and the supposed segments receive the same action, the distinction may be decorative. Do not merge merely to simplify a dashboard. A readable interface can group several distinct ICPs while preserving their definitions underneath.

5. Apply the retirement tests

Retire an ICP when one or more conditions are confirmed:

• the target application is incompatible with the offer;

• the commercial relationship is no longer supported;

• the segment cannot meet a mandatory access or service condition;

• repeated validation falsifies the core hypothesis;

• the offer or strategy has been withdrawn;

• another version fully supersedes the decision scope;

• required evidence cannot be collected responsibly and no safe proxy exists. Retirement is not a claim that every account is permanently unattractive. It means the current definition is no longer authorized for new decisions. Use "under review" when evidence is incomplete. GAO's data-reliability guide frames reliability in relation to accuracy, completeness, and applicability for the intended use. GAO, Assessing Data Reliability A source gap should become an evidence problem before it becomes a commercial conclusion.

6. Preserve history and migrate accounts

Every approved lifecycle event needs:

• old and new IDs and versions;

• effective date;

• owner and approver role;

• reason and evidence;

• mapping of criteria and states;

• treatment of active research and outreach;

• account migration rules;

• reporting restatement policy;

• rollback or review condition. Do not silently rescore historical accounts. Keep the prior decision and optionally add a current score under the new version. This separates "what the team knew then" from "how the account is assessed now." For named public companies, EDGAR can help verify ownership, segment, filing, and material-event evidence. SEC Search Filings Preserve filing date and scope. A parent-level disclosure may not resolve facility-level migration. For accounts in active contact, determine whether the change means continue, pause, suppress, hand off, or close. Avoid forcing a newly excluded account through the rest of an old sequence.

7. Use the lifecycle decision record

Complete these fields: Field Required entry Proposal Retain, split, merge, or retire Scope ICP IDs, versions, products, regions, relationship types Trigger Dated evidence or governance event Alternative explanations Data, exposure, timing, or classification effects Decision-rule difference What changes for eligibility or action Validation Sample, reviewer agreement, and unresolved states Migration Accounts, tasks, reporting, and active motions Effective date Start of new authorized use Review trigger Date or evidence threshold Owner Accountable role Approve the smallest lifecycle change that explains the evidence. A split that produces six thin categories may be worse than one ICP with an application subtype field.

8. Illustrative example

Illustrative example - fictional. A fictional exporter has one "regulated process plants" ICP. A review samples 80 fictional facilities: 40 food-processing and 40 pharmaceutical. Evidence result Food facilities Pharmaceutical facilities Confirmed application fit 30 / 40 = 75% 18 / 40 = 45% Service model pass 28 / 40 = 70% 20 / 40 = 50% Segment-specific documentation gate pass 34 / 40 = 85% 12 / 40 = 30% Unknown on critical evidence 3 / 40 = 7.5% 11 / 40 = 27.5% The pharmaceutical cell requires a different documentation gate, specialist evidence, service process, and review owner. The food cell does not. Reviewer agreement on the two application definitions is 35 of 40 independently checked accounts: 35 / 40 = 87.5%. The team chooses Split and retest: Food process retrofit ICP v1.0 and Pharmaceutical qualified-system ICP v1.0. It does not claim that the observed percentages estimate all facilities, because the sample was a fictional bounded review. Historical records retain the old ICP version, and unresolved pharmaceutical accounts remain on hold pending evidence.

9. Common mistakes

• Splitting on one campaign result: exposure or message differences are ignored.

• Creating a segment for every exception: taxonomy becomes unmanageable.

• Merging for dashboard simplicity: meaningful gates and service differences disappear.

• Retiring because data are missing: unknown becomes false non-fit.

• Overwriting history: old reports silently adopt today's definition.

• Migrating parents without facilities: site-level applications and ownership remain wrong.

• Changing codes without concordance: universe counts jump because classifications changed.

• Leaving active motions untouched: contacts continue under a retired hypothesis.

10. What this does not mean

Lifecycle governance does not freeze ICPs permanently. It makes change traceable and decision-relevant. A split is not evidence that one child segment is superior, and a merge is not permission to use one generic message. Retirement is not deletion. None of these actions identifies the buying committee or guarantees commercial performance.

11. Common questions

How much evidence is enough to split? Enough to show a reproducible difference in decision rules and actions, with material alternatives examined. Set the threshold before reviewing outcomes and match it to the cost of added complexity. Can one account belong to two ICPs? Yes, when different facilities, applications, products, or relationship types genuinely support separate motions. Store the assignment at the correct unit and prevent duplicate, conflicting contact. Should a renamed NAICS code create a new ICP version? Only if it changes the query, scope, evidence interpretation, or included population. Record even a technical mapping change when it affects reproducibility. What happens to reports after a merge? Keep historical results under the original IDs. Future reporting may roll them into the merged ICP while retaining a bridge table; do not restate past data without an explicit policy. Who should own retirement? The accountable business owner should approve it with research, operations, and affected commercial roles. The owner must also ensure new use stops and active motions are handled.

12. Sources and methodology

Official sources describe NAICS revisions, business-dynamics measures, public filing access, and data-reliability principles. SellersGPT created the lifecycle states, split/merge/retire tests, worksheet, and migration controls as editorial governance tools. The fictional example is not a customer result or population estimate. 1 U.S. Census Bureau, North American Industry Classification System. https://www.census.gov/naics/. Accessed August 24, 2026. 2 Statistics Canada, Notice of Release of NAICS Canada 2022 Version 1.0. https://www.statcan.gc.ca/en/subjects/standard/naics/2022/v1/notice. Accessed August 24, 2026. 3 U.S. Census Bureau, Business Dynamics Statistics. https://www.census.gov/programs-surveys/bds.html. Accessed August 24, 2026. 4 U.S. Government Accountability Office, Assessing Data Reliability (GAO-20-283G). https://www.gao.gov/products/gao-20-283g. Accessed August 24, 2026. 5 U.S. Securities and Exchange Commission, Search Filings. https://www.sec.gov/search-filings. Accessed August 24, 2026.