Design Controlled Vocabularies That Can Accept New Evidence
Keep controlled CRM vocabularies accurate with unknown and other states, proposed-value review, versioning and practical exception tests.
10/23/20263 min read
A controlled vocabulary can improve consistency while also forcing bad data into neat categories. When a researcher encounters evidence that does not fit the picklist, they may choose the least wrong option, type a workaround elsewhere or stop recording the distinction. The database looks standardized but its meaning becomes unreliable.
Design an exception route alongside the vocabulary. Allow a record to remain unresolved or carry a proposed value until an accountable reviewer decides whether the vocabulary should change. Consistency should make evidence comparable, not require it to fit an outdated model.
Define categories with inclusion and exclusion rules
Each category needs a plain-language definition, a matching example, a near-miss and an owner. A label such as “service company” is too broad unless the team knows which business activities qualify and why the distinction matters.
Keep unknown separate from other. Unknown means the evidence does not establish the classification. Other means the evidence is understood but falls outside defined categories. Neither should automatically mean rejected, low quality or not worth researching.
Add a proposed-value workflow
A proposal should contain the source record, observed evidence, existing categories considered, reason they do not fit, proposed label and decision affected. Require a useful distinction, not a new category for every unusual phrase on a website.
The reviewer can map the evidence to an existing value, clarify a definition, add a new value, keep the case unresolved or decide the field is unnecessary. Record the reasoning and any migration implications. Do not silently change the meaning of an existing label while leaving older records untouched.
Work through a fictional classification
Fictional Larkspur Research classifies potential service partners as referral, reseller or implementation. A company describes itself as an independent evaluator that does not sell, refer or implement the product. The researcher understands the role but cannot fit it honestly into the existing list.
The record receives “other, review pending,” with a proposal for “independent evaluator.” The proposal explains that the distinction affects commercial expectations and prohibits assuming a referral arrangement. The reviewer checks whether the team actually needs to track this relationship type.
If the role is relevant, the vocabulary gains a defined evaluator category with examples and exclusions. If it is outside the team's purpose, the record can be excluded from the partner workflow without pretending it is a reseller. This is a fictional governance decision, not an account finding.
Control changes without freezing the model
Assign a vocabulary version and an effective date for material changes. Explain whether a new label changes presentation only or changes classification meaning. Identify existing records that may require review, but do not bulk-reclassify them without sufficient evidence.
Keep an alias map for retired wording where needed for search or import. An alias should not hide a genuine semantic change. If two old categories merge, document what distinction is being lost and whether another field must preserve it for a relevant decision.
Test the exception path in the interface
A researcher should be able to choose unresolved, record a short reason and submit a proposal without entering an inaccurate value first. Test whether required-field rules or integrations defeat that path. Confirm product-specific enumeration and import behavior in the actual CRM before implementation.
Review a small synthetic set containing an ordinary match, a near-miss, insufficient evidence and a genuinely new category. Ask two reviewers to explain their decisions. Disagreement may reveal unclear definitions rather than careless users.
The finished vocabulary includes both its current categories and a fair way to challenge them. Monitor unresolved cases for recurring patterns, not as a target to eliminate at any cost. A picklist that can evolve through evidence is more reliable than one that achieves perfect completion by forcing every new fact into the wrong box.
Research access and CRM configuration are separate purchases. Leads Automation offers an independent US$15/month service and is not affiliated with LinkedIn. Confirm the actual plan, eligibility and terms; do not assume schema design, imports or integrations are included. Discuss the offer on WhatsApp.
Sources and scope
This is a proposed operating method, not a vendor-prescribed workflow. Verify product capabilities in your actual subscription before implementation.
NIST Privacy Framework: A Tool for Improving Privacy through Enterprise Risk Management, Version 1.0