Unknown Is a Useful Research Result

Define what an unknown result means

Use unknown to mean that the available evidence does not justify a supported value for the requested field at the required scope and time. That definition leaves room for useful partial observations. A page can identify a management team without establishing the current head of operations. A storefront can show products without establishing the enterprise software plan. The researcher may have learned something real even though the final attribute remains unresolved.

An unknown is not automatically a research failure. Public company information is selective, uneven, and sometimes internally inconsistent. Your specification may ask a question that the chosen sources cannot reliably answer. If the process has been completed and documented, an abstention can protect the buyer from a false decision. The buyer can then change the criterion, supply better input, or accept the uncertainty.

Make the distinction visible in the interface. Displaying an empty cell with no explanation forces users to guess whether the row is unfinished or deliberately unresolved. A short status and a concise note provide far more information. The goal is not to attach a lengthy essay to every blank, but to preserve enough context for the next person to make the correct choice.

Use a small vocabulary with non-overlapping states

Choose status labels according to causes, not researcher mood. 'Low confidence' alone does not explain whether the company identity is unclear, the source cannot be reached, or two pages disagree. A useful vocabulary gives each state a definition, a permitted next action, and an example. Avoid dozens of nearly synonymous labels that nobody can apply consistently.

The Government Data Quality Framework's guidance recommends communicating missing data and its reasons, including cases where a field is intentionally inapplicable. The taxonomy below is our practical implementation for company research. It is not a universal standard, and teams should adapt the labels to their schema while preserving the distinctions that affect decisions.

Allow a primary state and a secondary note when causes overlap. If a supplied company name matches two possible businesses and one site is inaccessible, the primary blocker is identity ambiguity. Retrying the inaccessible site will not resolve the company match by itself. Choosing the most consequential blocker makes work queues more useful than assigning every possible status.

References: Government Data Quality Framework: communicating data quality ↗

Separate values, evidence, and workflow

A compact field group can contain value, status, source_url, observed_at, source_date, note, and next_action. For example, a named-role field can have an empty value, status 'conflicting,' two evidence references, and a note explaining that different current pages name different people. The workflow can assign that record to review without pretending either person has been confirmed.

Do not put literal 'unknown' inside a numeric revenue column or a date field simply because the spreadsheet looks tidier. Keep the typed value empty and use the status column. The same principle applies to booleans. A technology indicator should be able to represent observed, not observed under a defined check, and unresolved; forcing those states into true or false destroys information.

Preserve evidence even when it does not meet the requested threshold. A historical founder biography may help a future researcher identify the right person, but it should not populate a current executive-role field automatically. Store it as context with a historical label. In a narrative dossier, place uncertainty directly beside the affected claim rather than in a general disclaimer at the end.

Make 'not found' a reproducible statement

A bounded search specifies where the researcher looked and what would count as an answer. For a company's primary business model, the boundary might include its homepage, product or service pages, about page, and relevant current company-authored material. For a named management role, it may include a leadership page and a company announcement. The boundary should fit the field, not force every question through the same page checklist.

Record the search date and meaningful obstacles. 'No current named operations leader found on the reviewed management and company-news pages' is more useful than 'no operations leader.' The first describes research evidence; the second asserts something about the organization that public pages may not reveal. Even the absence of a management page does not prove the absence of management.

Choose a stopping rule before the work becomes open-ended. Stop when accepted sources provide sufficient evidence, when the defined search has been completed without support, or when a blocker requires buyer input. If a record matters enough to justify deeper investigation, create an explicit escalation. Repeatedly revisiting the same uninformative pages is not a better uncertainty policy.

Keep unknown distinct from negative evidence

A negative claim requires evidence appropriate to that claim. 'The company has no online store' is stronger than 'no checkout was observed on the reviewed pages.' 'The business does not use a CRM' is much stronger than 'no public CRM evidence was found.' Public web research often supports bounded observations while leaving company-wide negatives unresolved.

When a filter requires a positive condition, unknown can be excluded from the confirmed segment without being relabeled false. For example, a campaign for active merchants can include confirmed merchants and hold unresolved companies. That routing decision does not establish that the held companies are non-merchants. Preserving the distinction lets later research add them without rewriting the historical record.

Negative evidence is especially tricky when websites change by region, authentication state, consent choice, or subdomain. Record the surface you reviewed. A missing feature on a marketing homepage may exist inside a customer portal or another corporate property. Do not broaden a local observation into a statement about the entire company merely to satisfy a binary filter.

Worked example: a founder field that stays empty

Consider the fictional company Northbank Components. The supplied list asks for the current founder or administrator's name and business email. Its about page names two founders in a company history, a recent management page names a managing director, and a contact page shows only a shared address. None of those sources establishes that a founder currently runs the company or that the shared address belongs to a named person.

A careless completion would attach one founder's name to the shared address and mark the contact found. A defensible research record keeps the historical founder names as context, leaves the named current contact unresolved, and records why the identity-email relationship is unsupported. The company may remain a good target, but the required contact deliverable has not been completed.

The next action depends on the order specification. If a named current business contact is essential, hold the row for additional accepted research or replacement under the agreed terms. If the task was only to enrich an existing company list, return the unresolved contact field without claiming new contact creation. The key is to let the deliverable definition control acceptance instead of manufacturing a person-email pair.

Worked example: platform evidence conflicts

Imagine the fictional retailer Cedar Basket. A current product page contains a trace associated with Platform A. A recent agency case study says the store moved to Platform B. The homepage is delivered through a separate site builder. The research question asks which commerce platform currently supports the transactional store. A single brand-level platform label cannot safely represent all three observations.

Resolve the surface first. Identify the product, cart, checkout entry, and marketing domains without submitting an order or bypassing access controls. Record the observation date for each. The case study may refer to a migration project that has not fully launched, a different regional store, or a historical implementation. Its recency alone does not establish that it describes the currently inspected checkout.

If the available public evidence still conflicts, report the store platform as unresolved and retain the observations by domain. Do not vote among sources or automatically prefer the most familiar detection label. A future reviewer can investigate the specific conflict. Meanwhile, the company can remain in a broader ecommerce segment and stay out of a platform-specific campaign that requires stronger confirmation.

Decide how missing evidence affects scores

Separate fit from research confidence. A company can appear relevant to an offer while important evidence remains incomplete. A confidence or coverage indicator tells the reviewer how much of the score rests on supported inputs. Combining everything into a single number makes a low score ambiguous: the company may be unsuitable, or the researcher may simply lack evidence.

Set rules for mandatory conditions before scoring. If a service requires a confirmed active ecommerce operation, unresolved merchant status should block a confirmed-fit label. For optional preferences, you may leave points unassigned, show a score range, or display the result as provisional. Choose one treatment and document it. Do not normalize away missing essential evidence so that a thinly researched account receives the same apparent certainty as a thoroughly researched one.

Use score ranges sparingly and explain their meaning. An illustrative account with 40 supported points and 20 unresolved possible points can be shown as 40 confirmed points with up to 20 pending review. That is not a probability of purchase. It is an accounting of the scorecard's unresolved inputs. The sales action should still depend on which conditions are missing.

Match the retry to the reason

Retry inaccessible sources when the expected benefit justifies the effort, but do not treat a transient problem as permission to circumvent access restrictions. Use normal accessible pages, an accepted alternative primary source, or buyer-provided evidence. Record the retry date and result. A repeated access failure should eventually become a documented unresolved state rather than an endless background task.

Resolve ambiguity by asking for the missing discriminator. A country, company website, registration identifier, or existing account ID may separate two businesses with the same name. More broad searching can add candidates without resolving the original uncertainty. When identity is uncertain, pause dependent enrichment rather than producing detailed research about whichever candidate appears first.

For conflicts, compare entity scope, observation date, and claim wording. A company history and a current leadership page may not truly disagree; they may describe different periods. A parent website and a subsidiary website may both be correct at their own level. Escalate only the remaining substantive conflict, with a concise note showing what would resolve it.

Evaluate unknowns without rewarding guesswork

Track unresolved rates by field and cause, but do not make the lowest unknown rate the sole quality target. A researcher can reduce unknowns by making unsupported assumptions. Review a sample of populated fields as well as a sample of abstentions. The first finds overclaiming; the second finds avoidable missed evidence or an overly narrow search method.

Compare like with like. Name-only inputs may produce more identity ambiguity than supplied verified domains. Executive roles may be less publicly visible in one business segment than another. If the input mix changes, the unknown rate can change without any decline in research skill. Report the population and meaningful subgroups beside the metric.

Investigate whether unknowns affect the most important decisions. An unresolved optional phone number and an unresolved company identity should not receive the same operational priority. A weekly review can inspect the highest-impact unknown fields, identify repeated causes, and decide whether to improve inputs, revise the scope, or accept the limitation. The objective is better decisions, not prettier completeness charts.

Preserve uncertainty through the handoff

Check what happens when the export enters the buyer's spreadsheet, CRM, or scoring process. An empty boolean may become false; a blank numeric field may become zero; an 'unknown' label may disappear from a dropdown mapping. These transformations can silently change research meaning. Test a small set of known, unknown, conflicting, and inapplicable records before a broad import.

Keep original research status even if the receiving system needs a simpler operational queue. The CRM can route a company to 'needs review' while retaining the more precise cause in a separate field or linked dossier. A narrative note should identify the exact attribute at issue. A generic 'data may be incomplete' sentence gives a salesperson little help.

Also preserve the distinction between internal research and external claims. A salesperson should not convert an unresolved hypothesis into a confident opening line. If a platform is uncertain, a discovery question may be appropriate when outreach is otherwise justified, but the message should not state that the company uses that platform. Research status should travel far enough to influence the wording.

Put a useful unknown policy into practice

Start with a handful of real fields from the next research order. Write the value definition, acceptable evidence, permitted unknown statuses, and next action for each one. Test the policy on a straightforward record, an ambiguous identity, an inaccessible source, and a genuine conflict. The test should reveal whether two reviewers would record the same state and send the row to the same queue.

Review the vocabulary after the first delivery, but resist adding a new label for every unusual case. Use notes for details and reserve statuses for distinctions that change work. A compact, stable policy is easier to preserve through exports and handoffs than a taxonomy that grows with every reviewer. The final record should tell the buyer what is known, what is unresolved, and what would make the uncertainty worth revisiting.

• Keep unsupported values empty and record missingness in a separate typed status field.
• Define the source scope and stopping rule for each important research question.
• Preserve row identity even when the company match cannot be resolved.
• Require evidence before converting unknown into a negative company-wide claim.
• Route essential unknowns to review and optional unknowns according to the accepted specification.
• Retain historical evidence with its date instead of presenting it as current.
• Audit both abstentions and populated fields for avoidable error.
• Test exports so blanks and uncertainty states survive downstream processing.