Score missing information without pretending it is a no

A blank field does not answer the question.

Suppose your qualification policy asks whether the requester has a relevant role. A verified irrelevant role can answer “false.” A blank or stale job title answers “unknown.” Treating both as zero hides the difference between a poor fit and a missing observation.

The rubric in this guide is an original synthetic example of QualiRelay’s deterministic policy approach. It is not a learned model, a buyer probability, or a validated scoring standard. Your team must choose and approve the criteria against representative examples.

Three-valued logic is an established way to distinguish true, false, and unknown; PostgreSQL’s logical-operator reference describes that distinction in SQL. The weighted score and routing rules below are our application example, not a feature or recommendation of PostgreSQL.

Preserve the possible range.

For each criterion, record its weight and one of three results:

• True: the permitted, sufficiently fresh evidence satisfies the rule. Add its weight to both ends of the score range.
• False: the evidence establishes that the rule is not satisfied. Add zero to both ends.
• Unknown: the evidence cannot establish either result. Add zero to the minimum and its weight to the maximum.

The minimum is the sum of known true weights. The maximum adds all unresolved weights. Do not divide by only the known criteria; that would make an incomplete record appear more certain than it is.

Work through a 65–85 result.

This example totals 100 possible points, with 70 allocated to fit and 30 to intent. The weights are illustrative business choices.

At a threshold of 80, this record is unresolved: its minimum is below 80, while its maximum is above it. Calling the inquiry a definite pass or definite fail invents knowledge. Display 65–85 out of 100, partial, with the role criterion visibly unresolved.

At a threshold of 60, the record passes even without knowing the role. At a threshold of 90, it fails that threshold even if the missing criterion becomes true. This gives you a simple test for the possible value of additional data.

Separate qualification from responsibility.

The request can need an immediate accountable owner even when qualification is incomplete. Existing customer ownership may resolve the route before the score matters. An explicit demo request can go to a standard handoff while an uncertain threshold receives review.

Document the precedence: known existing owner, ordered approved rules, then an accountable fallback. Uncertainty should not make a legitimate inquiry disappear, start an unapproved outbound campaign, or silently change an already acknowledged assignment.

Buy a missing fact only when it can matter.

Before an optional enrichment lookup, walk through four questions:

1. Is there an allowed unresolved field? A provider’s ability to return a field does not establish permission to use or retain it.
2. Could the field change an approved decision? In the example, a role lookup can affect the threshold of 80. It cannot change the 60 or 90 threshold outcome by itself.
3. Could existing permitted evidence answer it? Check the customer’s authoritative source and fresh CRM information first.
4. Is the purchase within an explicitly approved budget? Workflow allowance and supplier spend are separate authorizations.

Better explanatory prose alone is not a routing benefit. Nor is every threshold-crossing fact a reason to buy: if the existing account owner already determines the handoff, the additional score might not change the next action.

Keep evidence quality visible after a lookup.

Record where the result came from, when it was observed and retrieved, its permitted purpose, and whether it is known, stale, conflicting, unavailable, or not permitted. A provider no-match is not a universal statement that a company or role does not exist.

If a paid request times out after transmission, its cost can be unknown. Do not release the cost hold as if the request was free or immediately buy the same fact again. Reconciliation of the provider outcome is a separate operational task.

When late evidence arrives, preserve the original decision snapshot. A linked reevaluation can explain the difference. It should not silently reassign a prospect whose owner already accepted responsibility.

Review the rubric against decisions, not beautiful scores.

Choose a bounded sample with known fits, known mismatches, public-email contacts, stale company information, missing roles, conflicting associations, and repeat inquiries. Ask a responsible reviewer to label the expected disposition and explain disagreements.

Report the sample size, selection method, unresolved cases, and disagreements. A high agreement rate on a tiny selected sample is not proof of general predictive accuracy. Keep a separate set of examples for later checks when changing the policy.

The useful result is a defensible next action with visible uncertainty and a controlled cost. A larger profile or a narrower interval is supporting evidence, not the customer outcome itself.