Handle Missing and Conflicting Evidence

A GTM Lab account executive is told her deal has an engagement score of 8, which is deep in the danger zone. It says the deal is drifting and nobody has spoken to the buyer in months.

She had a call with them yesterday.

The call happened. It went well. It was not logged the way the scoring calculation expects, so the field that records the last meeting was never stamped, so the calculation ran exactly as designed and concluded that no meeting had ever occurred. The system did not fail. It answered a question about its inputs while everybody read it as an answer about the deal.

Absence is not evidence

This is the most expensive mistake in this guide, and it is almost invisible, because a record with missing evidence produces a number that looks like every other number.

The failure has a shape. Some input is unavailable. The calculation, having nothing, uses the default for that slot, which is usually zero or null. Zero is not neutral. In almost every scoring design, zero is the worst possible value, so the record is penalized for a fact nobody observed. The output is indistinguishable from a record that genuinely earned a low score.

Three things fix it, and they are all cheap. Distinguish absent from bad in the inputs: a missing meeting date is not the same as an old meeting date, and the calculation has to be able to tell. Decide, per input, what absence means: sometimes an absence is genuinely informative, since no email in six months on an active deal really is a bad sign, and sometimes it means nothing at all, since no revenue figure on a company record says something about your data rather than about the company. Make insufficiency an outcome: if enough required evidence is missing, the answer is not a low score. It is “not scored,” and it must be distinguishable in every report from both a low score and a calculation that failed.

GTM Lab's fix is to detect that a required activity date is absent, decline to score, and mark the record as needing its activity logged. The account executive sees a prompt rather than a verdict.

When two sources disagree

The other half of this chapter is the opposite problem: not missing evidence, but too much. A rep enters a job title by hand. An enrichment provider returns a different one. Both are present, both are plausible, and something has to decide.

Keep Your CRM Data Trustworthy settled the policy: classify each field family in advance, and never silently overwrite a value a human deliberately entered. What this guide adds is what the decision does with the disagreement while it is unresolved, which is a different question from what the record stores.

Prefer by authority. Take the value from the source with authority for this decision, and record which one you took. Correct when authority is genuinely settled. Treat the conflict as a signal. Sometimes a disagreement is itself informative: two sources disagreeing about company size may mean the company recently changed size, which is more interesting than either value. Refuse to decide. When the conflicting field is the deciding one, the honest answer is review. Picking a side quietly is how a decision system develops a bias nobody can find.

The rule underneath all three: whichever you choose, record that a conflict existed and which value you used. A decision that silently picked a side is one you cannot audit later, and the audit is the only way you will ever discover the rule was wrong.

Carry this into your business

Pick a scored population and count how many records sit at or near the bottom. Then check how many of those are missing an input the score depends on. In most systems that have not been checked, the answer is uncomfortable, and every one of them is a record somebody has quietly stopped working.