Insight yield: measuring what came back from the conversation

Insight yield measures the scientific substance a healthcare professional contributed during an interaction. It is scored for what came back rather than counted by how many records were filed, and that distinction is the difference between a measure that improves insight quality and one that quietly destroys it.

Talking-point coverage answers what was put to the clinician. Insight yield answers what came back. Together they are the two components of a depth measure, and the case for why it is those two and not a longer list belongs to why a depth measure is built from two components and not seven rather than to this piece.

This one is about the yield half: why it is scored instead of counted, what actually varies between one contribution and another, and what the pair tells you when they are read together. That last part is the reason both halves exist, and it is where most of the wider measurement problem becomes tractable.

What is insight yield?

Insight yield asks one question about a conversation: did it produce scientific substance the organisation did not already have?

The unit is the interaction. A meeting has a yield. An MSL does not have a yield and neither does a healthcare professional, although patterns across their interactions are readable and that is what section four is about. The canonical short definition sits in the glossary under insight yield.

Two things it is deliberately not.

It is not a count of insight records. The number of rows filed after a meeting is a measure of administrative effort. It tracks how diligent someone was with a form, which correlates with the value of the conversation only loosely and sometimes inversely.

It is not a rating of how the meeting felt. Agreeableness, receptiveness and enthusiasm are all things a person in the room can perceive, and none of them is scientific substance. A difficult conversation with a sceptical, well-read specialist routinely produces more usable information than a warm one.

One design requirement follows from this and it is not negotiable: zero has to be a legitimate, recordable result. Meetings happen that are well prepared, well covered and produce nothing the organisation did not know. That is a normal outcome and often not anyone's fault. If the system cannot record a zero — because a field is mandatory, or because a blank looks like a failure to complete the form — then something gets written in every time, and every value above zero is inflated by the ones that should have been zero.

Why counting insights produces worse insights

This is Goodhart's law applied one level up, and it moves faster than people expect.

The definitional version of the argument belongs to what counts as a medical insight: put a quota on insights and the bar for what qualifies drops until the quota is met. What matters here is the measurement consequence, which is what you have to build instead.

Counting rewards the act of filing. The cheapest ways to raise a count are all available within a quarter: split one observation into three records, file restatements of things the medical plan already says, file the topic that came up rather than the finding that came out of it. None of these require anyone to act in bad faith. They are simply what a reasonable person does when the number on the report is a count.

The damage is not that the measure fails. It is that the underlying asset degrades. Volume rises, mean quality falls, and the library becomes something nobody downstream is willing to read — at which point the genuinely good insights are lost inside it along with everything else. A count does not merely fail to measure insight quality; it actively consumes it.

So the measure has to be scored on substance rather than counted, and it has to be able to come back at zero. There is a second consequence that follows immediately: a scored measure cannot be self-reported. Anyone assessed on what they contributed will describe their contribution generously, and they will be right to, because the alternative is to disadvantage themselves for being honest. The score has to be derived from the conversation itself.

What makes one insight worth more than another

Four things, and none of them is how much the clinician liked the meeting.

Specificity. "She has concerns about safety" and "she will not move without comparator data in the population she actually treats, and has said so twice" are not the same contribution. The second can be acted on by someone who was not in the room. The first requires going back to the MSL to find out what it meant, which usually means it never gets acted on at all.

Novelty to the organisation. Not novelty to the MSL, and not novelty to that relationship. Something can be new in the room and thoroughly documented in the medical plan. The bar is what the company knows, and it is a higher bar than it looks.

Attributability. Traceable to something the healthcare professional actually said, rather than inferred from how the conversation went. Inference is not worthless — it is often right — but it is a hypothesis, and it cannot carry the same confidence into an evidence plan as a direct statement.

Reach. Who can act on it. A contribution only the field team can use is worth less than one that changes a publication plan, a piece of medical content or an evidence-generation decision. Reach is the quality most often overlooked, and it is the one that determines whether an insight leaves the library at all.

What is deliberately not published anywhere is how these combine into a number. That is not evasion. A published rubric becomes a target within a quarter, and the arithmetic is the least stable part of the design — the dimensions have held, the weighting has not and should not be treated as fixed. Anyone evaluating a measurement system should ask a vendor what the dimensions are and whether they can defend them. That is the answerable question.

How yield behaves differently by HCP and by topic

Yield is not evenly available, and treating it as though it were produces a ranking of territories rather than of conversations.

By healthcare professional. A senior specialist who has read the primary publication and the subgroup analysis will contribute more per meeting than a clinician newly working in the area. That is a fact about them. It says nothing about the quality of the MSL who met them, which is why yield cannot be used as a field scorecard — territory composition would decide the outcome before anyone left the office. This is the practical form of a commitment made elsewhere: the measure is of the conversation, not the person running it.

By topic. Where the evidence is settled, there is not much left for a clinician to contribute, and low yield is the expected result. Where it is contested or thin, yield should be high, and a run of low-yield conversations on a contested topic is a finding rather than a disappointment.

By stage of the relationship. Early interactions tend to produce more raw novelty, because almost nothing is on the record yet. Later ones produce less volume and more specificity, which is the healthier pattern and the one a count would score as decline.

The practical rule that falls out of all three: yield is compared like with like — same topic, comparable clinician, comparable stage — or it is not compared at all. An aggregate yield figure across an unlike population is not wrong so much as meaningless.

What high coverage with low yield is telling you

Coverage and yield were designed to be read together, and the reason is this section. Four combinations, and one of them is the reason the pair earns its keep.

High coverage, high yield. The planned points were put and the clinician engaged with them. The conversation worked. Nothing to investigate.

Low coverage, high yield. The clinician took the conversation somewhere other than the plan and it was worth going. Usually a good outcome, occasionally a sign the plan was aimed at the wrong thing. Worth reading what they went to instead.

Low coverage, low yield. Almost always logistics. The meeting was short, interrupted, or moved. This is the combination people most fear and the one that least often means anything.

High coverage, low yield. Everything was put and nothing came back. Once, this is a quiet meeting. Repeated across many clinicians on the same point, it is a content finding: the material is answering a question nobody is asking, and it is being delivered faithfully by people doing their job properly.

That last combination is the most actionable output of the whole measure, and it is invisible if either half is measured alone. A coverage-only system reports a strong quarter, because everything was delivered. An insight-count-only system reports a weak one, because little was filed, and it points at the field team. Neither of them tells you the thing that is actually true, which is that the material is wrong.

It is also a finding with an owner outside field medical. It goes to the people who plan and write the content, and it arrives as a by-product of conversations the organisation was already having.

RocketMSL derives insight yield from the interaction itself, scored on substance rather than counted by volume, and reads it alongside talking-point coverage. See how yield is derived from the conversation →