Talking-point coverage: measuring what was actually discussed

Talking-point coverage measures how many of the planned scientific talking points for a meeting were genuinely discussed, as distinct from mentioned in passing or skipped altogether. It is one half of a depth measure, and it is the half most often built badly, because a tick box records that a subject came up and then stops.

This piece is about mechanics. The argument for why a depth measure should be built from coverage and insight yield, and from nothing else, belongs to what a depth measure should and should not include and is not re-run here. Take the design as settled and look at how the coverage half actually works: where the points come from, how a discussion is separated from a mention, where the benchmark comes from, and what the number is not entitled to say.

It matters because coverage is the component that looks simplest and is therefore the one implemented most carelessly, which is how a function ends up with a 90% coverage figure that means nothing and the measurement gap it sits inside stays exactly where it was.

Where do planned talking points come from?

Not from the MSL on the morning of the meeting, and not from one global list applied to every HCP in the country.

A planned scientific talking point is a specific scientific position or piece of evidence that the medical plan intends to communicate about a product, to a particular kind of healthcare professional, at this point in the lifecycle. The canonical definition sits in the glossary under planned scientific talking points; what matters here is where the list for a given meeting comes from.

Three inputs, in order of weight. The scientific communication plan decides what is worth communicating this cycle. The HCP's own history narrows it — what has already been covered thoroughly, what they asked about and never received an answer to, what was raised last time and deflected. The lifecycle stage sets the register: a pre-launch conversation and a conversation about an established product are not covering the same ground even when the product is identical.

Two things follow, and both are easy to get wrong.

Approved content is not a talking point. A deck is a vehicle. The talking point is the scientific position the deck carries. Systems that measure which materials were shown are measuring the vehicle, and a clinician can sit through an entire deck without a single scientific position being put to them.

The list must be fixed before the meeting. If points can be added afterwards to match what happened, coverage is not a measure, it is a description. The denominator has to be chosen in advance and left alone. That single rule does more for the integrity of the number than anything else in this piece.

What is the difference between discussed, mentioned and skipped?

This is where most implementations collapse, and the collapse is always in the same direction.

Skipped. The point did not come up. Nothing in the conversation touches it.

Mentioned. The subject appeared and went one way. The MSL said it, a slide carried it, the words were spoken, and nothing came back. There is no evidence the HCP engaged with it at all.

Discussed. The point was put and the healthcare professional engaged with it. They responded, challenged it, asked a question about it, or related it to the patients they see. There is something on the HCP's side of the exchange.

A binary covered-or-not field makes mentioned and discussed identical, which is the single largest source of inflated coverage figures in the field. It also makes the measure trivially gameable: read the list aloud at speed and score the same as someone who had a conversation.

The three-state distinction holds up because it is observable rather than interpretive. Either there is something the HCP said in response to that point, or there is not. That is a fact about the conversation, not a judgement about the meeting, which is what makes it possible to derive coverage from the interaction itself rather than from a self-report typed up at the end of a long day.

One edge case worth settling in advance. A point that is put and actively deflected — raised, and moved away from — records as mentioned, not discussed. The deflection is useful information, and it should be captured, but the honest reading is that the point did not land.

Why the benchmark has to be derived rather than set

The instinct is to pick a number. Cover most of the planned points and the meeting was good. It fails for two separate reasons.

A fixed target sits on the wrong side of the equation. Coverage depends on how long the meeting ran, how senior the clinician is, and whether one genuinely good question consumed the whole slot. A corridor conversation and a scheduled sit-down cannot share a benchmark, and comparing them produces a ranking of meeting types rather than of conversations.

Targets on coverage produce coverage. A team told to hit a number will hit it, and the cheapest route is to shorten each point until everything on the list gets said. That is the exact behaviour the three states exist to catch, and it trades depth for breadth in the one measure built to stop that trade. It is counting what happened rather than what was said wearing a better name.

A derived benchmark asks a different question: what do comparable interactions in this therapeutic area, at this length, with this kind of HCP, actually achieve? That figure describes what is attainable rather than what someone hoped for, and it moves as the underlying data moves.

There is a cost to this and it is worth accepting openly. A derived benchmark cannot exist before there is data to derive it from. Early on, the honest position is that there is no benchmark yet and the coverage figure is descriptive only. Reporting should say that in plain words rather than borrowing a number from somewhere else and presenting it as a standard.

What coverage cannot tell you on its own

Coverage is a measure of the agenda. It is not a measure of the science, and three things follow that are worth stating before someone reads too much into a high number.

It cannot tell you what the HCP concluded. A point can be discussed at length, understood completely, and rejected on good clinical grounds. Coverage records that the exchange happened. It has nothing to say about where it landed, and it should not be read as though it does.

It cannot tell you whether the right points were planned. Coverage measures performance against the plan, and takes the plan on trust. High coverage of a poorly chosen set is a well-executed conversation about things that did not need discussing. The measure is structurally blind to this, which is an argument for reviewing the plan, not for distrusting the measure.

It is not an MSL scorecard. A meeting where the clinician took the conversation somewhere more valuable than the agenda will show low coverage, and may well have been the most useful meeting of the quarter. This is the practical form of a commitment made elsewhere: the measure is of the conversation, not the person running it.

Which is the whole reason coverage is half a measure rather than a measure. It answers what was put. Something else has to answer what came back.

How coverage behaves across a therapeutic area

At the level of one meeting a coverage figure is close to meaningless, and treating it as a verdict on that meeting is the most common misuse of it. Aggregated across a therapeutic area over a quarter it becomes genuinely diagnostic, and it exposes three patterns that are invisible any other way.

A point with consistently low coverage across many MSLs and many HCPs. Rarely a discipline problem. Usually the point is awkward to raise naturally, sits too late in the material to survive a short meeting, or does not answer anything clinicians are currently asking.

A point with high coverage and a high mention-to-discussion ratio. Raised everywhere, engaged with nowhere. This is the most actionable pattern in the whole measure, and it almost always means the material is answering a question nobody has. That is a content finding, and it goes to the people who write the content.

A point covered thoroughly in one region and not another. Usually environmental — a difference in pathway, formulary process or local guidance — and worth understanding before anyone tries to correct it.

None of these are visible in a binary covered field, and none are visible from a single interaction. They are the actual return on measuring coverage properly, and they arrive as a by-product of conversations the team was already having.

RocketMSL derives talking-point coverage from the interaction itself, in three states rather than two, against a list fixed before the meeting began. See how coverage is derived from the conversation →