What counts as a medical insight, and what does not

A medical insight is something a healthcare professional tells you that the organisation did not already know, and that someone who was not in the room could act on. Most of what field teams file under the word fails one of those two tests, which is why insight libraries fill up and nobody reads them.

Ask ten people in Medical Affairs to define an insight and you will get ten answers, most of them circular. It is a learning. It is something valuable from an interaction. It is what the Medical Science Liaison writes in the insight field.

The circularity is not harmless. An organisation that cannot say what an insight is cannot tell whether it is getting any, cannot train for them, and cannot improve at generating them. What it can do is count rows in a database and call that insight capture, which is a version of the wider measurement problem rather than a solution to it.

What is a medical insight?

A medical insight is a specific piece of information about clinical practice, the evidence base, or the environment a product operates in, learned from a healthcare professional, that changes what someone in the organisation would do.

Three components, and all three have to hold.

It comes from the HCP. Not from the MSL's interpretation of the mood in the room, and not from the MSL's own view of the market. The record should be able to point at something the clinician actually said.

It is new to the organisation. This is the component most often applied at the wrong level. Something can be genuinely new to the MSL, and new to that relationship, while being documented in the medical plan and known to every other person on the team. Novelty is assessed against what the company knows, not what the individual knew before the meeting started.

Someone can act on it. Not necessarily field medical. Usually not field medical, in fact — the value of a good insight is that it is useful to publications, medical strategy, evidence generation, medical information or medical content, none of whom were in the room.

Which gives the test worth applying to every entry:

Could someone who was not in the room do something different because of this?

If the answer is no, it is not an insight. It may still be worth recording somewhere. It is not this.

What does not count?

Three things get filed as insights constantly, and none of them survive the test.

A topic without a finding. "We discussed the safety profile." That is talking-point coverage, which is a separate measure and is already recorded separately. It tells you the subject came up. It tells you nothing about what was learned when it did.

A restatement of something already known. "She is interested in real-world evidence" is a fact about the specialty that the medical plan already names. It felt new in the conversation, which is why it gets written down. Organisation-level novelty is the bar, and it is a higher one than it first appears.

A judgement about the person rather than a fact about the science. "Dr A is a detractor." This is not verifiable, not specific, cannot be acted on by anyone, and decays within a quarter. It also drifts the record toward commercial targeting, which is the one direction a medical function cannot afford to drift.

Four worked examples, and what makes each one usable

Illustrative, not drawn from any real account.

1 · A specific evidence gap tied to a decision. She will not move without comparator data in the population she actually treats, and has said so in two consecutive meetings. It names a barrier, names who it applies to, and establishes a pattern rather than a one-off remark. Acts: evidence generation and publication planning.

2 · A fact about the system, not the science. The formulary review lands in Q4 and the criteria are being drafted by pharmacy rather than the specialist team. Neither clinical nor commercial. It is environmental, it is invisible from head office, and it changes the timing and the audience of everything planned for that account. Acts: field medical planning.

3 · A recurring misreading of the evidence. Three clinicians in the same specialty have interpreted the primary endpoint the same incorrect way. A single instance is noise. The pattern is the insight, which is only visible if insights are recorded in a form comparable across records — a point worth sitting with, because it is the reason free text fails. Acts: medical content and MLR.

4 · An unmet need stated in practice terms. The patients she sees most often are the ones the pivotal trial excluded. Not a complaint about the trial. A statement about the distance between the evidence and the clinic, which is precisely the information evidence planning exists to consume. Acts: medical strategy and evidence generation.

The pattern across all four: each is specific, each is traceable to something the HCP said, and each has an owner outside the field team. None of them requires knowing who the clinician was to be useful.

Why "positive discussion" is not an insight

Sentiment is the single most commonly captured thing in an insight field and the least useful, for four reasons that compound.

It is the MSL's reading rather than the HCP's statement. It does not vary — almost every meeting is logged as positive, and a field that never varies carries no information. Nobody downstream can act on it. And it quietly rewards the wrong behaviour, because an agreeable meeting scores better than a difficult one with a sceptical, well-read specialist, when the difficult meeting is usually where the usable information is.

The same argument is why sentiment should not drive a depth measure either, which is set out at more length in what scientific depth should mean.

There is a more practical version of this. A mandatory field at the end of a long day will get filled with something. If the definition is loose, that something is sentiment. The definition is not an academic exercise; it is the thing standing between a useful record and a field full of the word "engaged".

How many insights should a field team generate?

This is the wrong question, and it is worth resisting rather than answering.

The honest position is that there is no correct number. It varies with the maturity of the evidence base, the stage of the product lifecycle, the therapeutic area, and how well the HCP already knows the data. An early-phase asset in a crowded specialty and an established product in a settled one should not produce the same rate, and a benchmark that says otherwise is describing an average of organisations with nothing in common.

More importantly, a target destroys the definition. Set a quota per meeting and the field will meet it, because the field always meets it. The bar drops, the entries get vaguer, the volume goes up, and the signal goes down. That is counting activity instead of substance with an extra step.

Insight yield, as a measure, asks whether a conversation produced anything usable. It is not a count of rows and it should never be read as one.

If an audit question is wanted, here is one that works. Take twenty insights at random from the last quarter, and for each one ask who acted and what changed. If the answer is nobody for most of them, the problem is not that the team is generating too few insights. It is that the organisation has not agreed what one is.

RocketMSL captures insights from the conversation as it happens, structures them so they are comparable across records, and routes each one to the function that can act on it. See how insights are captured and routed →