The kickoff was a success. The launch video got a standing ovation, the brand team's energy was contagious, and the reps left the plenary genuinely excited. Three weeks later, a respiratory specialist asks one of them for the exacerbation reduction in the registration trial, and the answer comes out rounded, blended with the competitor's figure, or simply wrong.
That is the moment a launch is actually judged. Our view, after 100 programs with field forces, is blunt: in a regulated conversation, roughly right is wrong. Enthusiasm opens the door, but precision is what keeps it open, and precision does not come from motivation. It comes from knowing exactly which details your field gets wrong, why they get them wrong, and which of those errors are unlikely to fix themselves. The detail goes missing in predictable places, some errors resist repetition, and a launch program can be built around exactness rather than energy.
Launch conversations are won on four or five exact facts
Every launch story has a lot of words in it. Disease burden, unmet need, mechanism, trial design, results, safety, dosing, access. The visual aid runs to many pages. But when you listen to how specialists actually test a new product in the first weeks, the conversation tends to narrow to a handful of exact things.
The primary endpoint, stated with the right number and the right comparator. The population the label covers, and where it stops. The safety signal they have heard about at a congress. The dose adjustment for the patient they saw that morning. And, almost always, how this compares with the thing they already prescribe.
These are not areas where a rep can be vaguely correct. A number that is slightly off is not a small error to a specialist who read the paper last night; it is a signal that the rep has not. A benefit extended one step beyond the indication is not enthusiasm; it is an off-label claim that MLR spent weeks making sure the materials never contained. Precision is not pedantry here. It is credibility, and in a few places it is compliance.
PharmExec's recent work on launch drift, the gap between strategy and what actually happens in the field, names HCP belief barriers and variability in field execution among its root causes. We would put it more plainly. A strategy is only as precise as the least precise conversation that carries it.
Where the detail goes missing: four families of error
When we look at first-attempt answers across the 100 programs in our case library (about 9 million answers from 2016 to 2026), the same families of error show up again and again, whatever the therapy area.
- Numbers that drift: endpoints, percentages, thresholds and doses remembered approximately, often rounded towards whatever sounds more impressive.
- Mechanisms told backwards: a plausible explanation that sounds scientific and is the opposite of the real story, usually because it is simpler.
- Claims that slide off label: a benefit extended to a population, a line of therapy or a use the evidence does not support.
- Competitor confusion: our data and theirs, mixed up under pressure, especially when two trials have similar designs.
What is striking is not that these errors exist. Everyone who has sat through a ride-along knows they do. It is how concentrated they are. A single distractor, one wrong option on one question, can attract a large share of a field force at first attempt. That usually means the error was not random. Something in how the content was taught, or in how the competitor story is told, made the wrong answer feel right.
That is useful to know. A random error needs more practice. A shared error needs a different explanation.
Why numbers drift: the forgetting curve is not your friend
Exact values are the first thing to go. Murre and Dros's 2015 replication of Ebbinghaus, a single-participant relearning experiment with nonsense syllables, illustrates forgetting in that specific experiment; after 31 days the savings were about 4%.
We would not treat that as a forecast for a launch story. People learning clinical data they will use every day are not memorising nonsense syllables in a lab. What we take from it is the direction, and our own experience adds the rest: without retrieval, precise information tends to decay towards the gist. The rep remembers that the product "reduced exacerbations a lot". They do not remember by how much, in whom, against what.
And gist is exactly what a specialist will not accept.
This is why a single kickoff, however good, cannot carry a launch on its own. The detail has to be pulled back out of memory several times in the weeks after, in short bursts, with feedback on each attempt. Reading the slide again does not do the same job. Answering a question about it does.
Without retrieval, precise information decays towards the gist. And the gist is exactly what a specialist will not accept.
Some errors fix themselves. Some barely move.
Here is the finding from our data that changed how we design programs.
For pure recall, repetition works well. Once a person has seen the right value a few times, and has had to produce it themselves, it tends to stick. Most drifting numbers close this way. Across our case library, 43% of the 794,130 first-attempt gaps were verified closed, with the same person later answering the same question correctly. The 16 programs tied to a business moment such as a launch or kickoff closed 55%, against 41% for the other 84, an observed association rather than proof that the moment did it.
But some errors do not behave like memory lapses. The evidence here comes from a sample: the example questions we track in detail across programs, each followed from first attempt to last attempt within that program's window, which runs from a few days to several months. In one in five of those example questions, fewer than 20% of the people who were wrong corrected it. Some of those people were never retested, because they did not meet the question again before the program closed, so not all are proven misconceptions.
Where people did meet the question again and kept choosing the same wrong option, the error looks less like a gap in memory and more like a misconception: a belief that feels right, often because it fits an older mental model or a simpler story. A quiz, however well designed, is a weak tool against a belief. The tracked examples split roughly in thirds: questions where gaps closed, questions where they half closed, and questions where they stayed open at the close.
A scenario from a global immunology kickoff in our library makes it concrete. The program ran four franchise games at once for 1,330 players, with 404 questions over six days and around 301,000 answers. It closed 53% of 15,238 gaps, between 47 and 58% by franchise. Solid numbers. But one misconception about a mechanism, held by 125 reps, barely moved with repetition. Unless someone coached them, those reps were at risk of going into their next calls explaining the mechanism the wrong way round, with complete confidence.
That is the kind of error that damages trust, and it is invisible in a team average.
Team averages hide the conversations that hurt you
Most launch dashboards report knowledge, when they report it at all, as a score. The team scored well on the certification. One country is ahead of another. Everyone passed.
A strong average sounds like a field that is mostly ready. It can also mean that dozens of people hold the wrong mechanism, a few dozen are rounding the primary endpoint upwards, and a handful are quietly extending the indication. Averages are built to hide exactly the thing a launch lead most needs to see: which people are wrong about which facts.
A second example from the library shows the same pattern in MedTech. A vascular program across APAC ran seven congress challenges over 26 months, each in a three-day window, and recorded around 190,000 answers. Pooled across the waves, the five main content areas closed between 46% and 52% of their gaps; other content blocks closed 43%. And yet 110 of 345 players could not state the three factors behind device choice in complex anatomy. That is not a knowledge area you want a third of your field to be vague on in front of an interventional specialist.
This is why we insist on measuring at question level, person by person. Not because granular data is interesting in itself, but because the corrective action is different for each kind of error, and you cannot choose the action without seeing the error.
A fair objection: reps should have conversations, not recite numbers
A fair objection is that this all sounds like drilling. Good reps, the argument goes, do not recite figures. They listen, they ask about the patient, they connect the evidence to the specialist's practice. Precision drilling produces robots with a script.
We agree with the destination and disagree with the route. Nobody wants a rep who opens with a p-value. But the rep who can hold a real conversation, who can follow the specialist's line of thought and bring in the evidence when it is useful, is precisely the one who knows the facts cold. Fluency is built on precision. You cannot improvise around a number you are not sure of; you can only avoid it.
There is also a compliance angle that the objection misses. MLR review exists because the exact wording of a claim matters. A rep who paraphrases loosely is not being conversational, they are rewriting approved material in real time, without review. Precision is what lets a rep keep a conversation natural and within approved boundaries.
So yes to conversations. They just need to rest on exact facts.
How to design a launch program for precision
Precision is designed in, not added at the end. In our experience, a handful of choices make most of the difference.
- Build every question from the approved documents only, so the content the field practises is the content MLR signed off, and nothing else.
- Write distractors that mirror the real mistakes: the rounded number, the competitor's figure, the reversed mechanism, the neighbouring population. Easy wrong answers teach nothing.
- Flag the critical questions, the ones where a wrong answer is a credibility, safety or compliance risk, and weight readiness around them.
- Space the practice. Short challenges several times across the weeks after the kickoff do far more for exact values than one long test at the end.
- Close gaps person by person, and count a critical gap as closed only after repeated correct answers, not one lucky guess.
- Hand first-line managers the list of misconceptions that repetition did not fix, by person, before their next ride-along or cycle meeting.
That last step is the one most programs skip, and it is where the stubborn errors get resolved. A misconception rarely yields to another question. It yields to a conversation with someone who can explain why the intuitive answer is wrong.
A launch is judged in a few minutes. Make them exact.
Energy matters. A field force that believes in the product is a gift, and we would never argue for less of it. But energy without exactness is how a launch ends up sounding confident and slightly wrong in thousands of calls at once, which is worse than sounding unsure.
At Atrivity we build for that exactness: questions generated only from approved documents, with experts signing off and conflicts flagged; one-minute challenges that pull the detail back out of memory in the weeks after the kickoff; and Gap Fixing that sends each person the fix for their own mistakes and flags the misconceptions that need a manager. But the principle does not depend on any tool.
Before your next launch meeting, ask one question of your readiness data: which wrong answer does our field keep choosing? If nobody can tell you, that is the first detail to go and find.