04 · Simulations

Objections surfaced
before the real world asks.

A simulated panel of experts or patients reads your document or your direction and tells you where it pushes back, which statements draw challenge, and why. Run it when the question comes up, not when the logistics allow.

Tested against real-world KOL decisions

Calibrated to real-world KOLs.
Superior to general-purpose AI.

Simulated panels were run on questions real expert panels had already decided, and their votes compared with the recorded outcome.

226 recorded KOL decisions, against two AI systems
214of 226
Posteros simulated panelvotes matching the experts' recorded vote · 12 misses
146of 226
A general-purpose AI model, run the same wayvotes matching · 80 misses
What this does and does not show. It tests vote fidelity, not whether each simulated panelist reasons like a named expert. The counts are as recorded. The comparison was a widely used commercial general-purpose AI model, run with the same inputs and scored the same way; it is not named on this page. Method and counts available on request.
Three shapes

A panel that deliberates, or a broad read on specific statements.

Choose the shape by the question: a document to be read whole, or statements to be tested one at a time.

KOL Simulation
A seated panel of expertsNine to twelve simulated clinicians read a document or a direction, deliberate, and report where they push back, panelist by panelist.
Patient Simulation
A seated panel of patientsThe same read from the patient's side: what is reassuring, what is alarming, what is missing from the story as told.
KOL Reactions
A broad read on specific statementsMany simulated experts, one statement at a time: would they let it stand, and if not, what would they need to see.
Simulated panel · 9 endocrinologists illustrative

“Delivers the greatest weight reduction of any approved therapy in this class.” Would you let that stand?

Seven wouldwith the head-to-head cited
Two would notand said why
One of the two“Not as a superlative. The comparison went the other way in the population I treat. Say where it sits, and I'll accept it.”
The divided reads shown as divided. Full read attached →
Illustrative
✦ What arrives

Who pushed back, on what, and why.

A read is not a score. It is the panel, written down: who pushed back, on which statement, and the evidence that made them. Divided reads are shown as divided; a seven-to-two split on a superlative is the finding. Where a panelist asks for something, the request is quoted, so your writers know what to change.

Arrives as the objections, panelist by panelist, with what raised each; the full read attached.
Timing within the week for an agreed document and question.
Status in use; tested against real-world KOL decisions (above).
✦ When

Feedback as often as your strategy demands.

Expert feedback has always been rationed by logistics. An advisory board takes months to convene, so questions wait, and many are never asked. A simulated panel can read a scientific communication platform (SCP), a publication plan or a new direction the week the question comes up. Advisory boards and interviews still matter. You start them with the objections already on the table.

Reads a scientific platform · a publication plan · a lexicon · a new direction · a single claim.
Before an advisory board, a congress, a review, a launch.
Simulated panel · read of a draft · excerpt illustrative
Panelist 3Would not let the superlative stand. Raised by: head-to-head comparison, 2025 Panelist 5Accepts the tolerability statement; asks for the discontinuation rate beside it. Raised by: pooled safety analysis, 2025 Panelist 8Asks why the cardiovascular outcome is absent. Raised by: outcomes trial, 2024
Nine read it. Seven would let the claim stand, two would not. Full read attached →
Illustrative

Hear the objections first.
Then meet the experts.

Reads a platform, a plan, a direction or a single claim, before the experts do.

Talk to us All four products