Panel · Use case · Corporate Development

Probability of Success: Phase 3 Entry.

LiveEstimate Elicitation

What is the probability that this phase 2 asset meets its pivotal endpoint and obtains approval, given the data to date?

QuantityProbability of meeting the pivotal endpoint and obtaining approval

SeatsRegulatory, operations, evidence, medical

ReaderDevelopment and portfolio leads

Not askedA valuation

DepthStandard, four seats

Probability of Success: Phase 3 Entry is a ready-made use case for Corporate Development. It ships under the Estimate Elicitation template.

For corporate development leads. You get each seat's low, central and high with its anchors, the pooled range and what it is not, and what would narrow it. Claims are cited or marked as gaps.

Add it to your library to reuse from New → From template. Edit the saved copy in your library on the service form.

The decision

What is the probability that this Phase 2 asset meets its pivotal endpoint and obtains approval, given the data to date?

Who estimates

A probability of success usually depends on more than one thing: the endpoint, the regulatory acceptance and the commercial case. If two targets must both be met, treating them as independent misstates the result.

Seat	What it estimates from	Why it is at the table
Regulatory Strategist	The chance that the agency accepts the endpoint and the data for approval.^{2U.S. Food and Drug Administration. Accelerated Approval Program.}	Agency guidance and precedents.
Clinical Operations Director	The chance that the Phase 3 is executed as planned and reaches its endpoint.^{3BIO, Informa Pharma Intelligence and QLS. Clinical Development Success Rates and Contributing Factors 2011–2020. February 2021.}	Industry success rates for the phase and area.
Global Real-World Evidence Lead	How much the Phase 2 effect is likely to shrink in a larger trial.	Replication and regression patterns.
Global Medical Affairs Director	Whether the effect size is clinically meaningful and holds in the intended population.	Clinical context.

What the elicitation does

Say which targets must all be met. The quantity states whether approval needs one endpoint or two. In one published example the simulated probability of success was 4 percent when both targets were required and 41 percent when only one was.^{1Holzhauer, B. et al. Eliciting judgements about dependent quantities of interest: the SHELF extension and copula methods illustrated using an asthma case study. arXiv:2102.02852.}

Base rates anchor the seats. The dossier carries an industry analysis of phase transition rates as a reference class, and each seat says how the asset differs.^{3BIO, Informa Pharma Intelligence and QLS. Clinical Development Success Rates and Contributing Factors 2011–2020. February 2021.}

Range before point. Each seat gives a plausible low and high before the central value, to limit anchoring and overconfidence.^{4A review of prior knowledge elicitation: biases, task design, seed variables and aggregation. arXiv:2112.01380.}

Recommended length

Standard

Four angles bear on the number. Standard depth gives each a seat, with the dossier and the reasons for the spread, in about four pages.

You can pick another length on the form. The other two stay available.

Assumptions the elicitation checks

The elicitation states the assumptions each estimate rests on, and says what each gap limits. Examples for this quantity:

The reference class of past programs is similar to this asset.
The Phase 2 result is not a chance finding.
The agency’s view of the endpoint will not change.

Standards, guidance and sources

Method and anchors

Holzhauer et al., dependent quantities of interest in expert elicitation (arXiv).^{1Holzhauer, B. et al. Eliciting judgements about dependent quantities of interest: the SHELF extension and copula methods illustrated using an asthma case study. arXiv:2102.02852.}
Industry analysis of clinical development success rates, 2011–2020.^{3BIO, Informa Pharma Intelligence and QLS. Clinical Development Success Rates and Contributing Factors 2011–2020. February 2021.}
Review of prior knowledge elicitation: biases and task design.^{4A review of prior knowledge elicitation: biases, task design, seed variables and aggregation. arXiv:2112.01380.}

The decision practice behind the memo itself is on the template page.

What gets flagged

Any anchor with no source, any seat whose interval is much narrower than the others, and any probability that treats dependent targets as independent.

What arrives

The quantity and how it is judged
Evidence dossier
Seat judgments
Estimates side by side
Why the seats differ
Revised estimates after rebuttal
The pooled range and what it is not
What would narrow the range

What it will not do

The pool is a synthesis of AI judgments and not a calibrated distribution, and the seats are not weighted by performance. It does not narrow the range to look decisive.

It does not value the asset.

It is not a calibrated probability distribution. The pool is a synthesis of AI judgments, and the report says so.

It does not weight seats by performance. No calibration record exists for the AI seats.

It does not narrow the range to look decisive.

References

[1] Holzhauer, B. et al. Eliciting judgements about dependent quantities of interest: the SHELF extension and copula methods illustrated using an asthma case study. arXiv:2102.02852. Read in the template research via an ar5iv copy.
[2] U.S. Food and Drug Administration. Accelerated Approval Program. Known from FDA’s site; not read in full.
[3] BIO, Informa Pharma Intelligence and QLS. Clinical Development Success Rates and Contributing Factors 2011–2020. February 2021. Known from search results and secondary reports; the figures were not read.
[4] A review of prior knowledge elicitation: biases, task design, seed variables and aggregation. arXiv:2112.01380. Read in the template research via an ar5iv copy.

Links checked 9 October 2026

These are the sources the template drew on. None of their authors or issuers has reviewed it.

About the Panel service

Each panelist answers in their own voice, then the discussion is drawn together. You get the question and who was in the room, every position in full, what they agreed on, where they split, and what follows. Disagreements are written down rather than smoothed into a consensus nobody actually holds, because the split is usually the part you needed.

To edit a saved template, open New → Panel → From template, or the library rail on the service form.