Governance-First Launch Forecasting: A Decision Support Framework
Quantifying revenue risk from launch timing uncertainty using analog-based scenario analysis
🚀 Live app:
Oncology Launch Curve Forecaster
💻 Source code:
GitHub repository
All 60 revenue values behind the original ten analog curves were re-checked against the companies’ own reports: 35 changed, and 55 are now verified against a primary source. The original dataset had Tecentriq in Swiss francs labeled as US dollars, an Imbruvica series that mixed two definitions, and several values that did not match the filings (Calquence 2019, for example, was $164M reported vs $372M stored). Each value now carries its source and verification status in the repository, and the dashboard figures are recomputed from the corrected data.
The original ten analogs were all prominent launches, so the bands overstated a typical launch. Every new oncology molecular entity first approved by the FDA in 2013–2017 was then screened (56 products), and the set grew to 30 analogs, including modest launches and one withdrawn drug. The median (P50) Year-5 revenue fell from about $4.3B to $981M. The figures below use the 30-analog set.
The core challenge in launch planning isn’t predicting success—it’s quantifying how much uncertainty costs when timing shifts.
1. Framing the problem
Pre-launch planning in pharmaceutical commercialization involves decisions under substantial uncertainty. Brand teams must commit to launch timing, resource allocation, and forecast ranges while facing unknowns about regulatory approval timing, manufacturing readiness, and market reception.
The natural instinct is to seek precision: better models, more data, tighter confidence intervals. But in practice, the harder problem is often structuring the conversation around forecast ranges, revenue-at-risk, and timing shifts—rather than eliminating uncertainty entirely.
This project approaches launch forecasting as a governance and framing problem, rather than a prediction exercise. The objective is not to forecast whether a specific launch will succeed, but to quantify how analog selection and timing assumptions translate into revenue risk—making trade-offs explicit for cross-functional discussion.
In practice, this means shifting the question from “What will happen?” to “What are we implicitly assuming—and what does that assumption cost if it’s wrong?”
2. The core decision question
The framework is designed around a single, actionable question:
What is the revenue at risk if launch timing shifts under uncertainty?
This question matters because it connects analytical work directly to decisions that leadership teams actually face:
- How much revenue exposure exists in Year 1 if launch slips by one or two quarters?
- What forecast commitment range (P25/P50/P75) is defensible given analog variation?
- How sensitive are these conclusions to which analogs are included?
In practice, timing assumptions also propagate into manufacturing readiness, commercial resourcing, and financial guidance—making early uncertainty costly to unwind later.
Rather than producing point forecasts, the framework surfaces ranges and trade-offs that support structured executive discussion.
3. Why governance-first analog selection
Analog-based forecasting is common in pharmaceutical commercial planning. The challenge is that analog selection is inherently subjective—and post-hoc adjustments can undermine credibility.
This framework applies a governance-first methodology:
- Define inclusion criteria before examining outcomes — Therapeutic area, launch window, and data availability requirements are specified upfront. Filters that look at outcomes (a minimum Year-5 revenue, excluding the top analog) are offered only as sensitivity checks.
- Record the analog list — The app does not enforce this, since anyone can change a selection and see the results. Instead, the Scenario Builder exports a dated scenario record: criteria, included and excluded products with the reason for each exclusion, whether an outcome-based filter was on, the data version, inputs, results and a free-text rationale. That record is what gets filed with the approval, which happens outside the app.
- Document rationale for every inclusion/exclusion — Every one of the 56 screened drugs has a recorded keep/drop reason, which creates an audit trail for reviewers.
- Test sensitivity explicitly — Option to exclude the top-performing analog tests whether conclusions depend on outliers.
This approach addresses a common critique of analog analysis: that analysts cherry-pick comparators to support predetermined conclusions. By separating selection criteria from outcome examination, the framework survives skeptical review.
This governance-first approach is operationalized in the app through four components: Executive Brief (scenario comparison and trade-off framing), Scenario Builder (criteria configuration and sensitivity testing), Analog Explorer (individual product examination with guardrails), and Methods & Data (assumption documentation and scope boundaries).
4. What the framework quantifies
Delay Exposure
A delay usually does not move the patent or exclusivity end date. The months lost at launch are therefore lost at the end of exclusivity, not simply shifted later. The framework shows that exposure next to the smaller amount of first-year sales that is only deferred:
Delay exposure (fixed exclusivity end) = delay_quarters / 4 × median Year-5 revenue
Deferred first-year sales = delay_quarters × (median Year-1 revenue / 4)For a two-quarter delay across all 30 analogs, the exposure is about $490M, against about $87M of first-year sales deferred. The exposure figure assumes the Year-5 run rate applies to the final quarters of exclusivity, often ten or more years after launch. Sales then could be higher (continued growth) or lower (competition, label changes), so it is an exposure estimate, not a bound. A full comparison would value the delayed and on-time revenue paths over the same calendar horizon. Share lost to earlier competitors is not modeled.
Product Mode: Analogs as Shapes
Even a complete analog set’s dollar figures say little about a specific product. In practice, analogs supply the shape of a launch, and the size comes from the product’s own forecast. The Scenario Builder therefore takes an optional expected Year-5 revenue. Each analog’s curve is divided by its own Year-5 revenue, and the shapes are scaled to that figure. For an expected $1B at Year 5, the median shape implies about $129M in Year 1 (P25–P75: $91M–$232M), and a two-quarter delay exposes about $500M at the end of exclusivity. Lartruvo, withdrawn before Year 5, has no defined shape and is left out of this mode.
This is explicitly labeled as a Year 1 average-quarter approximation—a simplification that provides directional magnitude without implying DCF-level precision.
Scenario Bands from Analog Variation
Percentile bands (P25/P50/P75) are derived from observed analog performance:
- P50: historical median of the selected analogs
- P25 / P75: historical interquartile range
These represent observed variation across analogs, not statistical confidence intervals, and P50 is not a 50% probability for a new product. The distinction matters: they quantify historical heterogeneity, not forecasting uncertainty. The median curve also joins each year’s median, which can come from different products, so it summarizes the set rather than tracing a launch that happened; the chart shows how many analogs sit behind each year (26 to 30).
Revenue figures describe Year-5 revenue, not a lifetime peak. Every analog is observed for five years after approval, and most were still growing at Year 5. Labeling that value “peak” would understate products like Keytruda, whose revenue kept climbing well beyond the observation window.
Selection Statistics
Dynamic statistics update as criteria change:
- Solid tumor vs. hematology count
- Approval year range
- Median Year-5 revenue
- Number of analogs meeting criteria
This transparency helps users understand what’s driving the scenario range.
5. Key assumptions documented
The framework relies on several simplifying assumptions that users should understand:
Demand deferral, not destruction
Delayed demand is assumed to shift forward in time rather than being permanently lost. Competitive dynamics and access erosion are not modeled.
Label expansion bias
Some analogs experienced significant post-launch indication expansions, creating potential upward bias in Year-5 revenue estimates. Products without similar expansion potential may underperform these benchmarks.
Era effects
The analog set covers FDA approvals from 2013 to 2017 only. Launches after 2018 face different conditions (pricing pressure, value-based contracting, more crowded classes), which this set does not represent.
Data availability
Sixteen eligible drugs were dropped because their owner did not report their sales separately for every year, or because ownership changed in the first five years. They include small launches (Farydak, Odomzo) and large ones (Tafinlar + Mekinist, Zejula), so the direction and size of any remaining bias are unknown. Five launch years are also missing (for example Kadcyla’s first two), so early-year percentiles rest on 26–29 analogs.
Worldwide revenue and launch-year alignment
Values are company-reported worldwide product sales, converted to US dollars at Federal Reserve average rates; geographic mix differences are not adjusted. Launch Year 1 is the 12 months after each drug’s FDA approval date. Companies report by fiscal year, so each annual total is spread evenly over its days and split across the launch years it overlaps. Sales usually grow within a year, so early launch years are slightly overstated. Worldwide sales are aligned to the US approval date, so a drug sold elsewhere first (Alecensa in Japan from 2014) has higher early launch years, and one launched outside the US later has lower ones.
These assumptions are documented in the Methods & Data tab and surfaced in the dashboard interface to prevent over-interpretation.
6. What this framework does—and does not—do
What it does
✅ Structures analog-based launch scenario analysis
✅ Quantifies revenue ranges from analog variation (P25/P50/P75)
✅ Supports structured trade-off conversations under uncertainty
✅ Documents assumptions and limitations transparently
✅ Provides sensitivity checks to test analytical robustness
What it does not do
❌ Predict whether a specific launch will succeed
❌ Recommend specific investment or resourcing levels
❌ Model competitive response or market access dynamics
❌ Estimate NPV or financial returns
❌ Replace clinical, regulatory, or strategic judgment
By design: NPV and market access modeling are excluded because they require company-specific inputs that would compromise generalizability. Revenue scenarios are treated as foundational inputs to financial modeling, not final outputs.
7. Dashboard structure
The application is organized into four complementary views:
Executive Brief
An executive-facing summary that surfaces:
- Key metrics (analog count, historical median Year-5 revenue and range, delay exposure, median Year-1 revenue)
- Launch trajectory scenarios with P25/P50/P75 bands
- Two explicit decision trade-offs: launch timing and the forecast benchmark
- Selected analog summary table
The decision boxes explicitly frame trade-offs in terms stakeholders recognize: “Speed vs. readiness” and “Alignment vs. credibility risk.”
Scenario Builder
An interactive configuration view that allows users to:
- Select therapeutic focus (solid tumors, hematology, or both)
- Run sensitivity checks: exclude the top-performing analog, or analogs below a Year-5 revenue level
- Configure launch delay scenarios (0–4 quarters)
- Download a dated scenario record for the approval file
The Selection Rationale panel updates dynamically, showing live statistics and a governance note reminding users that criteria should be documented before examining outcomes.
Analog Explorer
A detailed view for examining individual analog characteristics:
- Side-by-side comparison of two selected analogs
- Launch curve visualization showing trajectory differences
- Characteristics table (company, indication, approval date, Year-5 revenue)
- Launch context factors that may have influenced trajectory
This tab includes guardrails against over-interpretation: it opens on a single default analog, comparisons are an explicit choice, and an interpretation note reminds users that the framework’s value lies in the distribution across analogs, not any single curve.
Methods & Data
A transparency-focused section documenting:
- Framework overview and workflow diagram
- Data sources (FDA approvals, company-reported revenue) and the status of every value
- Inclusion and exclusion criteria
- Key assumptions with explicit discussion
- “What This Framework Does NOT Do” section
- Technical implementation details
8. The analog set
The framework includes every new oncology molecular entity first approved by the FDA in 2013–2017 whose sales one owner reported separately for five years (30 of 56 screened; CAR-T cell therapies are excluded as a different launch model):
Solid tumors (20):
Keytruda, Opdivo, Ibrance, Tagrisso, Lynparza, Tecentriq, Kadcyla, Xofigo, Cyramza, Lenvima, Unituxin, Alecensa, Lartruvo, Rubraca, Kisqali, Bavencio, Alunbrig, Imfinzi, Nerlynx, Verzenio
Hematology (10):
Imbruvica, Darzalex, Venclexta, Calquence, Pomalyst, Gazyva, Zydelig, Blincyto, Ninlaro, Empliciti
Lartruvo, withdrawn in 2019 after its confirmatory trial failed, is kept on purpose: leaving out launches that failed would bias the bands upward. The repository lists every screened drug and why it was kept or dropped.
Approval years range from 2013–2017, so every analog has a full five-year observation window while remaining recent enough to reflect modern launch dynamics. Within that window, a few reported values are missing (mostly launch years), which leaves five launch years blank across the set.
9. Technical lessons from the build
This project surfaced an important technical insight about Shiny development with CSS-based UI frameworks.
The Problem
When using shiny.semantic (which relies on Semantic UI’s CSS-based tab switching), outputs don’t render automatically when users navigate to a tab. Shiny’s lazy evaluation assumes it knows when tabs become visible—but CSS-based tabs don’t notify Shiny of visibility changes.
The Solution
Adding outputOptions(output, "output_name", suspendWhenHidden = FALSE) forces Shiny to render regardless of perceived visibility. This pattern isn’t documented in standard Shiny guides but is essential for CSS-based tab frameworks.
The catch is that it has to cover every output that depends on shared state. A later review found six small text outputs (the range line and both decision cards) without it. After a selection change, the headline number updated while the range beside it stayed stale. The complete fix pairs suspendWhenHidden = FALSE on all dependent outputs with a shown event triggered on each tab switch, and verifies the result by changing a selection and checking that every number moves together.
This discovery became the most valuable technical lesson from the project.
10. Why this framing matters for commercial analytics
This project suggests that in analog-based launch planning, the choice of comparators and assumptions can move the answer more than the method: widening the analog set alone cut the benchmark by three quarters. The failure modes it is built to guard against:
- Analog selection was post-hoc rationalized
- Assumptions weren’t documented or tested
- Ranges were too narrow to capture actual uncertainty
- Trade-offs weren’t made explicit to decision-makers
This framework addresses these failure modes directly:
- Governance-first selection prevents cherry-picking
- Documented assumptions create audit trails
- P25/P50/P75 bands acknowledge inherent uncertainty
- Explicit trade-off framing connects analysis to decisions
The goal is not better predictions, but better-structured conversations about what forecasts can and cannot support.
Appendix: Methodology & Build Notes
Data Sources
- FDA Drug Approvals: Approval dates (T=0), initial indications
- Company-reported revenue: Full-year worldwide product sales from annual reports and results releases (SEC 10-K and 8-K filings; AstraZeneca and Roche results announcements), with a source link for every value
- Federal Reserve: Annual average CHF/USD rates to convert Roche’s Tecentriq sales
Technical Stack
- Language: R (4.3+)
- Framework: Shiny (modular architecture)
- UI: shiny.semantic (Appsilon)
- Visualization: ggplot2, ggiraph
- Tables: reactable
- Deployment: shinyapps.io
Key Design Choices
- Launch years counted from each drug’s FDA approval date, with fiscal-year sales split across them by days
- Percentiles calculated across analog set (not confidence intervals)
- Revenue scaled to millions USD for readability
- All data from public sources to ensure reproducibility
Purpose, Scope, and Disclaimer
This project is a portfolio case study intended to demonstrate analytical framing, governance, and decision-support design using publicly available data.
It is not intended for real-world commercial decision-making and is not affiliated with any pharmaceutical company. Product names are used for analytical illustration only.
All data are derived from publicly available sources including company filings and FDA databases. No proprietary or confidential information is included.
This work does not constitute investment advice, commercial guidance, or strategic recommendations.
Closing Reflection
The value of this framework lies not in its predictions, but in its structure. By making analog selection defensible, assumptions transparent, and trade-offs explicit, it provides a foundation for disciplined commercial planning conversations—conversations that connect forecast ranges to revenue-at-risk and timing decisions to resource commitments.
In launch planning, clarity about what we don’t know is often more valuable than false confidence about what we do.
Updated October 2026: revenue figures relabeled as Year-5 revenue rather than peak; all 60 revenue values re-sourced from company filings (35 changed; see the correction note at the top); corrected the data-source, era-effects, launch-year alignment and outcome-filter descriptions; replaced the deferred-only delay estimate with a cost of delay that assumes a fixed exclusivity end date; added a product mode that scales analog launch shapes to an expected Year-5 revenue; expanded the technical lesson on tab rendering; expanded the analog set from 10 to 30 (every eligible 2013–2017 oncology launch) and aligned launch years to exact approval dates. After an external review: relabeled the delay cost as an exposure estimate rather than a lower bound, described percentiles as historical reference points, added a downloadable scenario record (criteria are saved when a selection is applied, so the record always matches its results), and stated that the direction of the remaining selection bias is unknown.
Citation
@online{ponce2026,
author = {Ponce, Steven},
title = {Governance-First {Launch} {Forecasting:} {A} {Decision}
{Support} {Framework}},
date = {2026-01-28},
url = {https://stevenponce.netlify.app/projects/standalone_visualizations/sa_2026-01-28.html},
langid = {en}
}