Pharmaceutical companies decide which drug candidates to fund by modelling the odds of success at each development stage against the cost of getting there, then weighing that against the expected payoff if the drug reaches market. Most candidates fail. The question a decision model actually answers isn't which drug will succeed, but which bets are worth making given how steep those odds really are.
How Do Pharmaceutical Companies Decide Which Drug Candidates to Fund?
A single drug candidate can take a decade and cost hundreds of millions of dollars to move from early testing to approval, and the vast majority never make it that far. Given that reality, a company can't evaluate candidates one at a time in isolation. It has to compare them against each other, weighing each one's odds of clearing every remaining phase against the resources that phase will consume.
Analytica is built for exactly this kind of portfolio-level comparison, letting a company model dozens of candidates side by side rather than evaluating each one in a vacuum, then feeding uncertainty about trial outcomes, costs, and timelines directly into the comparison instead of treating each figure as fixed.
That shift, from evaluating one candidate at a time to comparing a full portfolio under uncertainty, is what actually changes funding decisions. A candidate that looks strong in isolation might still lose out to one with slightly lower odds but a much larger payoff if it succeeds, and a portfolio model surfaces that tradeoff clearly.
Why Is Drug Development Risk So Hard to Estimate?
The risk compounds at every stage. A candidate must clear preclinical testing, then three separate phases of clinical trials, then regulatory review, and failure at any point ends the effort. Each phase has its own attrition rate, and those rates vary significantly by therapeutic area, meaning an oncology candidate and a cardiovascular candidate face genuinely different odds even at the same development stage.
This is also where cost and risk interact in an unforgiving way. The later a candidate fails, the more money has already been spent, so a portfolio strategy must account not just for the odds of failure, but for when that failure is most likely to happen. Two candidates with identical overall success probabilities can carry very different financial risk if one tends to fail early and the other tends to fail expensively, late in Phase 3.
What Do the Actual Attrition Numbers Look Like?
BIO's report on clinical development success rates breaks down these odds by therapeutic area and development phase, based on a decade of aggregated trial data. The figures make the scale of the problem concrete rather than abstract: overall success rates from Phase 1 to approval vary widely by disease area, and Phase 2 in particular tends to be where the steepest drop-off happens across most categories.
That phase-by-phase breakdown is exactly what a decision model needs as its input. A company modelling a portfolio isn't just plugging in a single overall success rate for each candidate. It builds in the specific odds of clearing each phase, since a candidate with strong preclinical data but a historically difficult Phase 2 for its therapeutic area carries a different risk profile than the raw numbers alone would suggest.
How Does This Change What Actually Gets Funded?
Without a formal model, funding decisions tend to gravitate toward whichever candidate has the most exciting early data or the loudest internal advocate. That's not a reliable way to allocate resources when most candidates fail regardless of how promising they looked at the start.
A decision model forces a more disciplined comparison. It weighs the same set of factors for every candidate: phase-by-phase odds, remaining cost to reach the next milestone, and the value of the outcome if the candidate ultimately succeeds. That doesn't guarantee better picks, since the underlying uncertainty is real and no model eliminates it. What it does is make sure a promising-sounding candidate with genuinely poor odds doesn't crowd out a less flashy one with a stronger risk-adjusted case.
One limitation is worth stating honestly. These models depend on historical attrition data that reflects past candidates, and a genuinely novel mechanism or target may not fit neatly into those historical categories. In those cases, the model is still useful for structuring the comparison, but the inputs deserve extra scrutiny rather than blind trust in the historical averages.
FAQ
How do pharmaceutical companies decide which drug candidates to fund?
They compare candidates as a portfolio rather than individually, weighing each one's phase-by-phase odds of success against the cost required to reach the next milestone and the expected payoff if it eventually succeeds. The goal is allocating limited resources across the strongest risk-adjusted bets.
Why do so many drug candidates fail?
Each candidate has to clear preclinical testing, three clinical trial phases, and regulatory review, and a failure at any single stage ends the effort. Attrition rates compound across these stages, which is why overall success rates from first-in-human trials to approval remain low across most therapeutic areas.
Does every therapeutic area carry the same level of risk?
No. Success rates vary meaningfully by disease area and development phase, based on aggregated historical trial data. This is part of why portfolio decision models weigh candidates individually rather than applying one blanket probability to every drug in development.
Can a decision model actually predict whether a specific drug will succeed?
No model can predict an individual outcome with certainty. What it can do is compare the relative odds and financial risk across a portfolio of candidates, helping a company allocate funding toward the strongest risk-adjusted bets rather than the most exciting-sounding one.