The Missing Price Tag: What a Trusted Rating for Drug Assets Would Do to Biotech
In 1909, John Moody published a book that graded American railroad bonds with letters. Railroads were the technology boom of their age, and they were financed by a market that ran on reputation, rumor and whoever had the loudest promoter. Most investors could not tell a sound line from a doomed one, so most of them stayed out.
Moody’s letters didn’t make railroads safer. They made railroad risk legible. An insurance company in Hartford could now compare a bond from Kansas with one from Ohio without sending an engineer to walk the track. Once risk could be compared, it could be priced. Once it could be priced, generalist money could own it. The bond market we know today, where pension funds and insurers hold most of the paper, grew out of that one idea: a shared, independent language of risk.
Biotech is still in the pre-1909 era.
The asset nobody outside the room can price
A drug candidate is one of the most valuable and least legible assets in the economy. Roughly one in ten programs that enter human trials ends up approved. The other nine fail for reasons that are often visible in hindsight: a toxic chemical motif, a target that never mattered in the disease, a dosing problem nobody wanted to look at.
Today, judging which is which happens inside a small number of rooms. A pharma business-development team, a specialist biotech fund, a few dozen experienced partners. Each runs its own diligence, in its own format, reaching conclusions that cannot be compared with anyone else’s. Nothing accumulates. Two funds can look at the same asset and reach opposite answers, and neither has to show its work.
That has a cost beyond bad bets. If only specialists can price drug risk, only specialists can fund it. Biotech ends up financed almost entirely by a few hundred venture firms and the balance sheets of large pharma. When that pool tightens, as it did after 2021, good programs die for lack of money rather than lack of merit.
Now imagine a rating for drug assets that people actually trust. It would need to be independent of the money, published before outcomes are known so its record can be checked, and consistent enough that a grade on one asset means the same as a grade on another. Here is what that would change.
1. Generalist capital can finally come in
Pension funds, insurers and sovereign funds manage tens of trillions of dollars. They are not allowed, and are not equipped, to make bets they cannot benchmark. Their investment committees need a number they can compare, a methodology they can explain to trustees, and a track record they can audit.
A trusted rating gives them all three. A portfolio of forty rated Phase 2 assets becomes something an investment committee can model: expected losses, correlation across therapeutic areas, return under stress. Trial outcomes are also largely independent of the economic cycle, which is exactly what a pension fund looks for when it diversifies.
That is the precondition for structures that already exist on paper. Andrew Lo and colleagues proposed financing large, diversified portfolios of drug programs with securitized debt more than a decade ago. The idea stalled for the reason every structured product stalls: nobody could agree on the probability of success of the underlying assets. Fix that, and the “megafund” stops being a thought experiment.
Even a small allocation matters. One percent of global pension assets is many times what venture capital invests in biotech in a year.
2. Lenders can lend against pipelines
Biotech runs mostly on equity, the most expensive money there is. Founders give away a large share of the company to fund each trial, because banks cannot underwrite a loan against an asset whose value hinges on a readout they cannot assess. The venture debt that does exist comes from a few specialist lenders who price in their own uncertainty.
A rating turns a pipeline into something closer to collateral. A lender can set covenants against a rated asset, price the loan to its risk and syndicate it to other lenders who never met the company. Cheaper capital means founders keep more of their companies, and programs that are good but unglamorous can be financed without a headline-grabbing equity round.
The same logic reaches royalty buyers, who already pay cash today for a share of future drug sales. With a comparable rating, they could move earlier in development, where the money is needed most.
3. Landlords can diligence their tenants
This one sounds small but isn’t. Life-science real estate is a big asset class: wet labs, cleanrooms and vivariums built at great cost and leased to companies that may not exist in three years. After the post-2021 funding crunch, lab vacancy climbed and landlords learned the hard way that a tenant’s lease is only as good as its next readout.
Today a landlord sees cash runway and investor names. With rated assets, it can see the real driver of credit risk: how likely the tenant’s lead program is to succeed, and when the decisive data will land. A company with a highly rated asset and eighteen months of cash is a better tenant than one with more cash and a weak lead program, and a landlord could finally tell the difference.
4. Hospitals and trial sites can choose their trials
Hospitals and academic centers are the factories of clinical development. They recruit the patients, staff the protocols and absorb the operational risk. When a small sponsor runs out of money mid-study, the site is left with stranded work and, worse, patients whose treatment stops.
A rating lets sites diligence a trial before committing beds, coordinators and patient trust to it. A health system weighing two oncology studies could favor the one with stronger evidence behind it, and a site taking over a study from a failing sponsor could judge whether it is worth rescuing. Over time, that steers scarce clinical capacity, and scarce patients, toward the programs most likely to help them.
5. Shelved drugs get a second life
Large pharma companies sit on thousands of shelved programs. Most were not abandoned because they failed. They lost an internal portfolio review, or a merger made them redundant, or the strategy changed. Universities have the mirror-image problem: their technology-transfer offices hold promising early assets they have no way to value, so many never get licensed.
Some investors already make a business of rescuing these assets. They read old data packages by hand and bet on the ones that were shelved for the wrong reasons. A trusted rating would turn that craft into a market. A shelved asset with a strong rating is an asset someone will pay for, and a university that can show a credible rating can license on better terms. Drugs that would have gathered dust reach patients instead.
6. Borders stop mattering
A growing share of the world’s drug candidates now comes from outside the US and Europe, especially from China, Korea and Japan, and a growing share of licensing deals crosses those borders. Every one of those deals starts with the same problem: the buyer has to trust data generated in another system, often in another language, by a company it barely knows.
A shared rating is a common language. An asset from Shanghai becomes as comparable as one from Boston, and a seller in Seoul can show a Western buyer a number both sides already understand. That speeds up deals, and it widens the pool of assets the global market will seriously consider.
7. Accounting, tax and audit get real numbers
This is the quietest effect and possibly the largest market. Every acquisition of a drug company has to put a fair value on programs still in development, and every year those values are tested for impairment. Companies that move drug patents between their own subsidiaries need valuations that tax authorities will accept. Private biotechs need defensible valuations for employee stock options.
Today almost all of this rests on generic industry success rates: the average Phase 2 oncology drug, applied to a specific molecule. Auditors know it is crude, and tax disputes often turn on exactly these assumptions. An asset-specific, independently produced probability of success would give accountants, auditors and tax authorities something better to argue from.
8. Fewer patients in doomed trials
Every failed Phase 3 trial enrolled hundreds or thousands of patients who hoped it would work, and most early programs consume animals before they consume people. When a program was predictably doomed, that is not just wasted money. It is an ethical cost.
A trusted rating makes it easier to stop earlier. A founder who sees an outside view before the next $50 million, or a board that can point to an independent assessment, can kill a program before it reaches patients. Fewer people spend months of their lives on drugs that were never going to help them.
9. The ripple effects
Once one standard exists, every other participant can lean on it:
- Public markets. Small-cap biotech stocks swing wildly around readouts, and insiders have an information edge over everyone else. A public rating narrows that edge and gives ordinary investors a fairer starting point.
- Public and nonprofit funders. Agencies such as NIH, ARPA-H and BARDA, and patient foundations that fund drug development directly, can use ratings to decide which programs to back.
- Payers. Insurers and health systems can see years ahead which expensive drugs are likely to reach the market and plan their budgets accordingly.
- Insurers can write cover for trial programs and for the sponsors that depend on them.
- Acquirers and licensors get a common starting point for negotiating, so deals close faster with fewer arguments about whose model is right.
- Talent. Scientists deciding which startup to join, and what their stock options are worth, finally have something to go on.
- Patients and policymakers get a public track record of which kinds of programs work, accumulated year after year instead of buried in private data rooms.
10. Turning AI’s promise into cures
The leaders of AI drug discovery are not thinking small. Isomorphic Labs describes its mission as solving all disease, and Demis Hassabis has said that ending disease with the help of AI is within reach, “maybe within the next decade or so.” Behind the ambition is real progress. Models now predict protein structures that took years to solve experimentally, design molecules against targets once thought undruggable, and run through chemical space faster than any team of chemists. An analysis of the first wave of AI-discovered molecules found that 80 to 90 percent passed Phase 1 safety trials, well above the historical average.
But discovery was never the main bottleneck, and AI makes that obvious. Models can now propose credible candidates far faster than the industry can afford to test them. Every one of them still has to go through years of clinical trials costing hundreds of millions of dollars, and the same analysis found that AI-discovered molecules succeeded about 40 percent of the time in Phase 2, where efficacy is tested, roughly the same as conventional drugs. Making molecules is getting cheap. Deciding which ones deserve a trial, and finding the money to run it, is not.
That is where the promise stalls without a rating. When candidates were scarce, a handful of expert rooms could keep up. When candidates are abundant, the same rooms become the choke point, and the capital behind them stays roughly as limited as before. Investors who cannot tell a strong AI-designed molecule from a weak one either back the best story or back nothing. Neither gets the strongest candidates into patients.
A trusted rating changes the economics on both sides of the bottleneck. On the decision side, it lets the field filter thousands of candidates consistently and send the best to the clinic. On the capital side, it brings in the pension funds, insurers and lenders described above, so the number of trials the world can afford grows with the number of good candidates rather than staying fixed.
It also closes a loop that AI discovery badly needs. Discovery models today mostly learn from lab assays, not from clinical outcomes, because that knowledge is scattered across private data rooms. A rating that publishes its predictions and its hit rate, year after year, becomes a shared record of what actually works in people. That is precisely the signal the next generation of discovery models needs to learn from.
AI may well give us the molecules to cure most diseases. Turning them into medicines depends on something more mundane: knowing which ones to bet on, and having enough money to place the bets.
What could go wrong
Ratings can fail, and the last time they failed it nearly broke the financial system. A rating for drug assets would bring its own risks, and the honest case for one has to name them.
- Herding. If money flows to whatever rates well, it will crowd into mechanisms and modalities that have worked before. Truly new biology has no precedent to learn from, so a rating may score it unfairly low. The biggest danger of a drug rating is that it funds the tenth drug in a proven class and starves the first drug in a new one.
- Self-fulfilling grades. A low rating means no money, and no money means the program fails, whether or not the science was sound. A rating that becomes a gatekeeper can manufacture its own track record.
- Gaming. Once a number unlocks capital, people optimize for the number. Companies may choose indications, endpoints or trial designs to score well rather than to help patients.
- Forced selling. If funds are allowed to hold only assets above a certain grade, a downgrade forces them all to sell at once. That is exactly how bond downgrades turned into fire sales in 2008.
- Conflicts of interest. Before 2008, agencies paid by the issuers of mortgage securities stamped AAA on paper that collapsed. A drug rater paid by the companies it rates faces the same temptation.
None of these is a reason not to build one. They are design requirements:
- Independence. The rater cannot own the assets it rates or be paid more for a better grade.
- A public, timestamped record. Every rating issued before the outcome is known, and kept, so anyone can check the hit rate. A rating agency is only as good as the predictions it cannot quietly delete.
- Honest uncertainty. A calibrated probability with its error bars, not a letter grade that hides how little anyone knows about an early asset, and explicit flags when an asset is too novel to judge well.
- Competition. More than one credible rater, so no single model decides what gets funded.
None of this requires anyone to be perfect at predicting drug outcomes. Moody’s wasn’t perfect about railroads. It requires being better than the opinions available today, consistent enough to compare, and honest enough to be checked.
The prize
Biotech does not have a science problem as much as a capital-allocation problem. Discovery is getting cheaper and faster, increasingly with AI, and the number of candidates is exploding. The money and judgment needed to decide which to carry forward have not kept up.
A trusted rating would not create good drugs. It would make it possible for far more of the world’s capital to fund them, and to send it to the ones most likely to work. A century ago, a shared language of risk turned railroads from a speculator’s game into the backbone of institutional finance. Drug development deserves the same, and the patients waiting at the end of the pipeline need it more.
Disclosure: I’m co-founder and CTO of Saisho.ai, which is building a rating for drug assets.