The Human Bottleneck

Drug discovery is being commoditised; human truth is the new scarce resource
Phase-0’s leverage isn’t de-risking - it’s surfacing (and fixing) human delivery/exposure constraints early enough to change efficacy
The bottleneck in pharma is clinical learning speed, not idea generation - Phase-0 is the highest-ROI “human check” to collapse uncertainty fast
The investable opportunity is a platform: standardised, decentralised execution + instrumented analytics + a compounding PK/PD dataset flywheel
None of it matters without decision discipline: pre-committed thresholds and action paths that make “stop/prioritise/progress” non-negotiable

In October 2025, HealthPad published a Commentary titled, “Phase-0 Goes Mainstream.” The reaction was immediate - and strategically revealing. The debate was not whether Phase-0 matters. It was about two sharper questions.

First: what does a Phase-0 “microdose” strategy look like when it does more than de-risk - when it materially improves downstream outcomes by collapsing uncertainty in molecule selection early enough to change which candidate is taken forward?

Second: what must be true for Phase-0 to become a real investment category - not a niche service line, but a compounding, defensible capability?

These questions land because the ground has shifted. Targets and hypotheses are no longer scarce. We are industrialising discovery - and commoditising parts of it. The scarce resource is human truth: early, high-signal evidence that a candidate reaches the right tissue, achieves sufficient exposure, engages the target, and produces the intended biology at a dose people can tolerate.

In plain terms, the question is no longer “does it bind?” It is: “does it work in a body that matters - and why?”

That tension defines modern drug development. Timelines remain stubbornly long, and costs are dominated by failure - not because teams lack intelligence or effort, but because preclinical plausibility does not reliably translate into clinical benefit. We can be right in vitro, compelling in animals, and still wrong where it counts. As Teslo and Scannell and others have argued, the true bottleneck is not idea generation; it is clinical development - the only stage that produces evidence regulators, investors, and patients accept.

This is where Phase-0 changes status.

Properly conceived, Phase-0 is not “a smaller Phase I.” It is an early, information-dense human experiment - often using microdoses or tightly limited exposure in a small number of participants - designed to answer a narrow but decisive set of questions:

Does the drug reach the right place in the human body?
At what concentrations, and with what distribution?
Is there early evidence of target engagement or pharmacology?
Are the exposures required for biological activity feasible in practice?

The goal is not to treat disease at that moment. The goal is to compress learning about delivery, distribution, exposure, and early biology into the earliest possible window - when decisions can still change outcomes.

Done well, Phase-0 does not just reduce uncertainty. It can change the trajectory of efficacy by revealing the constraint early - and making that constraint actionable. Often the hidden failure mode is not the target or the molecule in theory; it is what happens after dosing: insufficient exposure, wrong tissue distribution, unexpected metabolism, or a delivery problem that no animal model reliably predicts. Phase-0 is the fastest way to surface those truths - and to iterate while the programme still has room to move.

That is where the investment thesis becomes coherent.

Phase-0 becomes investable when it is more than bespoke studies sold one-by-one. It becomes investable when it behaves like a repeatable learning system: standardised protocols, fast cycle times, robust instrumentation and analytics, and a growing proprietary dataset that improves decisions over time.

In that world, Phase-0 is not just a risk filter. It is a value-creation engine - converting early human studies into decision-grade evidence with compounding returns: better capital allocation, fewer late failures, and - most importantly - a higher probability that programmes are engineered to work in humans, not just in models.

In This Commentary

This Commentary has one purpose: to make the Phase-0 opportunity legible by answering a simple question raised by HealthPad’s earlier piece: What does a Phase-0 strategy look like when it is not just a de-risking step, but a commercially decisive way to collapse uncertainty in molecule selection and improve the odds of downstream clinical success? It sets out what a credible Phase-0 “play” must include: the core capabilities, operating model, unit economics, and data flywheel required to build a repeatable human-signal engine - one that generates early, decision-grade evidence on exposure, delivery, and biological engagement, and converts it rapidly into clear action. Executed well, Phase-0 shortens iteration cycles, safeguards scarce clinical capacity, and compounds learning across a portfolio - turning “human truth” into an institutional capability rather than a downstream bottleneck, and into an investable advantage. To make this concrete, the argument is built around a strategic roadmap:

1. Make Phase-0 clinically consequential (not performative): design it to answer the questions that determine whether efficacy is plausible in humans.

2. Make it operationally routine: remove fixed overhead so “small, fast, high signal” is achievable repeatedly, not occasionally.
3. Make it clinically productive: use early human data to identify and fix delivery/exposure constraints while the programme can still change form.

4. Make it commercially scalable: standardise workflows, build repeat customers, and convert each study into a compounding dataset and defensible operating advantage.

5. Make decisions non-negotiable: pre-commit to action paths so Phase-0 outcomes reliably shape portfolio behaviour.

The Paradox: Scientific Acceleration, Clinical Deceleration

Discovery is accelerating at a rate few R&D leaders imagined a decade ago. We can read biology more cheaply, generate candidates faster, and iterate designs with something close to an engineering cadence. Yet the moment a programme crosses into humans, progress slows to a crawl.

Clinical throughput - the rate at which we convert hypotheses into reliable human evidence - remains slow, administratively heavy, capacity-constrained, and brutally expensive.

That mismatch is not a footnote. It is the operating constraint of modern drug development, and a primary reason R&D productivity remains uneven, often captured by Eroom’s Law. Portfolio-level failure follows a predictable pattern: organisations get better at producing “promising” assets while the clinic remains rate-limiting - and uncertainty accrues interest until it becomes catastrophic in Phase II and Phase III.

For healthcare systems, the consequences are tangible: trials that arrive late, oversized, and under-instrumented for learning; operational burden that competes with care delivery; and finite clinical capacity consumed by programmes that should have stopped earlier.

For investors, the consequence is structural capital inefficiency: long cycles, binary readouts, and value inflection points pushed years downstream. The cost is not only failure. It is time spent being wrong, and the compounding opportunity cost of being wrong at scale.

Two realities dominate drug R&D economics:

Attrition is structural: most programmes fail in humans, regardless of how compelling preclinical results look.
Returns are heavy-tailed: a small number of winners drive most patient benefit and commercial value.

In a heavy-tailed world, you do not win by perfecting narratives. You win by taking more credible shots - and by building a system that produces earlier, cleaner signals about what deserves the next tranche of capital, time, and patient exposure.

And there is only one source of those signals: structured learning in humans.

Medical misinformation isn’t new, but today it scales at speed. This episode of HealthPadTalks, When Medical Misinformation Becomes a Public Health Crisis, tracks the shift from fringe vaccine resistance to algorithm-amplified mythmaking, and how institutional failures turn mistrust into harm. From the UK infected blood scandal to the US opioid crisis, we unpack what broke, and what must change.

The Seduction of the Map

Modern biopharma has a recurring risk: confusing the map for the world. A persuasive mechanism, a clean pathway diagram, or a compelling computational model can start to feel like proof - especially when those stories help raise capital and align teams.

But biology does not negotiate with narratives. Many valuable medicines were not born from mechanistic certainty; they were discovered, improved, and positioned through iterative contact with human data. Clinical research is not the “final exam” at the end of a linear pipeline. It is an evolutionary engine: candidates meet real-world human variation, and only those that produce meaningful effects at tolerable doses survive.

GLP-1 medicines (a class of drugs that help regulate appetite and blood sugar) illustrate this pattern. Early human studies produced clear, decision-worthy signals. What followed was not certainty, but optimisation: dose finding, delivery improvements, and side-effect mitigation so more people could stay on treatment. The scientific explanation expanded and sharpened as human exposure accumulated.

The lesson is both warning and strategy: do not confuse plausibility with proof. Build systems that pull human feedback earlier and more routinely.

Phase-0: The Highest-Leverage Human Check

When leaders hear “run more trials,” it often triggers the wrong reflex: cost panic, risk control, and a retreat into bigger preclinical packages - as if more assays can substitute for human evidence.

But the strategic case is not for larger, slower late-stage programmes. It is for earlier learning: small, fast, high-signal experiments in humans that collapse the uncertainties that drive failure before you place a nine-figure bet.

That is the leverage of Phase-0 when executed with discipline. It is the highest-ROI human check you can run because it tells you whether the programme is playing the right game.

At its best, Phase-0 is a focused decision instrument:

Microdosing where appropriate (to study distribution/exposure with little pharmacological risk),
measurement of human exposure through pharmacokinetics (PK),
and where feasible, evidence of target engagement or pharmacodynamic (PD) effect.

The goal is not to prove efficacy. It is to answer a handful of narrow, high-leverage questions that determine whether benefit is plausible:

Is human exposure aligned with expectations, or is translation already breaking?
Are required exposures feasible and tolerable, or does the margin vanish the moment you dose a person?
Can the drug reach relevant tissue and engage the intended biology in humans at practical doses?

These are not academic curiosities. They are the fault lines along which programmes fail expensively later.

It is just as important to state what Phase-0 is not. It does not establish clinical efficacy. It does not, by itself, validate a target. It does not magically “de-risk Phase II biology.” What it does – strategically - is reduce the chance you spend years and tens of millions learning something you could have learned in weeks.

In a world where most drug candidates fail, the most valuable early trial is often the one that tells you to stop - quickly, clearly, and for the right reasons. That is not pessimism. It is portfolio hygiene.

So why is Phase-0 not routine? Because traditional clinical operations impose large, fixed overheads even on small studies. Site bottlenecks, start-up bureaucracy, contracting and monitoring, complex sampling logistics, and slow data reconciliation can turn a modest human check into a months-long project - costly and brittle - which defeats the point.

This is where decentralisation matters - not as a scientific shortcut, but as an operational unlock: remove friction, preserve rigour, and make early human learning fast enough and repeatable enough to become standard capability, not occasional luxury.

What Decentralised Phase-0 Buys

Separate two kinds of value that are often blurred:

Operational value: speed, access, repeatability, lower fixed overhead
Scientific value: decision-grade evidence - which must be earned by design

Decentralisation buys the operational side: remote pre-screening, eConsent, participant-centric scheduling, local or home-based procedures where appropriate, mobile visits where needed, and reserving specialist sites for what truly requires them.

You might also like:

Phase-0 Goes Mainstream

But speed is not truth. A study can run quickly and still produce weak data if endpoints are ill-chosen, assays are not validated, chain-of-custody is sloppy, or sampling is mis-specified. The platform thesis is not that logistics magically create insight. It is that repeatable infrastructure removes friction so teams can run good studies more consistently - and can afford to be disciplined about what each study is meant to resolve.

For readers new to decentralised trials, the intuition is straightforward: Phase-0 studies are small by design. They do not need the same site footprint as large efficacy trials. Yet traditional trial infrastructure imposes “fixed costs” that dominate small studies. Decentralisation converts those fixed burdens into scalable workflows:

participants are screened and consented remotely,
sampling is scheduled around participants rather than site calendars,
routine procedures move closer to the participant,
data capture and reconciliation are digitised end-to-end,
site time is reserved for what must be done at specialised centres.

This is not about lowering standards. It is about making high standards routine.

The Clinical Opportunity: Phase-0 as an Efficacy Engine, Not Just a Filter

The most important misunderstanding about Phase-0 is that it is “just de-risking.” That framing is too narrow.

Many programmes fail not because the target is wrong, but because the medicine cannot reliably achieve the right exposure in the right tissue at a tolerable dose and feasible delivery route. Preclinical models often miss practical human constraints: absorption variability, tissue penetration, metabolism, formulation limits, drug-drug interactions, transporter effects, unexpected clearance.

In short: the molecule may be conceptually elegant, but human delivery physics breaks the story.

Phase-0 enables a different posture: learn the constraint early, then engineer around it while you still can.

Clinical value emerges when Phase-0 is used to do three things:

Reveal the bottleneck. Is the limiting factor exposure, distribution, metabolism, or engagement? Even small studies can indicate whether human PK aligns with expectations and whether variability is manageable.
Convert bottlenecks into design choices. Once visible, constraints become actionable: formulation changes, prodrugs, delivery route redesign, depot strategies, combinations, dose scheduling, or patient stratification. The goal is not to confirm the original plan. It is to make a better one.
Protect the path to efficacy. Early human evidence improves the odds that Phase I/II programmes are properly dosed, properly instrumented, and not set up to fail.

In this sense, Phase-0 can be clinically creative. It can prevent the common tragedy where a medicine that could have worked is abandoned because early clinical execution was built on the wrong assumptions about human delivery.

What Makes Phase-0 an Investable Opportunity

If Phase-0 remains a one-off service - bespoke studies executed on demand - it remains a narrow market. The investable opportunity is the platform: repeatable unit economics with compounding advantage.

A decentralised Phase-0 platform creates commercial value in three ways.

1. It removes the “start-up tax.” Early studies are still treated as custom projects: assemble teams, pick sites, renegotiate contracts, bolt vendors together, unwind it all at the end. Every programme pays the same overhead before a single participant is dosed. Platforms standardise what should be standard: contracts, quality systems, audit-ready workflows, lab logistics, chain-of-custody, data integrity, and reporting. The molecule is bespoke. The operating system is not.

2. It turns execution into a reusable asset. Each study improves the system: SOPs, cycle time, monitoring, data pipelines, and decision playbooks. Over time, execution becomes not only faster, but more reliable. Reliability is commercial: sponsors return to the system that delivers decision-grade evidence without drama.

3. It builds a proprietary “human truth” dataset. The defensible moat is not “we can run a study.” It is “we can interpret and act on early human evidence better than others because we have seen more of it - cleanly, comparably, and at known quality.” A growing dataset of early human PK/PD patterns, operational benchmarks, assay performance, and design outcomes becomes a durable decision advantage.

This is the compounding loop investors should care about:
More studies → more proprietary, comparable human data → better design and triage → better sponsor outcomes → more repeat business → more studies.

Why AI Won’t Replace Human Trials - and Why That’s the Strategy

AI will improve drug development. It will not remove the need to test in humans. Therapeutic benefit is not a pure prediction problem. The path from “binds a target” to “helps a person” is shaped by adaptive biology, evolving disease, and human variability that cannot be fully modelled in advance.

This is not bad news for AI. It is strategic clarity. AI’s defensible role is not as an oracle, but as a force multiplier that makes human learning faster, cleaner, and cheaper.

In a Phase-0 platform, AI’s highest value is instrumental:

strengthening design by selecting informative timepoints and sampling schedules within practical constraints,
reducing overhead by automating reconciliation, monitoring, and reporting work that consumes coordinators and monitors,
protecting data integrity by flagging anomalies early - missing samples, timing errors, protocol drift - before datasets become unusable,
supporting decisions by surfacing patterns without false certainty: what the evidence suggests, what it does not, and what closes the loop next.

Used this way, AI increase’s reliability, reduces avoidable noise, and compresses cycle time - concentrating spend on programmes with credible human signal.

The prize is not AI that claims authority over biology. The prize is an AI-enabled decentralised Phase-0 capability that repeatedly converts uncertainty into decision-grade evidence earlier in the portfolio, at lower cost, with less burden on sites and participants - so patient benefit and capital efficiency improve together.

You might also like to listen to:

Diversification Is a Trap

The Hidden Constraint: Decision Culture

Phase-0 only creates value if organisations are prepared to act on what it shows. Many companies do not fail because they lack data. They fail because decisions become sticky: sunk cost, narrative commitment, internal momentum, and the default choice of “not yet.”

In that environment, Phase-0 can degrade into a checkbox: a quick study followed by slow rationalisation. The fix is governance by design:

define the decision question up front: what uncertainty is this Phase-0 check meant to retire?
where feasible, pre-commit to thresholds and action paths: what would “stop”, “prioritise”, or “progress” look like?
align incentives so disciplined stopping is treated as progress, not failure
instrument the study to produce a decision, not a report

A platform can widen the aperture of human learning. Only decision discipline makes that learning consequential.

Ethics and Regulation: Don’t Fight It - Instrument It

Any argument for more human trials must earn ethical legitimacy. “More” cannot mean more burden, more opacity, or lower standards. The goal is better experiments undertaken earlier - with clearer purpose, stronger protections, and more participant agency.

Done properly, decentralisation can strengthen ethics: less travel burden, broader access, participant-centric scheduling, real-time safety monitoring, and auditable consent. But trust must be designed in: privacy, secure bio-sample handling, chain-of-custody, endpoint integrity, and clear governance for secondary data use.

The strategic move is not to evade regulation. Medicines win on credible evidence. The play is to outperform within regulation by making strong evidence cheaper and earlier - instrumenting compliance so quality happens by default.

Takeaways: A Roadmap to Clinical and Commercial Success

Drug development is no longer constrained by imagination. It is constrained by human learning - how quickly and cleanly we can convert plausible mechanisms into decision-grade evidence in people. We made discovery cheap and scalable, then acted surprised when the clinic became the choke point. The predictable result is bloated portfolios, uncertainty carried too far downstream, and patient capacity, clinical bandwidth, and capital spent answering questions that should have been resolved earlier.

Phase-0 is the highest-leverage countermeasure - not because it proves efficacy, but because it resolves the translational uncertainties that decide a programme’s fate: exposure, feasibility, and early engagement in humans. It is underused for a reason: traditional operations impose large, fixed overheads even on small studies, stripping Phase-0 of its strategic advantage - speed. Phase-0 pays only when it stays small, fast, high-signal, and leadership has the discipline to act on the result, including the hardest call: stop.

That is why clinically serious, properly governed, AI-enabled decentralised Phase-0 platforms are not a “nice innovation.” They are a structural upgrade. They:

cut the start-up tax that makes early studies slow,
broaden access beyond narrow site bottlenecks,
protect measurement integrity in real time,
and make early human experimentation repeatable rather than bespoke.

In this model, AI is neither the product nor an oracle. It is the force multiplier that makes the learning engine reliable: tightening designs, enforcing quality, accelerating review, catching deviations early, and stripping operational waste so small studies can stay small - and decisions can stay timely.

The provocation is straightforward:

If you care about patients, you should want more early human learning, not less - because the most ethical trial is often the one that ends a weak programme quickly and redirects resources to something that can help.
If you care about ROI, you should want the same thing - because the edge comes from collapsing uncertainty sooner, taking more credible shots, and concentrating resources on real human signal rather than preclinical stories.

Done well, an AI-enabled decentralised Phase-0 platform creates rare alignment: patients get better-targeted medicines sooner, and investors back a system that wastes less time being wrong - while finding winners faster.

Comments

comments

Author

Details

Tags

Sponsored

The Human Bottleneck

Comments