The Death of the Grant Proposal: Why Scientists Spend 40% of Their Time Writing Instead of Researching

Getting your Trinity Audio player ready...

Scientists are trained to formulate hypotheses, conduct experiments, prove theorems, build instruments, and discover facts. Yet the modern research system increasingly requires them to perform another job: marketing unfunded ideas to grant committees.

The commonly repeated claim that scientists spend “40% of their time writing grants” requires qualification. A Nature survey found that academic researchers spent only about 40% of their working time conducting research—not that grant writing alone consumed the remaining 60%. Separate studies, however, show that grant preparation and administration can absorb an extraordinary share of researchers’ productive time. One analysis found that preparing a new proposal required an average of 38 working days of researcher effort, while another estimated approximately 116 principal-investigator hours per proposal.

The exact percentage varies by country, discipline, career stage, and funding system. The structural problem does not: scientists must repeatedly describe research they hope to perform, while research they have already performed may remain unfunded and unnoticed.

A better funding architecture would gradually replace proposal-centered competition with AI-assisted, retroactive evaluation of real scientific output. Researchers would receive funding after producing verifiable contributions—or through small preliminary payments followed by larger output-based rewards—rather than winning money primarily through persuasive forecasts.

Grant Proposals Turn Scientific Labor into a Competition in Persuasion

A conventional proposal asks researchers to predict several uncertain things:

• what they will discover;
• how long discovery will take;
• which methods will work;
• what measurable impact the research will produce;
• how the project will fit a funding agency’s current priorities.

This creates a fundamental mismatch. Genuine research investigates what is not yet known, while a successful proposal often needs to present the future as orderly, predictable, and administratively manageable.

The scientist is therefore rewarded not only for having a valuable idea, but for transforming uncertainty into a convincing narrative.

This skill is commonly called grantsmanship. It includes identifying fashionable terminology, interpreting a call’s institutional priorities, constructing deliverables, anticipating reviewers’ preferences, and presenting uncertain research as a plausible sequence of milestones.

Some planning is necessary. Researchers should explain their methods, budgets, safety requirements, and expected contribution. But when funding depends heavily on proposal rhetoric, the system begins selecting for the ability to describe future science, not necessarily the ability to produce important science.

The Hidden Cost of an Unsuccessful Proposal

Grant applications are expensive even when no money changes hands.

A study of Australian researchers examined 632 proposals and found a success rate of approximately 21%. A new proposal consumed an average of 38 working days across the research team, while a resubmission consumed 28 days.

These numbers expose the real economic cost of competitive grants.

Suppose five research teams each spend several weeks applying and only one receives funding. The agency records one funded project. It does not normally record the months of specialist labor consumed by the four unsuccessful applications.

That labor cannot be recovered. During those hours, researchers are not:

• collecting data;
• checking proofs;
• reproducing published results;
• maintaining scientific software;
• training junior researchers;
• improving laboratory methods;
• publishing negative findings;
• exploring unconventional questions.

The loss is especially serious because research labor is highly specialized. A month of a senior mathematician’s, chemist’s, or experimental physicist’s time cannot simply be replaced by hiring a general administrative worker.

The funding system is consuming precisely the scarce intellectual resource it claims to support.

Low Funding Rates Multiply the Waste

Proposal costs become more damaging when success rates are low.

For example, the US National Science Foundation reported a 27% overall funding rate for its Mathematical and Physical Sciences directorate in fiscal year 2024. Rates differed substantially by division: 18% in astronomy, 26% in mathematical sciences, and 45% in physics.

A low funding rate does not prove that peer review is useless. Funding agencies have finite budgets and must choose among competing projects. Nevertheless, when four or five credible teams prepare extensive submissions for one award, the aggregate application cost can become disproportionate to the value of the selection process.

More detailed proposals do not automatically solve the problem. Research on grant-writing effort found that the number of hours spent preparing an application was not clearly associated with whether it was funded.

This suggests a disturbing possibility: part of the extra work may improve presentation without materially improving selection accuracy.

The Proposal System Favors Institutional Insiders

Proposal-based funding does not affect every scientist equally.

Established researchers often possess:

• institutional grant offices;
• experienced co-investigators;
• preliminary data financed by earlier grants;
• reputational signals;
• professional networks;
• knowledge of reviewers’ expectations;
• paid time allocated to proposal development.

Independent researchers, small laboratories, researchers in poorer countries, and scholars pursuing unconventional subjects may possess none of these advantages.

A proposal may therefore measure two different things at once:

1. the potential value of the proposed research;
2. the applicant’s access to the infrastructure required to produce a competitive application.

These are not equivalent.

A polished application can be evidence of organizational competence. But it can also be evidence that the applicant already belongs to a well-funded institution. When funding follows the strongest proposal-writing infrastructure, existing inequality becomes self-reinforcing.

This is one reason AI-based scientific meritocracy should evaluate contributions using transparent, consistently applied criteria rather than relying primarily on institutional prestige or narrative polish.

Why Prediction Is a Poor Basis for Scientific Funding

Grant committees are asked to rank projects before the decisive evidence exists.

That is inherently difficult. Reviewers may assess whether a method is coherent, whether the team is qualified, and whether the question is important. They cannot reliably know which project will produce a breakthrough.

The safest proposal is often not the most transformative proposal. It is the one that resembles work the field already understands.

This can disadvantage:

• new fields without established reviewers;
• interdisciplinary research;
• negative or corrective work;
• long-term theoretical projects;
• replication studies;
• research conducted outside universities;
• ideas that challenge a discipline’s prevailing framework.

The result is a funding paradox: applicants need evidence that their idea will work before receiving the money needed to test whether it works.

Retroactive Funding Reverses the Logic

Retroactive research funding rewards demonstrated scientific contributions instead of relying exclusively on promises about future work.

Under this model, a researcher first produces an assessable output, such as:

• a paper or preprint;
• a mathematical proof;
• a curated dataset;
• reproducible experimental results;
• research software;
• a documented replication;
• an open protocol;
• a useful negative result;
• a review that identifies a major error;
• a scientific dependency used by later projects.

The funding system then evaluates the output and rewards its demonstrated or probable value.

Retroactive funding does not mean that researchers must personally finance expensive laboratories. A hybrid architecture can combine:

1. microgrants for initial work;
2. milestone payments for verifiable progress;
3. retroactive rewards for completed contributions;
4. continuing funding based on demonstrated reliability and impact.

This changes the central question from:

“Can this researcher persuade us that the proposed work will matter?”

to:

“What has this researcher produced, how reliable is it, and what other knowledge depends on it?”

Retroactive public-goods funding has already been tested in blockchain ecosystems. Optimism, for example, developed a mechanism intended to reward contributions after their positive impact could be assessed. This is not itself a scientific funding system, but it demonstrates the broader economic principle: valuable public goods can be rewarded retrospectively rather than sold in advance.

How AI Can Evaluate Real Scientific Output

Manual retroactive evaluation would still require substantial reviewer labor. Automated evaluation can reduce that burden, although it cannot responsibly eliminate human oversight.

An AI-assisted funding engine could examine multiple dimensions of scientific value:

Methodological quality

The system can check whether methods are clearly documented, whether conclusions follow from reported evidence, and whether important limitations are disclosed.

Reproducibility

AI can inspect the availability of code, data, protocols, computational environments, and intermediate results. It can also compare claims with attempted replications.

Originality

Semantic analysis can identify how a contribution differs from existing literature. This is more informative than counting keywords or accepting the author’s own claim of novelty.

Dependency and upstream value

Some contributions are important because many later projects depend on them. A dataset, proof, software library, classification system, or experimental technique may be upstream of numerous visible results.

AI-powered public-goods mechanisms are already exploring this type of dependency analysis. Gitcoin’s Deep Funding model, for example, uses machine learning to assess relationships among projects and identify upstream contributions.

Expert and community review

Post-publication reviews, replication reports, technical criticism, and expert assessments can be incorporated as evidence. The algorithm should evaluate reviewers as well as researchers, assigning greater weight to reviewers with a history of accurate and technically substantive judgments.

Public and scientific use

Scientific outputs can be assessed partly through their use in subsequent publications, government documents, technologies, educational resources, and other research. Research on science as a public good shows that downstream use can be measured across multiple domains rather than reduced to a single citation count.

No single metric should determine funding. Citation counts alone favor older work and established fields. Popular voting alone can reward visibility. AI summaries alone can contain errors. A credible system must combine independent signals and expose the reasoning behind allocations.

AI Evaluation Must Not Become Automated Bureaucracy

Replacing grant committees with an opaque model would not solve the underlying problem. It would merely encode the bureaucracy in software.

A responsible system requires:

• published evaluation criteria;
• auditable inputs and outputs;
• explanations for funding decisions;
• mechanisms for correction and appeal;
• protection against coordinated manipulation;
• detection of plagiarism and fabricated evidence;
• field-specific assessment standards;
• human review for high-value or disputed decisions;
• regular testing for institutional, linguistic, and geographic bias.

AI has no magical ability to define scientific value. Its advantage is narrower and more practical: it can process far more evidence than a conventional panel and apply the same declared procedures repeatedly.

The AI alignment and anti-gaming architecture proposed by World Science DAO is therefore essential. Automated funding must be designed to resist metric gaming, fabricated citations, review rings, superficial output inflation, and attempts to manipulate evaluation prompts.

What Happens to Research That Needs Advance Capital?

Retroactive funding has an obvious limitation: some research cannot begin without expensive equipment, materials, staff, clinical infrastructure, or long-term data collection.

The answer is not to abolish all prospective grants. It is to reduce their role and improve their design.

Advance funding can remain available when the applicant demonstrates that:

• the work cannot reasonably be self-started;
• the proposed cost is proportionate;
• the methods are technically credible;
• outputs will be openly documented where possible;
• payments can be divided into auditable stages.

Past verified output can then replace much of the speculative proposal. A researcher who has repeatedly delivered reliable work should not need to rewrite their intellectual biography for every application.

The system could use reputation as evidence, not status as authority. A strong record would increase access to advance capital, while completed outputs would continue to receive retroactive rewards.

From Proposal Writing to Scientific Production

The grant proposal will probably not disappear completely. Large, hazardous, capital-intensive, or coordinated projects require prospective planning.

But it should stop being the default gateway through which almost every serious researcher must pass.

The present system often rewards three activities:

1. predicting uncertain results;
2. adapting ideas to institutional priorities;
3. persuading a small group of reviewers.

An output-based system would instead reward:

1. producing useful knowledge;
2. making results verifiable;
3. contributing to other researchers’ work;
4. correcting errors;
5. solving neglected problems.

That transition would not merely save administrative time. It would change what scientists are incentivized to do.

Researchers would spend less time marketing possible discoveries and more time creating assessable ones. Independent scientists could compete through visible output rather than institutional affiliation. Negative results and infrastructure work could receive explicit value. Funding decisions could be continuously updated as new evidence appears.

The Grant Proposal Should Become the Exception

Scientific funding currently attempts to identify the future through documents written before the evidence exists. This guarantees uncertainty, encourages strategic language, and consumes enormous quantities of expert labor.

AI-assisted retroactive funding offers a different principle:

Fund scientific value where it can be observed, reward contributors after evidence appears, and use advance grants only where advance capital is genuinely necessary.

The purpose of science funding is not to produce excellent applications. It is to produce reliable knowledge.

World Science DAO’s proposed automated research-funding system aims to make output, dependency, reproducibility, and scientific usefulness central to allocation. Such a system should not be treated as an unquestionable machine judge. It should be an auditable mechanism through which evidence, expert review, and public funding can interact at a scale that traditional committees cannot manage.

The death of the grant proposal would not mean the death of scientific accountability. Properly designed, it would mean the opposite: accountability based less on promises and more on what researchers actually contribute.

Researchers who are already producing valuable work should not need to become professional marketers to survive. They should be able to document their results, submit them for transparent evaluation, and receive funding proportional to their demonstrated scientific value.

That is the transition from funding proposals to funding science.