Commercial proposal teams have an entire software category. Why do research proposals still appear to live largely in Word, SharePoint and the heads of PIs?
I came across a surprisingly difficult question while researching proposal automation.
I have worked at A*STAR in Singapore for around eight years and have been involved in multiple research proposals, including as a Principal Investigator. I have written proposals, coordinated technical contributors, interpreted calls, assembled supporting material, reviewed requirements and worked through the usual cycle of drafting and revision.
Yet until I started researching the commercial RFP market, I had never even heard of products such as Loopio or AutogenAI.
That initially seemed strange.
Commercial proposal teams now have an entire category of specialised software designed to analyse RFPs, retrieve previous content, generate drafts, coordinate contributors, maintain reusable knowledge and check compliance. AutogenAI even explicitly markets its platform for grant writing. (AutogenAI – Grant Writing)
Meanwhile, my own research environment has historically felt much more familiar:
Word or PowerPoint → files on local computers or SharePoint → emails and meetings → increasingly ChatGPT or Copilot.
This is only my personal experience inside one research organisation. It does not establish how every A*STAR institute operates, much less every university.
But after looking for comparable software adoption elsewhere, I no longer think the observation should simply be dismissed.
The more interesting question is:
Why did commercial proposal work develop a visible specialised software ecosystem, while the actual authoring of research proposals appears much less institutionalised as a software category?
I do not yet have a confident answer.
But several explanations I initially considered no longer survive scrutiny.
1. Research Proposals Are Not Obviously Less Reusable
My first instinct was that commercial RFP responses may simply contain much more reusable material.
A company responding repeatedly to customer questionnaires can reuse product descriptions, security answers, implementation methodologies, case studies, resumes, certifications and previous responses. Loopio's entire content-management model is built partly around maintaining material that can be reused across future responses. (Loopio Platform)
Perhaps research proposals are simply too novel for that model.
I could not find good evidence for that conclusion.
Universities themselves maintain reusable proposal material. The University of Alabama at Birmingham, for example, operates a Boilerplate Library containing institutional information intended to reduce repeated proposal-writing work. (UAB Boilerplate Library)
Anyone who has written several research proposals will also recognise the pattern. The scientific question changes, but many surrounding inputs may recur: institutional capabilities, facilities, team biographies, previous achievements, methodological descriptions, impact material, administrative information and parts of the broader research narrative.
That does not mean research proposals are 60% reusable, or 30%, or any other convenient number.
I do not have reliable comparative data measuring content reuse in commercial RFPs against research grants.
So I would now reject the easy explanation that commercial proposal software exists because commercial proposals are repetitive whereas research proposals are not.
The evidence is not there.
2. PIs Can Function Remarkably Like Proposal Managers
A second explanation initially sounded plausible: commercial organisations have dedicated proposal managers, whereas universities do not.
But that also fails to match reality particularly well.
A PI may not carry the title “Proposal Manager,” but consider what often happens around a substantial research proposal.
The PI may need to:
-
Interpret and structure the opportunity
- understand the call;
- determine what the funder is actually asking for;
- establish the work packages, objectives and proposal structure;
- identify eligibility, submission and evaluation requirements.
-
Coordinate the response
- recruit collaborators;
- obtain technical inputs;
- reconcile different scientific viewpoints;
- collect CVs, track records, budgets and institutional information;
- manage revisions and deadlines.
-
Develop and defend the proposal
- write or integrate major sections;
- ensure the research plan remains coherent;
- align the work with the funding programme;
- respond to reviewer or internal feedback;
- produce the final submission package.
That is not identical to commercial proposal management.
But functionally, there is substantial overlap.
The absence of a “Proposal Manager” job title therefore does not explain why specialised proposal-authoring software would be unfamiliar to someone who has repeatedly performed much of that work.
3. I Was Also Comparing The Wrong Software Categories
This became one of the most important corrections in my research.
Universities and research institutes absolutely do use specialised grant software.
But much of it belongs to a different category.
Research-administration systems concentrate heavily on activities such as applications, budgets, institutional approvals, ethics, compliance, routing, grant records and post-award administration. Singapore's Research Grants Portal similarly exists to support grant administration across participating public research funders. (Singapore Research Grants Portal)
That is valuable infrastructure.
It is not the same thing as software asking:
What does this call require?
Which previous institutional material is relevant?
What should the proposal structure look like?
Which sections are weak?
What evidence supports this claim?
Which requirement have we failed to answer?
What can be drafted from our existing material?
A Singapore PUB research-funding instruction provides a simple illustration of the separation. Researchers prepare the research proposal in Microsoft Word and then upload the required documents through the grant system. (PUB Open RFP Fact Sheet)
In other words:
The institutional system manages the grant. Word still manages much of the thinking and writing.
That distinction matters enormously.
Research administration software should not be counted as evidence that universities already have an AutogenAI equivalent.
4. Genuine Research-Proposal AI Does Exist
Once I applied that stricter definition, I did find real examples.
But there were fewer than I expected.
4.1. Colorado School Of Mines
The Colorado School of Mines Research Office invited researchers to beta-test Intellectible, an AI proposal-writing system.
This is a genuine match to the category I was looking for. The system can use materials such as previous proposals, publications and CVs alongside the funding solicitation, analyse the opportunity and help produce a draft. (Colorado School of Mines – Intellectible AI)
That is much closer to AutogenAI than to conventional grant administration.
4.2. Duke University
Duke developed an internal tool called Devil's Advocate.
Its role is somewhat different. Rather than primarily replacing the initial writer, it can evaluate research proposals against funding requirements, identify compliance issues, review citations and priorities, and perform forms of red-team or critical proposal review. (Duke OIT – Devil's Advocate)
Again, this is actual proposal intelligence—not routing a grant application through an approval chain.
4.3. William & Mary
William & Mary's Corporate & Foundation Relations operation has publicly been presented as a user of Instrumentl Apply, which can use historical proposal information and organisational inputs to help develop new grant responses. (Instrumentl – William & Mary)
The qualification is important: this example sits closer to institutional fundraising and foundation grants than to a PI developing a complex NSF-style scientific programme.
It nevertheless demonstrates that university grant-writing teams are experimenting with specialised authoring AI.
4.4. AutogenAI Itself
AutogenAI has published a case study describing deployment at a large seven-campus US public university.
The university reportedly used historical proposal material and institutional knowledge to generate outlines, accelerate first drafts and support reuse across grant-writing activity. (AutogenAI public-university case study)
This is perhaps the closest direct match to the question.
But the university is anonymous.
That makes the evidence useful while limiting what can be independently verified about the scale and institutional context of the deployment.
5. What Surprised Me Was What I Did Not Find
Before doing this research, I expected to find something closer to:
Stanford uses X.
Oxford uses Y.
ANU standardised on Z.
Major research universities routinely provide proposal AI to their researchers.
Instead, the public evidence I found was much more fragmented.
There are university pilots.
There are internally developed tools.
There are specialised teams using grant-writing AI.
There are vendor case studies.
There are individual researchers using advanced AI tools.
But within the research already completed for this investigation, I did not find strong evidence that specialised proposal-authoring software has become a normal, institution-wide layer of infrastructure across major US, UK or Australian research universities.
For Singapore, the evidence was even thinner.
That does not mean those deployments do not exist. Enterprise software adoption is often private, and absence from public search results is not proof of absence.
But it does change my prior assumption.
I had implicitly assumed that because commercial proposal AI is already a substantial market, research institutions must have developed something comparable behind the scenes.
I can no longer confidently make that assumption.
6. So Why Did These Markets Develop Differently?
This is where the investigation becomes more interesting—and much less certain.
Several hypotheses seem plausible.
6.1. Hypothesis 1: The Buyer And The User Are Too Far Apart
The PI experiences the authoring pain.
The institution may pay for the software.
Those are not necessarily the same decision-maker.
A research administration office can justify a platform because every grant must pass through institutional workflows. A tool that makes an individual PI 30% faster may have a less obvious institutional owner, even if the productivity gain is real.
I find this explanation plausible.
I do not yet have enough evidence to say it is the primary cause.
6.2. Hypothesis 2: Existing Tools Are Already “Good Enough”
Researchers already have Word, SharePoint, institutional repositories and increasingly general-purpose AI.
If ChatGPT or Copilot can summarise a funding call, improve writing and help generate first drafts, the incremental value of buying another specialist system may simply not be large enough.
This would mirror something I have seen elsewhere in proposal research: specialised software is not competing against doing nothing.
It is competing against:
existing workflow + general AI + human expertise.
That is a substantially harder economic benchmark.
6.3. Hypothesis 3: Institutional Adoption Costs Overwhelm Individual Productivity Gains
Research proposals can contain unpublished scientific ideas, collaborator information, intellectual property, sensitive technical material and other non-public information.
Introducing another external AI platform can therefore involve security review, data governance, procurement, legal review, authentication, integrations, support and training.
A tool might save a researcher ten hours and still fail the institutional cost-benefit calculation.
Again, plausible.
Not yet proven as the dominant explanation.
6.4. Hypothesis 4: Research Proposal Workflows Genuinely Differ In Ways Current Commercial Products Handle Poorly
This possibility should not be dismissed simply because PIs perform proposal-management functions.
A complex government-contracting RFP and a frontier scientific grant are not identical products.
Novel scientific reasoning, experimental design, literature positioning, preliminary data, investigator credibility and intellectual originality may occupy a much larger fraction of the value in some research proposals.
If those differences reduce the usefulness of reusable organisational knowledge and structured response workflows, commercial proposal software may translate poorly.
But this takes me back to the evidence problem.
I do not yet have comparative data showing where those workflows differ enough to change software economics.
That is exactly what would need to be investigated.
7. The More Interesting Market Question
I therefore think the useful question is no longer:
Can AI write research proposals?
Clearly, AI can already assist substantially with research proposal work.
Nor is the useful question:
Can universities buy grant software?
They already do.
The more interesting question is:
Why has proposal authoring itself not become the same kind of institutional software category in research that response management became in commercial bidding?
That distinction has implications beyond universities.
It may tell us something fundamental about where proposal software becomes economically valuable.
Perhaps specialist proposal systems emerge only when several conditions coincide:
- proposal activity is frequent enough;
- substantial knowledge is repeatedly reused;
- multiple contributors need coordination;
- mistakes have meaningful economic consequences;
- a recognisable organisational function owns the workflow;
- and the value created is large enough to justify changing established behaviour.
Commercial bid teams may satisfy those conditions more consistently.
Research institutions may satisfy only some of them—or may satisfy all of them while lacking the organisational structure needed to buy the solution.
I do not yet know.
8. What I Would Now Ask Research And Proposal Practitioners
This investigation has left me with questions I would rather have practitioners answer than resolve through speculation.
For researchers and research-development professionals:
- When preparing a major proposal, which parts are genuinely new intellectual work and which parts are recurring proposal operations?
- How much useful material is actually reused across grants, and how is that material found and validated today?
- Would a specialist proposal system materially outperform Word + institutional repositories + ChatGPT/Copilot? Where?
- Who inside the institution would actually own and pay for such a system?
- Have institutions already tried these products and rejected them—or has the category simply failed to reach most researchers?
For commercial proposal professionals, I think there is an equally interesting comparison.
What happened inside commercial bidding that eventually justified dedicated proposal managers, content libraries, response-management platforms and increasingly sophisticated AI?
And which of those conditions are missing from research?
9. My Current Conclusion
I started with what looked like an odd personal observation.
After years of writing research proposals, specialised proposal AI was effectively invisible to me.
I initially tried to explain that away by assuming research proposals were less reusable or that academia lacked proposal-management roles.
I no longer think either explanation is adequate.
There are genuine research-proposal AI products and genuine university deployments. But based on the evidence I have found so far, they still look more like an emerging collection of pilots, specialised teams, internal tools and early deployments than a universally recognised institutional software category.
That makes the absence itself interesting.
Perhaps research proposal authoring represents an underserved market.
Perhaps Word, general AI and human expertise are already economically good enough.
Perhaps institutional procurement and governance make the category unattractive.
Or perhaps research proposal work differs from commercial RFP response in ways that specialised software companies have not yet understood well enough.
At this stage, I would rather leave those possibilities open than force one into a convenient market thesis.
The question I now find most useful is therefore not whether AI can help researchers write proposals.
It clearly can.
It is why an activity that consumes so much highly qualified human attention has still not produced the same visible proposal-software ecosystem that emerged around commercial bidding—and whether practitioners believe that difference is accidental or fundamental.