What practitioner discussions changed about how I think proposals actually get won
I started this research by asking how much of the RFP and proposal process could be automated.
That question increasingly feels incomplete.
After spending more time reading discussions between proposal professionals, examining existing software, and looking at the kinds of proposal work companies still outsource, I have started to see a different hierarchy. AI can already perform a surprising amount of useful proposal work. But whether that work is actually valuable appears to depend much more heavily on the system surrounding the AI.
That system includes how knowledge is maintained, how requirements are decomposed, how drafts are developed, how assumptions are challenged, how previous performance is interpreted, how SMEs contribute, how compliance is checked, and ultimately how the team decides what story it is trying to tell.
My current hypothesis is therefore becoming much simpler:
The competitive advantage in proposal work may increasingly come from the quality of the proposal operating system, not the quality of the language model.
AI matters. But it may be one relatively small component inside a much larger professional discipline.
1. A Proposal Is Not A Writing Task
One of my mistakes earlier in this investigation was giving too much conceptual weight to what the AI could do.
Modern general-purpose systems such as ChatGPT, Claude and Copilot can already summarize solicitations, extract requirements, create initial compliance structures, draft capability statements, rewrite sections and produce reasonable first drafts. In one detailed discussion among government-contracting practitioners, people described using general AI for exactly these kinds of tasks. Several seemed perfectly comfortable using mainstream tools rather than buying specialised proposal AI for every operation. Reddit discussion: Anyone here using AI for proposals?
But experienced practitioners also described something much more important around those tools.
The work is not simply:
RFP → AI → proposal.
A serious response appears much closer to:
-
Understand and structure the opportunity
- extract requirements;
- build the compliance matrix;
- break the response into manageable sections or “shreds”;
- identify what information and evidence each section requires.
-
Build and challenge the response
- create the initial structure and first draft;
- develop individual sections;
- check whether each answer actually addresses the question;
- challenge unsupported assumptions;
- ensure terminology and claims remain consistent across sections.
-
Make the proposal competitive
- align with the company's strategy;
- select appropriate past performance;
- develop positioning and win themes;
- distinguish generic capability from what matters to this particular buyer;
- repeatedly review the proposal as one coherent argument rather than a collection of AI-generated answers.
AI can assist throughout that workflow.
But the workflow itself appears to be doing much of the heavy lifting.
A practitioner with a disciplined proposal system can switch between models and still produce useful work. A team with a weak process can give a sophisticated model hundreds of documents and still produce something generic, inconsistent or strategically irrelevant.
That distinction is becoming central to my research.
2. The Better The System, The Less Magical The AI Looks
The same Reddit discussion contained a recurring observation that I initially underweighted: AI becomes substantially more useful when the organisation already has good material behind it.
One participant described the useful role of AI as essentially a junior writer connected to the organisation's own library. Others emphasised that organised content, past performance, opportunity information and company documents determine what the model can meaningfully produce. One particularly useful observation was that, once the content library is organised, the problem becomes increasingly about the company's data rather than the particular AI tool. Reddit discussion
This helps explain something that otherwise looks contradictory.
On one hand, sophisticated proposal platforms already exist. Loopio, Responsive, GovDash and others can retrieve organisational content, generate responses, support review workflows and increasingly perform requirements and compliance work. Loopio Responsive GovDash
On the other hand, companies still hire freelancers for surprisingly basic work.
In previous research I found clients paying freelancers to research relevant opportunities, construct compliance matrices, coordinate proposal material and even proofread responses for wrong customer names, inconsistent numbers, incorrect dates, broken cross-references and unanswered requirements. RFP Proposal Proofreader – Upwork
Those jobs do not prove that AI cannot perform those tasks.
They show that having the capability available somewhere is different from having a dependable system that makes the capability happen correctly every time.
That may be the more interesting market problem.
3. Knowledge Management May Be The Hidden Infrastructure
The concept that now interests me most is knowledge management.
At first I interpreted this mainly as search: can the proposal team find the right previous answer, case study, CV, technical description or past-performance example?
That definition is much too shallow.
Suppose an RFP asks whether a system can meet a particular recovery-time commitment. Inside the bidder's organisation there may be an old proposal promising eight hours, a newer technical architecture suggesting four hours, a marketing deck saying “as little as two,” and a customer-specific implementation that achieved something different again.
Finding all four documents is not the hard part anymore.
The real questions are:
- Which information is still true?
- Which source is authoritative?
- Which statement applied only to a particular customer or product?
- Which information can safely be reused?
- Who owns the answer?
- What has changed since the last proposal?
- When must an SME make a new decision rather than reuse an old one?
That is a much more demanding problem than semantic search.
I now think of the pipeline as:
Documents → trusted organisational knowledge → opportunity context → proposal
The middle layers matter enormously.
This is not just an inference from AI discussions. APMP itself treats content-library maintenance as a serious professional discipline. Its August 2026 AI conference included a session titled “Learn from a Librarian: Tips for Creating Your AI Content Library,” explicitly focused on organising, governing and maintaining content so AI can use it effectively. APMP Winning AI in Proposals agenda
CAPTRUST provides another revealing example. Its proposal operation includes an RFP Knowledge Management function supporting numerous internal teams and SMEs, with responsibility for maintaining usable organisational knowledge and SME relationships. CAPTRUST case study
That tells me knowledge management is not merely a software feature.
For mature proposal organisations, it can be an operating function.
4. Knowledge Alone Still Does Not Win The Bid
Even excellent knowledge management does not solve the entire proposal problem.
A company may have beautifully organised SharePoint folders, current past-performance records, approved technical answers and reliable search. That can make drafting dramatically easier. It still does not automatically tell the team how to win.
Experienced practitioners repeatedly distinguish between producing an answer and building a competitive response.
What past performance should be emphasised? Which differentiators genuinely matter to this evaluator? What assumptions should the team challenge? What does the buyer appear to care about beyond literal compliance? Which weaknesses need to be mitigated? What should the evaluator remember after reading another five proposals?
These are questions of strategy and positioning.
Generic AI can help interrogate them. It can offer candidate themes, compare evidence, challenge a draft and act as another reviewer. But a company that has repeatedly competed in the same market may possess far stronger knowledge of its customers, competitors, strengths and historical winning patterns than an AI could infer from a solicitation alone.
This produces a useful hierarchy:
AI capability
sits underneath
organisational knowledge
which sits inside
proposal workflow and review discipline
which ultimately serves
competitive strategy.
The mistake would be to optimise the bottom layer while assuming everything above it takes care of itself.
5. The Small-Team Problem May Be Different
This creates an interesting contrast between mature proposal organisations and much smaller bidders.
Large organisations can justify dedicated proposal managers, content owners, knowledge-management processes and expensive specialist platforms. Their problem may increasingly be maintaining, governing and applying a huge amount of organisational knowledge correctly.
A company with five, ten or twenty employees has a different constraint.
The founder may also handle business development. A technical lead may become the proposal SME. Previous responses may live in folders rather than a carefully governed library. There may be no proposal manager and certainly no dedicated knowledge-management team.
Yet these companies can increasingly access surprisingly capable AI for almost nothing.
That leads to a hypothesis I now want to investigate:
Could small proposal teams gain a disproportionate advantage from better systems rather than better AI?
The useful intervention might not be another enterprise proposal platform. It may instead be a much smaller system that helps a team apply professional proposal discipline consistently.
For example:
-
Before drafting
- organise the solicitation and supporting inputs;
- extract requirements and create the working compliance structure;
- retrieve previous evidence and identify missing information.
-
During drafting and review
- keep sections tied to explicit requirements;
- flag claims that lack supporting evidence;
- detect inconsistent names, terminology and commitments;
- ask whether the response actually answered the evaluator's question;
- surface sections that need an SME rather than generating confident filler.
-
After submission
- capture newly approved knowledge;
- record which content should become reusable;
- preserve decisions and evidence so the next proposal does not start from zero.
None of these ideas is proven whitespace.
Existing platforms already address pieces of them, and a disciplined team can accomplish surprising amounts using SharePoint, spreadsheets and general AI.
The question is whether small teams still experience enough friction assembling these pieces into a repeatable operating system.
6. This Is Not Yet A Product Thesis
I am deliberately resisting the temptation to jump from “interesting problem” to “build software.”
My current research focus is moving toward companies with very small proposal organisations—perhaps firms with roughly 1–20 employees initially, while still looking up to around 50. That is not because I have proven they are the optimal customers. It is because they may provide a clearer environment in which proposal work is important but the company cannot solve every problem by hiring specialist proposal and knowledge-management staff.
A low-cost product—perhaps something on the order of $50–100 per month—could theoretically be more accessible than enterprise proposal software.
But that price is currently only a hypothesis.
I do not yet know whether small firms submit enough meaningful proposals to pay even that amount, whether the important workflows are sufficiently similar across companies, or whether the necessary customisation would destroy the economics.
That last issue worries me particularly.
A good proposal system is shaped by the company's products, customers, sector, internal approval structure, past performance and way of working. If every customer requires a different operating system, then the value may be real but difficult to convert into scalable software.
Perhaps that is why the strongest proposal platforms increasingly combine multiple layers—knowledge, workflow, opportunity intelligence, drafting, review and human collaboration—rather than treating AI generation as the product itself.
But I do not yet have enough evidence to make a stronger claim.
7. The Question I Now Want Practitioners To Break
My earlier question was:
How much of proposal work can AI automate?
I no longer think that is the most useful framing.
I am more interested in:
What system allows a proposal team to repeatedly turn organisational knowledge and professional judgment into a competitive, compliant response—and which parts of that system remain unnecessarily difficult for smaller teams?
Several questions now seem particularly important:
- When you reuse previous material, how do you know it is still authoritative and applicable to this opportunity?
- Where does your proposal workflow rely on professional discipline that software does not enforce well today?
- Which review steps genuinely improve the chance of winning, versus merely correcting avoidable operational mistakes?
- If a small team had perfect search and good general AI tomorrow, what important proposal problem would still remain?
- Which parts of a mature proposal operation could realistically be packaged for a company that cannot hire proposal managers, librarians and dedicated knowledge-management staff?
I expect experienced practitioners to disagree with parts of this framing.
That would be useful.
The strongest lesson from this stage of the research has been that I should spend less time asking what AI is capable of doing and more time understanding how good proposal teams actually work.
The AI may change every six months.
The system that turns organisational knowledge, strategy and human judgment into a winning proposal is the more interesting thing to understand.