By December, every proposal pro will use AI. Will you lead—or follow?

Shipley Associates.

https://empower.shipleywins.com/courses/advanced-ai-topics-and-techniques-for-proposal-development?activecontentid=6eba97f9-b72d-454d-91c6-cfbbb4335747&meetingtime=null&utm_medium=email&utm_source=nwsl_09_09_25&utm_campaign=sai&utm_content=nwsl_pr_st

The AI essay-writing contest.

It's heeeere!!

I have to see it and results before saying anything pro or con. I took a course from this company and found it very informative and beneficial. The course got me to critically focus on what the solicitation asks for rather than reacting what I quickly thought was needed. Their instruction stresses submitting evidence of each assertion and citing examples results produced. It’s easy to write proposals that simply restate requirements or say “we’ve done this numerous times” or “this is how we do it” and not provide proof.

Edit to add comments: So I know this company understands what is a good proposal and how to write a responsive one. It will be interesting to see if AI can improve on offeror submissions by use of AI.

15 hours ago, Vern Edwards said:

It's heeeere!!

Yes, it is and it's here to stay as us humans roll off and out of various fields. It is just a matter of (short) time before various AI tools take over most of the procurement process in the Govt and related industries. The naysayers will be left behind for sure, so doubt AI at your own peril. Bit of a sidebar here but it reminds me of my father who worked for forty years in the auto industry as a designer. Computers had been utilized for various design inputs for years, at least since 1960 (Plymouth Valiant) but they "arrived" for good in the early 80s. Many engineers and designers just refused to believe that a computer could take over the role of 2D manual drafting efforts. Lee Iacocca mandated a hard shift to CADCAM in maybe 1978 or so right after he came onboard. Many engineers of all ages took to it with great enthusiasm, but quite a few did not and stuck to their old manual ways. When the company removed the manual drafting boards for good in the late 80s, those who didn't calibrate their skills were put out to pasture. The new system required far fewer designers and took much less time and effort. The only major company to continue using manual boards was Daimler and I think they were forced to transition with the times when they merged in the mid 90s.

3 hours ago, Motorcity said:

Yes, it is and it's here to stay as us humans roll off and out of various fields. It is just a matter of (short) time before various AI tools take over most of the procurement process in the Govt and related industries. The naysayers will be left behind for sure, so doubt AI at your own peril. Bit of a sidebar here but it reminds me of my father who worked for forty years in the auto industry as a designer. Computers had been utilized for various design inputs for years, at least since 1960 (Plymouth Valiant) but they "arrived" for good in the early 80s. Many engineers and designers just refused to believe that a computer could take over the role of 2D manual drafting efforts. Lee Iacocca mandated a hard shift to CADCAM in maybe 1978 or so right after he came onboard. Many engineers of all ages took to it with great enthusiasm, but quite a few did not and stuck to their old manual ways. When the company removed the manual drafting boards for good in the late 80s, those who didn't calibrate their skills were put out to pasture. The new system required far fewer designers and took much less time and effort. The only major company to continue using manual boards was Daimler and I think they were forced to transition with the times when they merged in the mid 90s.

I love how any discussion of AI is framed as a snake oil sales pitch: "Step right up and get your AI! You, my young sir, will lose your house, family, and everyone you love if you fail to use this miracle cure for what ails you! Want to succeed? AI is the only way to not be left behind!" I am always skeptical of technology that claims to be a cure all or a solution to every problem known to mankind. Sure, proposals are one area that may be able to use AI successfully, but let's see what happens before we get ahead of ourselves. In terms of taking over all of procurement, without any evidence or plan to implement these tools except, "AI will manifest itself because it's inevitable," I think a bit of skepticism here is warranted, especially when you consider the amount of CAPEX being thrown at the AI industry and the need for ROI. The business proposition that investing in AI will gain you entry into every market is certainly a tempting thought to private capital, a particularly gullible and sociopathic crowd, but also consider that these individuals/consortiums/groups are perfect marks--pockets full of cash and little in the way of critical thinking skills. As of now, our "AI" can't even select the correct clauses for our organization's contract actions.

I can see a contractor's history of project data, costs, on-time deliveries, positive customer correspondence, CPARs, etc., being fed into a Large Language Model. If the contractor is large and experienced, that would be a huge amount of information for an AI to mull over. Given a solicitation, it could identify similar projects that had been successfully performed. It could write up the benefits received--formerfed's "evidence" in support of the offeror's assertions. The LLM could look at solicitation requirements and identify gaps that need to be filled, in terms of either personnel to be hired or else subcontractors that need to be brought on board. The LLM could look at prior cost history for similar projects and make accurate estimates of future costs to be incurred. I can see all that happening;

I expect it will happen.

That said, I think people will still be needed, especially if what is being proposed is new and different from the past.

1 hour ago, opsyscons said:

I love how any discussion of AI is framed as a snake oil sales pitch: "Step right up and get your AI! You, my young sir, will lose your house, family, and everyone you love if you fail to use this miracle cure for what ails you! Want to succeed? AI is the only way to not be left behind!" I am always skeptical of technology that claims to be a cure all or a solution to every problem known to mankind. Sure, proposals are one area that may be able to use AI successfully, but let's see what happens before we get ahead of ourselves. In terms of taking over all of procurement, without any evidence or plan to implement these tools except, "AI will manifest itself because it's inevitable," I think a bit of skepticism here is warranted, especially when you consider the amount of CAPEX being thrown at the AI industry and the need for ROI. The business proposition that investing in AI will gain you entry into every market is certainly a tempting thought to private capital, a particularly gullible and sociopathic crowd, but also consider that these individuals/consortiums/groups are perfect marks--pockets full of cash and little in the way of critical thinking skills. As of now, our "AI" can't even select the correct clauses for our organization's contract actions.

I agree with the current AI tools available at some agencies. On the flip side, I have sat through a number of procurement-related AI demos where the clause selection was 98% accurate in some cases and perfect in others, and these weren't routine commercial buys, either. The room was floored.

12 minutes ago, Motorcity said:

I agree with the current AI tools available at some agencies. On the flip side, I have sat through a number of procurement-related AI demos where the clause selection was 98% accurate in some cases and perfect in others, and these weren't routine commercial buys, either. The room was floored.

Do you mean it was perfect in picking all provisions and clauses for a hypothetical solicitation?

3 hours ago, here_2_help said:

I can see a contractor's history of project data, costs, on-time deliveries, positive customer correspondence, CPARs, etc., being fed into a Large Language Model. If the contractor is large and experienced, that would be a huge amount of information for an AI to mull over. Given a solicitation, it could identify similar projects that had been successfully performed. It could write up the benefits received--formerfed's "evidence" in support of the offeror's assertions. The LLM could look at solicitation requirements and identify gaps that need to be filled, in terms of either personnel to be hired or else subcontractors that need to be brought on board. The LLM could look at prior cost history for similar projects and make accurate estimates of future costs to be incurred. I can see all that happening;

I expect it will happen.

That said, I think people will still be needed, especially if what is being proposed is new and different from the past.

All fair points, and I think this is more of what I expect vs. the hyperbolic assertions that all contracting professionals will be out of a job unless we start outsourcing our thinking to AI right now. I mean, maybe we will, I can't predict the future (hedging my bets here), but I don't really follow the world-domination sales pitch. That said, one of our source selection teams is going to participate in some training that uses the "Acqbot Platform" to assist in solicitation development, so I'll be interested in what they have to show us. The training providers are touting it as the "cure for everything," so we shall see. I'll report back if anyone is interested.

AI makes the process much, much easier.

The upside is both huge and obvious, I think.

• AI-assisted SOWs/requirements documents are much, much better.

• "AI, here are 7,000 contract files. These 27 have caused major problems, and these 350 minor problems. Analyze the contract files and determine what are the risk factors for problem contracts than can be identified pre-solicitation and pre-award." - AI is really good at answering questions like this.

• Its good at reading a 50-page contract and finding stuff like this (real example) - "The place of performance is Hawaii, but you have EST office hours, that might be a mistake."

I am, however, more cautious about the benefits of applying AI/LLMs to the document-intensive proposal -solicitation - evaluation process, specifically.

• AI may drastically increase the number and 'quality' of proposals. Every vendor can now, or soon will be able to, write a perfect proposal in seconds. Search theory suggests this fact may make both buyers and sellers take longer to make decisions, aka paradox-of-choice. In other markets - dating, college, jobs - this has been a big problem. We'll figure out how to deal with the tsunami, eventually, but it'll be a pain short-term. Some CS is foolishly going to use sam.gov and get 1,000 proposals for her $100K 're-paint the parking lots' solicitation.

• All proposals will be perfect, which will make differentiating them difficult. The fundamentally important but implicit assumption - better offerors write better offers - will no longer be true (although it never was completely true). This will be hard to deal with. Different evals maybe - the welcome death of "understands the requirement" as an evaluation factor? My pet solution is face-to-face meetings where the offerors can't BS you by typing your questions into ChatGPT, aka 'oral presentations.' When real-time video deep-fakes become undetectably good, if this hasn't happened already, in-person meetings.

In every profession, AI is best at doing the tasks that junior people did at the start of their career. The career and expertise pipeline, as it has existed for decades, is or soon will be broken. This will have to be dealt with. This a huge deal with software developers and lawyers right now, today, so we will learn from them before it hits us full-on.

On 9/10/2025 at 10:58 AM, here_2_help said:

I can see a contractor's history of project data, costs, on-time deliveries, positive customer correspondence, CPARs, etc., being fed into a Large Language Model. If the contractor is large and experienced, that would be a huge amount of information for an AI to mull over. Given a solicitation, it could identify similar projects that had been successfully performed. It could write up the benefits received--formerfed's "evidence" in support of the offeror's assertions. The LLM could look at solicitation requirements and identify gaps that need to be filled, in terms of either personnel to be hired or else subcontractors that need to be brought on board. The LLM could look at prior cost history for similar projects and make accurate estimates of future costs to be incurred. I can see all that happening;

Exactly. This is what any experienced proposal consultant or in-house business development specialist will tell a company to do. The benefit of a LLMis it can crunch all that data very quickly and thoroughly for the company to consider. Plus AI may tailor the information to the specific requirement in the solicitation.

When I was writing proposals, I dreaded being told to find and include this data. It takes a long time and I’m certain I missed a lot on each instance. A specialist then took my attempts and rewrote it to be right on target. AI might be capable of doing, minimally, the equivalent if not better and identify the gaps you mentioned.

On 9/10/2025 at 12:18 PM, Don Mansfield said:

Do you mean it was perfect in picking all provisions and clauses for a hypothetical solicitation?

Yes, it was so very close.

28 minutes ago, Motorcity said:

Yes, it was so very close.

How did they validate the accuracy of the output?

On 9/11/2025 at 2:55 PM, Don Mansfield said:

How did they validate the accuracy of the output?

The team looked at the FAR prescriptions and manually reviewed/compared each clause to verify accuracy. One if the hiccups was fill-in sections of various clauses. For instance, the spaces in 52.217-8 and -9, but I guess this firm was actively working on "flagging" anything that requires a fill-in. Nonetheless, this technology is only going to get better and we are absolutely going to be relying on it heavily on the procurement realm. Matter of fact, I see its usage as being mandatory very soon.

2 hours ago, Motorcity said:

Matter of fact, I see its usage as being mandatory very soon.

I hope it stays just as an option for an after the fact secondary check.

A bit off topic but I saw this post from Ryan Connell. He’s saying how he uses AI to prepare prototype project agreements:

“Conversational Contracting.

Just have a conversation, let #AI do the rest; here's the workflow everyone in United States Department of War can use today:

Contracting/Agreements Officer gets on a microsoft teams call with the requirement owner/PM.

Records it (Microsoft creates transcript)

Take that transcript to NIPRGPT, CAMOGPT, ASKSAGE, CORAS (Gary) [Any LLM approved for CUI]

with prompt:

"You are an 1102 #contracting officer well versed in the software acquisition pathway, other transactions, and writing 4022 #prototype project agreements which test and deliver outcomes. Review the attached transcript from a meeting that discussed the requirement below and populate the Project Agreement template attached [your local offices template] to best reflect the PMs requirements.

80-90% solution if you ask the right questions during a ~15 minute call...”

1 hour ago, formerfed said:

80-90% solution

Agree. I did this using our in-house AI (a version of Claude) for a FAR 13 BPA, with minimal prompting and no reference material (our AI does much better with references). We do this a lot in my office, exploring AI's capabilities. If you are curious, I uploaded it.

AI-generated BPA 16.09.2025.pdf

21 hours ago, General.Zhukov said:

If you are curious, I uploaded it.

I am curious. I am going to guess that your prompt was a for a quick demostration. In looking at the upload here is my first thought after spending about 5 minutes looking at the document.

Why was not your prompt more like this and if it was what would the result be? Emphasis added noting that "commercial" seems to be a key element of the exercise example. Likewise having now experience in such an exercise on use of something like "Claude" my wording may be off but I hope you get the thought process.

"PROMPT: You are an 1102 #contracting officer well versed in the procurement of high-volume commercial laboratory supplies. Prepare a FAR 13.303 Blanket Purchase Agreement following FAR part 12 for purchase of consumables like personal protective equipment, cleaning supplies, containers (bottles, jars, jugs, flasks, tubes, vials), dispensers, pipette tips, chemical tests, etc......"

It was just a quick demo - showing that even a really basic prompt can do a decent job, a pretty good first draft. I entered your prompt and uploaded results. You can see the difference. If I were to run the same prompt again, with changed punctuation and capitalization, the result would also be different. AIs are weird.

Regarding commercial - I didn't think about it. We buy lots and lots of small $ stuff for our many bio-pharma laboratories. These products are 100.00% commercial. Commercial is the water in which we swim, so us fish don't think about it too much. Also, I don't actually write BPAs these days. But you are correct.

AI-generated BPA 17.09.2025 - Commercial.pdf

1 hour ago, General.Zhukov said:

It was just a quick demo - showing that even a really basic prompt can do a decent job, a pretty good first draft. I entered your prompt and uploaded results. You can see the difference. If I were to run the same prompt again, with changed punctuation and capitalization, the result would also be different. AIs are weird.

Regarding commercial - I didn't think about it. We buy lots and lots of small $ stuff for our many bio-pharma laboratories. These products are 100.00% commercial. Commercial is the water in which we swim, so us fish don't think about it too much. Also, I don't actually write BPAs these days. But you are correct.

AI-generated BPA 17.09.2025 - Commercial.pdf

Thank you! I agree in both cases a good starting point. As I understand more about AI it would seem "Prompt" wording is important and output can be biased by the prompt wording.

Specific to my quick review of both results it would seem output is biased by what is already out there which in my view is not necessarily good. Again please understand my review was quick. Just a few of thoughts -

Why didn't combined synopsis/solicitation creep in?

Uniform Contract Format seems to be followed. I know the debate on its use but still it creeped in for both.

"Fair opportunity"? How did it creep in in the second effort especially when there was no mention in the prompt of FAR part 16?

Evaluation responses in both cases seem more like FAR part 15 than FAR part 13 and/or 12 making me wonder again about the how the prompt does or does not limit the result.

If I really spent some time I would probably find more but these were the quick ones.

Conclusion on my part - Yes AI is a tool of some value but one should not depend on its result completely. I question the ability of AI to sort out what is right by the FAR, what is not, and what is questionable.

Thanks again, much appreciated.

I’m in the “use it, but verify it” camp.

From what we’ve seen across proposal teams, AI helps most when it’s constrained by procurement-specific guardrails and a disciplined review loop. A few practices that have worked well:

1. Requirement-first, not prose-first. Start with a structured compliance matrix (every shall/should/must mapped to Section L&M). Let AI propose fills, but require a human checker to mark each line as “addressed / evidence attached / gap.” This avoids the “pretty essay, missed requirement” problem.

2. RAG over free-write. Retrieval-augmented generation using your own corpus (past proposals, CPARS excerpts, resumes, tech volumes, SOPs) + FAR/DFARS prescriptions reduces hallucinations. When we test clause selection, we measure precision/recall against the prescription and log every fill-in (e.g., 52.217-8/-9) so a CO/PM can review the parameters explicitly.

3. Citations and provenance. Every assertion should link to a source (CPARS ID, contract #, deliverable, KPI). Reviewers can then spot-check instead of re-write.

4. Prompt discipline ≠ policy discipline. A good prompt doesn’t fix wrong basis (e.g., FAR 15 language creeping into a Part 13 buy). We run checks for “part drift” (did Part 16/UC format show up where it shouldn’t?) and flag it.

5. Validation like any other tool. Treat models like interns with superpowers: benchmark them. For clause packs, we build a small gold standard set and record confusion matrices; for narratives, we score to L&M and evidence completeness. If it can’t beat your baseline, it shouldn’t be in the workflow.

On the concern about a “tsunami of perfect proposals”: agree that volume will rise. Two mitigations we’ve seen help on the buyer side (and you can encourage in Q&A):

• Evidence-gated instructions (require verifiable identifiers for past performance/results, not just text).

• Orals/working sessions focused on scenario execution (harder to “AI through” without real capability).

Disclosure for context: I work with Vamstar on AI-assisted proposal workflows in regulated markets (MedTech/Pharma). Not here to pitch—just noting that the biggest gains haven’t been “auto-writing,” but cycle-time and error reduction from requirement mapping, clause/attachment tracking, and evidence reuse under audit logs. If it’s helpful and within forum rules, happy to share a public brief on the validation approach we use (precision/recall against prescriptions, fill-in tracking, and reviewer sign-off).