AI Won't Fix Your Win Rate. But Ignoring it Won't Save You Either.

Stargazy's 2026 Proposal & Bid Software Report published a quote I gave them that's gotten a lot of pickup:

"AI replacing proposal teams is like handing someone a scalpel and calling them a surgeon. The tool doesn't do the work… the expertise does."

I stand by every word. And I want to be just as clear about the flip side, because the two-sided truth is what actually matters.

Firms that treat AI as a replacement for pursuit expertise are going to lose. Firms that refuse to use AI at all are also going to lose.

Both things are true. Neither one is a contradiction. The firms pulling ahead in GovCon right now are the ones who figured out the difference — using AI where it genuinely helps, keeping skilled humans on the work that matters, and building the validation discipline that keeps AI errors from shipping into proposals.

This post is the longer argument behind the quote. Because the quote by itself — taken too far in either direction — leads people to the wrong conclusion.

Where AI actually helps federal pursuits — with guardrails

There's a category of work in GovCon and federal AEC pursuits where AI can meaningfully accelerate what your team is already doing. Not replace the team. Accelerate their mechanics so they have more time for the judgment work AI can't touch.

The firms getting real value from these tools share one thing: discipline about what AI can do unsupervised versus what it can only do as a first pass under human review. Every category below comes with that caveat built in.

Opportunity intelligence and market analysis. Parsing hundreds of federal solicitations across SAM.gov, eBuy, GSA schedules, agency-specific portals, and industry feeds — looking for the ones that fit your firm — is work AI does well. Procurement-specific tools also surface patterns humans miss: agency buying cycles, teaming signals, price points on similar awards, incumbent vulnerability indicators.

The guardrail: AI-generated opportunity signals are leads, not conclusions. A tool flagging a pursuit as a strong fit is the start of a capture conversation, not the end. The judgment about whether to pursue still belongs to your capture team, who bring context the tool doesn't have — your relationships, your teaming strategy, your current capacity, your recompete obligations.

Solicitation shred and compliance matrix development — with validation. A 300-page RFP with multiple attachments, cross-references, and ambiguous requirements is grinding, detail-intensive work. AI accelerates the mechanics of it dramatically. What used to take a proposal team two or three days of manual extraction can get to a first-pass shred in under an hour.

But this is also where AI can hurt you badly if you're not disciplined. AI tools hallucinate. They miss requirements buried in attachments. They pull language that sounds like a requirement but isn't. They occasionally invent requirements that aren't in the document at all. And if an AI mislabels a Section L instruction as a Section M evaluation factor — or vice versa — and that error survives into your compliance matrix, you've got a serious problem.

The teams getting real value from AI on shred work treat the AI output as a first pass that a trained proposal professional then validates against the actual solicitation. Line by line. Section by section. The AI saves the hours of initial extraction. The human does the work only a human should do — confirming every requirement is captured, correctly classified, and tied to the right section of the response.

That validation step is not optional. It's the entire reason this use of AI is safe. Skip it, and you ship a compliance matrix with errors that evaluators will find before you do. And we all know what happens next.

Past performance mining — with verification. Most firms have strong past performance examples buried in a document management system nobody's catalogued properly. The project that closely matches the current RFP's requirements exists — it's in a folder somewhere nobody can remember. AI tools that index your project library and surface relevant past performance based on current requirements unlock value that's already there.

The guardrail: AI can surface the right project and still misrepresent it. It may pull outdated contract values, misstate the period of performance, conflate two projects, or describe scope that doesn't match what was actually delivered. Past performance claims are ones evaluators can and do verify. Every AI-surfaced past performance entry needs validation against the underlying contract documentation before it goes into a proposal. Don't trust the summary. Check the source.

First-draft content from structured inputs — heavily edited. Turning a capture plan, SME interview transcript, or meeting notes into a starting draft of a proposal section. This is where AI saves the most writing time, and also where overreliance on AI produces the most obviously generic proposals.

The honest version: AI-generated first drafts are a time-saver for writers who know what good looks like, and a trap for writers who don't. An experienced proposal writer can take an AI-generated draft and turn it into strong, voice-appropriate, scoring-aligned content in a fraction of the time it would have taken to draft from scratch. An inexperienced writer — or a writer under deadline pressure who doesn't push back on what the AI produced — ships AI-flavored content that sounds like every other AI-drafted proposal the evaluator has seen this quarter.

AI also hallucinates in technical content. It will confidently describe methodologies that don't exist, cite standards that don't apply, or attribute capabilities to your firm that you don't actually have. Every factual claim in AI-generated draft content needs verification. This is especially true in technical and past performance sections, where errors can cost you more than just scoring points.

Proposal boilerplate management and consistency checking. Version control. Terminology consistency. Acronym definitions. Cross-reference validation. Template adherence. These are tasks where AI is genuinely strong and the error modes are more forgiving — an AI miss on an acronym definition is embarrassing but not usually fatal.

The guardrail here is lighter but still real: AI isn't a substitute for your final quality control pass. It catches most inconsistencies but misses some, and the ones it misses tend to be the edge cases that matter. A trained human eye on the final document is still the last line of defense.

Capture plan foundation and competitive intelligence. Pulling together customer profile data, incumbent analysis, competitive landscape, and historical award patterns into a capture plan starting framework. AI is good at the research aggregation. It's not good at the judgment.

The guardrail: treat AI-generated capture intelligence as input to your analysis, not the analysis itself. The competitive assessment that matters — which competitor will actually bid aggressively, whose relationship with this customer is genuinely strong, what the political context looks like — requires human judgment applied to information AI doesn't have access to. AI can give you the data foundation. Your capture team has to build the strategy on top of it.

Where AI is still bad — and probably always will be

Here's the other half. The half that matters just as much.

Strategy. AI does not understand your customer's political context, their unwritten priorities, or the things they told you off the record at industry day. It doesn't know which competitor is hungry for this work and which one will phone it in. It can't assess whether your incumbent advantage is real or fragile. It has no idea why you're the right firm for this specific opportunity in ways that would actually resonate with this specific evaluator.

Strategy requires judgment applied to context. AI can describe context. It can't make the judgment.

Customer-specific insight. The moments in proposals that win — the sentences where the evaluator thinks "they get it" — come from deep human understanding of the customer's world. What their program office is trying to achieve. What leadership has said publicly and what's only been said in hallway conversations. How the contracting office's patterns shape proposal strategy. Why this pursuit is different from the five other pursuits the team has worked this month.

AI produces generic language because AI is trained on generic patterns. Customer-specific insight is a human product.

Voice. Your firm's voice. Your principal's voice. The voice your past customers recognize in a cold-submitted proposal. AI imitates the pattern of voice but not the specifics — the proof points only your team knows, the stories only your principals have lived, the opinions only your firm holds because of who you are. Proposals written predominantly by AI sound like everybody else's AI-written proposals. Evaluators are increasingly calibrated to spot the pattern.

Editorial judgment. When page count is tight, what do you cut? Which win theme gets more real estate? Which past performance example replaces which other one? Which risk deserves a paragraph versus a sentence? These are editorial decisions that depend on understanding the evaluation criteria, the customer, the competitive landscape, and the specific strategic decisions your capture team has made.

AI does not make these calls well. Experienced proposal managers and writers do.

Technical accuracy in specialized domains. AI hallucinates. Especially in specialized federal technical domains — specific agency regulations, specific defense requirements, specific scientific or engineering standards. Confident, fluent, wrong. If you're using AI to draft technical content, human expert validation is not optional. It's the entire safeguard. The fluency of AI output creates a false sense of accuracy. A paragraph that reads smoothly and cites real-sounding references can be completely wrong on the substance. Your SMEs still have to check the work.

Relationships. This one's obvious but worth stating. Federal procurement still runs on relationships. AI does not build them. Capture professionals do.

The honest framing

Stop asking whether AI will replace your proposal team. That's the wrong question, and it leads to two equally wrong answers.

The right question is this: is your team using AI to accelerate work that needs acceleration, with the validation discipline to catch the mistakes AI will make?

If yes, you're going to win more. You'll run more pursuits. You'll qualify opportunities faster. You'll enter RFP response with better capture maturity. You'll have more time for customer conversations, competitive analysis, and strategy. Your people will spend their hours on the work that moves win rates.

If no, one of two things will happen. Either you'll keep up by burning out your team — running the same pursuit pace with manual workflows that your competitors have automated. Or you'll fall behind because you're producing less, with more effort.

But there's a third failure mode that's worth naming. Teams that adopt AI without validation discipline. They're the ones who ship proposals with hallucinated past performance, misclassified compliance requirements, or technical content that sounds right but is wrong. They're moving faster than their competitors, but they're also introducing errors that can cost contracts or damage customer relationships.

The scalpel line holds. The tool doesn't make someone a surgeon. A surgeon without good tools is operating with a rusty knife. And a person with a scalpel who hasn't been trained to use it does real damage fast.

How to actually adopt AI in GovCon pursuit work

For firms building AI capability into their pursuit operations, a few practical guidelines.

Start with the tasks where error cost is lowest. Boilerplate management, formatting, consistency checking. Build confidence with the tools on work where mistakes are easy to catch and fix. Move to higher-stakes tasks — shred, past performance, technical drafting — only after your team has developed instincts for where AI fails.

Use procurement-specific tools for procurement-specific work. General-purpose tools like ChatGPT, Claude, and Gemini are excellent for many things. For procurement-specific work, purpose-built tools trained on federal procurement data produce meaningfully better results and have guardrails designed for this use case. The difference is real. Don't try to make a general tool do what a specialized tool does better.

Build validation into every workflow. For every AI-assisted task, define who validates the output and what they're checking for. Shred gets a trained proposal pro validating line by line. Past performance gets verified against contract docs. Technical content gets SME sign-off. Voice gets editor review. No AI output ships unvalidated. Write this into your proposal process. Enforce it.

Train your team on AI failure modes, not just AI capabilities. Most AI training focuses on what the tools can do. The more important training is what they get wrong and how to catch it. Teams that know where AI fails are much safer users than teams that just know what AI can produce.

Protect your voice. Decide how AI fits with your firm's voice before you use it at scale. Some firms will use AI extensively for boilerplate, compliance, and internal-facing content while keeping the customer-facing proposal prose human-authored. Others will use AI more broadly but invest heavily in voice-specific editing. Both work. Drifting into "whatever AI produces is what we ship" does not.

The Summit Win System™ view

Within the Win System, AI is most useful in Plan and Propose — where research, analysis, and document production workflows benefit from acceleration. AI has some role in Position and Persuade but cannot replace the human work of building differentiation and crafting customer-specific content. And in the deeply human work of capture — the Position peak — AI supports but does not drive.

The firms winning right now have figured out this distribution. They use AI to accelerate the mechanical parts of pursuits, which frees up their team's time to invest more in positioning, relationships, and strategy. The net effect is more bids, better-shaped bids, and higher win rates. But they got there by building validation discipline first, not by adopting AI tools and assuming the tools would work cleanly out of the box.