I've spent the last few weeks listening to recorded RFP discovery calls, demos, and customer conversations and there's an interesting pattern I’ve noticed. Everyone buys an RFP software tool for automation, right? What legacy RFP tools have done though, is created more work for proposal teams.

Most teams today have some version of Loopio or Responsive or something else, but the way legacy tools are built have forced teams to create layers of review and additional processes to compensate for what the tool should be automating but can’t.

Here are a few examples:

In one call, a team ran clean, well-organized source material through their new AI RFP platform (not ours!) and still got back answers that were just plain wrong or were written in a way that didn’t feel convincing to a seasoned proposal writer. 

In another call, someone asked the AI in their tool for a shorter response and got the exact same length back with a few words swapped in. 

My favorite example was when a customer fed a win theme into the system so the AI would lean into their strengths and watched it get jammed into nearly every single answer in that RFP, whether the question called for it or not. He liked the idea behind it, but he also had to go back through the entire questionnaire hunting for a phrase that was inserted in places it didn’t really make sense.

None of what I just described made anyone's process faster. What it did was add a job that didn't exist before: checking the AI's work on top of the review the team was already doing. 

AI in this process was supposed to remove layers of work. Instead, in a lot of what I heard from RFP teams, it's adding new ones. So here’s what I did.

I went stage by stage through the actual RFP workflow, from the moment a request lands to the moment it goes out the door, to see where AI RFP software is actually automating work and making teams more effective, and where teams are still doing the same work they always did, just with an extra step of double-checking the AI’s work or adding on additional processes due to the limitations of their current tool.

‍

Stages of RFP process and how AI should help

When you think about all the processes you’ve built around your RFP workflow, it begs the question: “Is my software actually helping me or creating more work?” I pulled examples from calls to show where the process breaks down and also put in the ‘ideal’ AI state (What AI should do) that teams should strive for. 

1. Intake & processing of RFP requests

What AI should do: Your team gets an RFP request in whatever tool you use for this (email, Salesforce, Slack, etc). The AI agent or system watches the queue, pulls the file, assesses it against pre-determined guardrails and strategy and it gets automatically uploaded into the RFP tool for answering.

Limitations of existing AI RFP tools: One large enterprise trust team already routes every request through Salesforce and what happens next is incredibly manual. Someone has to open each case, review the attachments, figure out how much effort the RFP needs, who is going to respond or take a first pass, and upload it to the tool they use. That team handles 700+ requests a month with a team of six people running that process manually and they still miss their own 15-day turnaround.

‍

2. Go/no-go screening

What AI should do: Scan a request for dealbreakers before anyone spends time writing RFP answers and summarize them for you.

Limitations of existing AI RFP tools: This is the one stage where AI is already delivering what it should be able to do. One rep told us they were able to catch three unwinnable requirements in a large RFP packet within a couple of minutes because their tool scanned for these and flagged them to the reviewer. This work would have normally taken an hour of reading if done manually. 

‍

3. Win themes and strategy

What AI should do: Give the AI in your RFP tool your instructions, deal brief, background materials, product slides, preferences, and have it come back with a real strategic approach before it generates answers. It should do things like tell you what to lead with, what to tie together across questions, and what this buyer cares about. You should also be able to chat with it and give it more context to refine the strategy that it will use for answer generation.

Limitations of existing AI RFP tools: Teams typically have to determine these on their own and put them into a static text box or section in the software. Even when these themes or strategies are input into the software, the AI doesn’t always apply these cohesively across the RFP. One team told us that the custom instructions and win themes are applied after answers are already drafted so AI is re-writing its own first draft. Most solutions don’t have a way for AI to intelligently strategize on win themes taking in context and source material and also apply the themes to answer generation. 

‍

4. Getting the file into the RFP software exactly as formatted

What AI should do: You should be able to drag and drop any file into an RFP solution and the AI should read any format, dropdown columns, merged cells, locked PDFs, and a mix of narrative and grid layouts in the same file, without anyone having to manually manipulate the file before uploading.

Limitations of existing AI RFP tools:

• One team uploaded an RFP document to their tool where each answer field was meant to stay separate like name, address, contact details, etc but the tool merged all of it into one long paragraph instead. It completely lost the structure of the file upon upload.
• Another team’s tool only supported two answer columns when their file had three columns and they had to do this one by hand. In these instances and more, teams spend time manually manipulating a file to fit the software’s limitations and many give up because it isn’t worth the time. 

‍

5. Getting a first draft of answers

What AI should do: Feed it your source material, context, preferences, win themes, and instructions and get answers back for an entire RFP that takes all of these things into account while also sounding natural and cohesive.

‍Limitations of existing AI RFP tools: It's a mixed bag. Some tools can produce a decent first draft, especially the newer, AI-native RFP solutions, but some still run into issues when it comes to AI answer quality. 

• One team described running clean, well-organized material through a tool and still getting answers that were wrong. 
• Another asked for a shorter response to certain answers and still got the same length back with different words swapped in. 
• One customer described feeding a win theme into the system and watching it get jammed into every answer that followed, whether the question called for it or not. He had to go back through the whole RFP looking for phrases that didn't belong which added time to his review.

‍

6. Delegation & cross-team collaboration

What AI should do: AI understands who is responsible for a question or section and automatically routes it to the right person or team. Instead of making them log into a separate platform to respond, that SME gets notified and can answer the question in the tool they’re already working in like Slack, Teams, email or even LLMs like Claude and CoPilot.

‍Limitations of existing AI RFP tools: Some tools can make AI assignee suggestions now and many tools allow for notifications through Slack or Teams, but many still take the reviewer back into the platform to respond. Most RFP software today still requires you to manually assign sections or questions to reviewers.

• One RFP manager said he spent years trying to get legal, finance, compliance, privacy, and even HR teams into his team’s RFP platform and they still refused to log in. He had to spend hours chasing down every one of their answers himself.

• Another team realized three or four days into a seven-day deadline that 75% of the RFP needed sales engineering work and no one from that team had been notified or routed any of the work. 

‍

7. Reviewing RFP answers

What AI should do: No ideal AI state for this as we think this step still deserves human time spent on review and polish. The ways AI can help are to be descriptive about why it flags a low confidence answer or doesn’t answer a question. 

‍Limitations of existing AI RFP tools: Most proposal and RFP teams use RFP response automation tools today and this step is still largely manual because teams prefer to do a human review to improve win rate and so unapproved answers don’t go out to the customer. 

‍

8. Maintaining your knowledge base

What AI should do: The system should stay current on its own with minimal involvement from a human. The AI knows when an answer is getting stale and flags it for re-verification. It learns from every edit and past answer without anyone having to toggle or tag something to be re-used and it weighs recent edits more heavily than older ones when two answers conflict. 

It also connects to various source materials and understands which source material applies to which product line, so a certification, architecture detail, or feature that only applies to one product doesn't bleed into answers about a different one. It pulls current information from whatever's connected, SharePoint, Confluence, an internal wiki, instead of working off a static export so its memory stays updated. Also, you should be able to use an LLM to actually review, categorize, and clean up the knowledge base itself, not just to pull answers.

‍Limitations of existing AI RFP tools: Maintenance of a knowledge library comes up on almost every call. It’s one of the most painful steps in this whole process, yet everything hinges on your library to be up to date and accurate to populate the correct answers. 

• One RFP manager described a daily 30-minute check-in with his team, manually re-entering answers from recently completed RFP projects into their shared repository, dating and time-stamping each one by hand.

‍

9. Final export and the narrative writing

What AI should do: The approved answers are inserted into the original document in the correct narrative format with no errors. Teams should be able to easily export or click a button to attach and send the completed RFP back into their ticketing tools, LLMs, or CRMs.

‍Limitations of existing AI RFP tools: Filling the original file back in works well almost everywhere now, but it’s still a manual process to export and send it where it needs to go. Also, writing the executive summary or cover letter that ties the whole response together is one area existing tools could improve on. One RFP lead said that her plan was to export the finished answers and then write that part in a different tool entirely.

‍

Score your own process

Is your tool actually automating work and making your team more efficient or is it adding more manual work and review for your team? Score your own tool and process in our mini quiz here and compare your score to this guide:

• 22–27: Rare. Most teams don't get here on more than two or three stages.
• 14–21: Normal. Real progress in a few spots but you might still be doing hours of manual work in some places.
• Below 14: Worth asking whether "AI-powered" is working for your team.

If you're tool isn't working for you, check out Conveyor's AI RFP software for intelligent AI that actually delivers on the AI promise of better answers and less manual work.

‍