RFPs are one of the most time-consuming documents in project management — and one of the most underutilized opportunities for AI to help.
Whether you’re writing an RFP, reviewing one, or responding to one, AI can reduce the time it takes to do the work well. I’ve been on all three sides of this, and the use cases are different enough to be worth breaking out.
Writing an RFP
When a city I’m involved with needed to replace a roof, the RFP process started from scratch. Someone had to define what contractors would be evaluated on, what information they needed to submit, and what the scope of work actually was.
If I’d been asked to help draft it, the first thing I would have done is give AI the basics — what we needed, any known requirements, and the context — and asked it to draft the structure. Scope of work. Contractor qualifications. Bid submission requirements. Evaluation criteria. Timeline.
AI is good at this kind of structured drafting. It knows what a well-formed RFP looks like. It won’t forget the insurance and bonding requirements. It will include evaluation criteria even if you didn’t think to mention them.
What it won’t do is know your specific situation. It doesn’t know that your jurisdiction requires prevailing wage, or that your last roof contractor left without finishing the flashing. That’s the context you bring. AI gives you a professional starting point; you make it accurate.
Reviewing an RFP
This is where I think AI is most underused. Before an RFP goes out, someone should be asking:
- Are there gaps — things a contractor would need to know that aren’t here?
- Are there inconsistencies — places where the document contradicts itself?
- Is anything too specific?
That last one matters more than people realize. In public procurement especially, an RFP that requires a very specific product, brand, or configuration can be challenged as steering toward a particular vendor. It also limits your pool of qualified bidders in ways that may not serve the project.
I serve on a local government committee, and when we needed to replace a roof, I didn’t see the RFP until we were already reviewing bids. By then it was too late to fix the gaps. Looking back, I would have pasted it into Claude — or any AI tool — and asked: What gaps does this have? Are there any inconsistencies? Is anything written in a way that could be seen as too restrictive or steering toward a specific vendor?
That’s a fast, low-cost review that catches things a tired committee might miss at 6pm after a two-hour meeting.
Responding to an RFP
I worked with a startup — Rohirrim — that built a tool specifically for this problem. What’s worth noting: they built this before ChatGPT launched, before Copilot rolled out, before most people had any frame of reference for what generative AI could do in a business context. They saw the problem early.
The concept: ingest a company’s past RFPs, product documentation, and institutional knowledge, then use AI to generate responses to new RFPs.
It’s a real problem. Responding to RFPs is expensive and time-consuming. For companies that respond to dozens or hundreds of RFPs a year, a lot of that work is repetitive — pulling the same boilerplate, the same certifications, the same case studies — and reassembling them for each new response.
The AI approach makes sense: if the knowledge exists in the company, you shouldn’t have to manually dig it up every time. You give the tool the RFP, it finds the relevant answers from your existing materials, and you review and refine rather than drafting from scratch.
The quality of the output depends entirely on what you’ve fed it. Garbage in, garbage out — but the reverse is also true. A company with well-documented past work and solid product materials can generate a much stronger first draft than one starting from a blank page.
Today, enterprise tools like Copilot can do some version of this if your organization’s knowledge is organized in SharePoint. The underlying idea Rohirrim was building toward is now table stakes for AI vendors. They just got there a few years early.
You don’t need an enterprise tool to do a version of this. If you’re a PM or a small team responding to an RFP, you likely have the raw material already — past project summaries, lessons learned documents, case studies, budget actuals, client testimonials. This is called grounding: giving AI your specific context so its output reflects your actual work, not generic advice. Feed those materials into Claude or any capable AI along with the RFP, and ask it to identify which past projects are most relevant, what evidence supports your qualifications, and where the gaps are. The quality of the response depends entirely on what you’ve grounded it with — but a well-documented project history turns a blank page into a strong first draft.
The Through-Line
All three use cases have something in common: AI handles the structure and the recall so you can focus on the judgment.
When you’re writing, AI gives you the scaffolding. You supply the specifics.
When you’re reviewing, AI asks the questions a thorough reviewer would ask. You decide which answers to change.
When you’re responding, AI retrieves what your company already knows. You make sure it’s accurate and competitive.
The RFP process exists to create a fair, structured way to evaluate options. AI doesn’t replace that judgment — it removes the friction so you can apply it better.