RFP Genie: What We Learned from Twilio's Internal RFP Response Tool 

You know that sinking feeling when a 200-page RFP lands in your inbox? At Twilio, they used to feel it too. Nine people would clear their calendars, brew endless pots of coffee, and disappear into conference rooms for weeks. And here's the kicker—these RFPs were bringing in over a third of their revenue, but they had to turn many away because they simply couldn't keep up.

That was before RFP Genie.

I first heard about this when Wired dropped that article back in April about AI coming for sales jobs. But instead of the usual doom and gloom, sales teams were actually celebrating. Weird, right? Then I watched Pruthvi Shetty's presentation from the AI Summit in October, and suddenly it all made sense.

Why Nine People and Three Weeks Were Never Going to Cut It

Let me paint you a picture of Twilio's old RFP process. You've got product managers digging through documentation, sales engineers fact-checking technical specs, legal reviewing every clause, and somehow everyone's supposed to coordinate their input into one coherent document. By the time you're done, the opportunity might have passed.

Shetty laid out the brutal math: several weeks per RFP, nine people involved, and they're declining deals left and right because they're maxed out. Meanwhile, over 33% of revenue comes from these RFPs. Something had to give.

Here's where it gets interesting. Instead of buying some off-the-shelf RFP automation software, Twilio decided to build RFP Genie internally. And honestly? After seeing what they created, I get why.

The tech stack reads like a who's who of AI buzzwords, but stick with me because it actually makes sense:

GPT-4 does the writing. But not in that generic, obviously-AI way you're thinking. They trained it on Twilio's actual documentation, past proposals, the works.

FAISS handles the search. Never heard of it? It's Facebook's tool for finding stuff really, really fast. When you ask "What's our uptime SLA?", FAISS finds the answer in milliseconds from thousands of documents.

RAG keeps it honest. This is the secret sauce. Instead of letting the AI make stuff up (technical term: hallucinate), it forces every answer to come from real sources. Each response includes where it pulled the information from.

The whole thing works as an AI-powered sales intelligence tool that actually knows what it's talking about. Wild.

The Numbers Are Honestly Ridiculous

When Shetty showed the metrics, I had to rewind the video. Check this out:

• Three weeks → under 10 minutes
• 600-question monster RFPs? 8 minutes flat
• Cost per RFP: less than 10 bucks for the AI processing
• About 80% of answers need zero human editing
• Only 15-20% require the sales team to step in

How It Actually Works 

Here's what actually happens when someone at Twilio uses this AI proposal generation system:

First, you upload the RFP. The system reads it, figures out what they're asking for (harder than it sounds with some RFPs), and maps questions to topics.

Then you tell it which Twilio products this is about and add any context about the customer. Are they worried about compliance? Do they care more about scalability? This stuff matters.

Hit generate, and here's where the magic happens. Instead of sending each question to GPT-4 one by one, they batch them up. Think of it like carpooling for API requests. Smart.

Within minutes, you've got draft answers pulling from Twilio's knowledge base. But—and this is important—it shows its work. Every answer links back to source documents. No black box mystery meat.

The sales team reviews everything, tweaks what needs tweaking (remember that 15-20%?), and boom. What used to take weeks now takes less time than a coffee break.

Plot Twist: It's Not Just for RFPs Anymore

This is where things get really interesting. Once Twilio had this revenue execution platform component working, they realized something. Those compliance questionnaires that took a whole quarter? Same underlying problem. Customer support documentation? Yep. Operations procedures? You guessed it.

Suddenly their sales enablement platform became a Swiss Army knife for any process involving repetitive questions and document creation. The security questionnaire automation alone probably saved hundreds of hours.

One example Shetty mentioned: compliance reviews that took three months now wrap up in days. Days! Same tech, different use case.

Let's Talk About What Could Go Wrong (Because It's Not All Perfect)

Look, I love a good success story, but Shetty was refreshingly honest about the challenges. This isn't plug-and-play magic.

Data quality is everything. If your documentation is a mess, your answers will be, too. Twilio had to get serious about keeping its knowledge base clean and current. That auto-embedding pipeline they built? It's not optional.

You're dependent on external AI. When OpenAI has issues, you have issues. They designed it to work with different models (Claude, etc.), but you're still relying on someone else's servers.

Privacy gets complicated. You're feeding your company's secret sauce into an AI. How comfortable is legal with that? What about customers' data in those RFPs?

Integration is always a pain. Getting this to play nice with Salesforce and other tools? Not trivial. Legacy systems don't care about your fancy AI.

What This Means for the Rest of Us

The RFP Genie story isn't just about Twilio getting faster at proposals. It's a glimpse at how sales execution platforms are evolving. We're moving past the era of "let me help you format that document" into "let me handle this entire workflow while you grab lunch."

But here's what really struck me: the 80/20 split. They didn't try to automate everything. That last 20% where human judgment matters? They left it alone. The AI-powered sales intelligence tools that succeed aren't trying to replace people—they're trying to make people superhuman at the boring stuff.

For companies evaluating revenue enablement platforms, this case study is gold. It shows what's actually possible versus what vendors promise. It also shows the investment required—this wasn't a weekend project.

The Part Where I Mention SparrowGenie (Bear With Me)

Okay, real talk. RFP Genie sounds amazing, but unless you work at Twilio, you can't use it. It's their baby, built for their needs, running on their infrastructure.

But here's the thing—you don't need to build your own from scratch. At SparrowGenie, we saw what Twilio achieved and thought, "Why should only big companies with AI teams get this superpower?"

We're not revolutionizing anything. We just took the proven approach (RAG architecture, human-in-the-loop, the whole nine yards) and packaged it so you don't need six months and a team of engineers. Our proposal creation tools work pretty much the same way—upload RFP, add context, generate answers, review with your team, done.

No need to figure out FAISS or worry about rate limits or build your own embedding pipeline. We handled the plumbing so you can focus on winning deals. If that sounds useful, we should talk. If not, no worries—at least now you know what's possible.

Where This Is All Heading

Watching Shetty's presentation, one thing was crystal clear: this is just the beginning. Better language models are coming. Faster retrieval systems are in development. The gap between "AI-generated" and "human-quality" is shrinking daily.

The companies that figure out how to ride this wave—whether building their own like Twilio or using platforms like ours—are going to have a massive advantage. We're talking about reclaiming literally weeks of time per deal.

The real question isn't whether RFP automation software will become standard. It's whether you'll adopt it while it's still a competitive advantage, or wait until you need it just to keep up.

Based on what I saw from Twilio's results, I know which side I'd rather be on.

Sources

1. Twilio's Journey with RFP Genie - AI Summit Presentation by Pruthvi Shetty, October 26, 2024
2. Generative AI Is Coming for Sales Execs' Jobs—and They're Celebrating - Wired, April 2024