(A)I Know What You Did Last Summer

By Frederick Lowe, October 13, 2025

In November 2024, the CEO of a global digital advertising company called me about a problem potentially costing his firm millions: lost Requests For Proposal (RFP).

Opportunities were getting drowned out in the noise of the modern Inbox, buried in departed seller accounts, or routed to the wrong teams. The issue wasn't the sales organization's professionalism. It was organizational scale, coupled with the realities of Inbox noise and industry churn.

Eventually, the inevitable question came: could we use AI to fix it? As I sometimes do, I said "yes", despite having a very limited sense about the scope associated with that answer.

This wasn't a request for a feel-good PR AI project or a flashy-but-meaningless demo, it was an opportunity to build a complete, practical AI platform that would address a systemic business problem. Perfect.

Thus began the project that would define my Summer of 2025, under the title "Entreprenur In Residence For Applied Artificial Intelligence."

The Approach

When this project landed, I had enough LLM experience to know any frontier model could classify an RFP reliably given proper context and prompting. Given the scale of the organization, I also knew the hard parts were going to be logistical and architectual:

  1. Conducting stakeholder interviews across a scaled, global organization
  2. Designing and implementing an agentic pipeline to mirror discovered workflows
  3. Reducing email noise while preserving critical context for scoring and extraction
  4. Conforming LLM outputs to dependent system requirements (Saleforce, Google)
  5. Linking extracted data to authoritative sources
  6. Designing low cognitive overhead Human-in-the-Loop confirmation steps
  7. Weaving all considerations into a platform that enhanced, rather than disrupted, existing work
  8. Building instrumentation to measure what mattered

Optimizing Excellence

Early in the project, I met with executive teams, not to discover the org chart, but to identify organizational Rock Stars. Maybe it goes without saying, but when you're modeling workflows, it's sensible to model them after the habits of top performers.

I then interviewed the identified account-level resources in detail about their day-to-day, and their managers about departmental challenges.

Account executives were spending 30 to 60 minutes per RFP on manual work: copying prospect information into Salesforce, cross-referencing past interactions, and scheduling brainstorm meetings across time zones.

The mean window from RFP arrival to first team meeting was 54 hours. That's long enough for aggressive or nimble competitors to already be in conversation with a prospect.

What I Delivered

The serivces I deliverved included, at various stages:

Call it "Enterprise AI Process Optimization" if you need a lanyard. In the end, I honored my mantra:

The best workflow solutions change... nothing at all.

Instead of changing workflows, I reduced drudgery and freed sellers to do the most beneficial thing they do for themselves and their organizations: sell.

The completed platform:

  1. Uses Enterprise OpenAI GPT to classify and extract information from incoming RFPs
  2. Confirms high-confidence outcomes with sellers via a push-button UI in Slack
  3. Programmatically creates prospect-stage opportunities in Salesforce with full context
  4. Automatically schedules a kickoff meeting for the account team accross time zones

Now, instead sifting through Inbox chaff, Sellers receive a Slack notification within seconds of receiving an RFP.

The Results

While still in beta, the system recovered hundreds of thousands of dollars of lost opportunities. We won't know for a quarter or so whether those opportunties will close, but on a projected basis the project has already acheived ROI.

Reviewing project goals, we had:

It's intellectually honest to note that while the time reduction is linear when sellers move immediately, in practice they sometimes take extra time to chat internally, gather additional RFP requirements, or structure their schedules for focused attention.

Still, we believe the metric is solid.

What I Learned

The lesson of this project wasn't that AI can perform a complex task like RFPs identification and detail extraction, it's that AI's highest value in enterprise settings comes from eliminating administrative taxes on human expertise.

I didn't replace human decision-making. I compressed the time between signal and action, letting experts do what they're already excellent at faster, and with less friction. That's not flashy. But it's worth millions.

Over the next several articles, I'll get into the technical details of my approach.