How a hospitality tech company turned RFPs from an expert scramble into a review job, with altspark

TJ Coyle, Co-founder and CEO·3 min read

The company's RFPs land on short notice: hundreds of questions in a customer's own workbook, sometimes due within a week. The answers depend on subject-matter experts spread across the business, who are some of the scarcest and most valuable people the company has. A technology leader at the company built RFP Responder on altspark. It pulls every question out word for word, drafts answers only from known good answers, and leaves anything it can't source blank for an expert. RFP Responder is now in beta.

At a glance

  • 150–600 questions per RFP, sometimes due within a week
  • 0 invented answers: with no source, the answer stays blank and goes to an expert
  • 100% of questions copied word for word, checked by code against the customer's file
  • 85% of RFP questions given a quality answer on the first pass.

The challenge

Every RFP that reaches the company's sales team looks similar. It's a customer-authored workbook of 150 to 600 questions, often with instructions and context documents attached, and it's sometimes due within a week. Customers choose their words carefully, so a paraphrased or skipped question is a problem, and a made-up answer is worse.

The old process put all the weight on the experts:

  • Many experts, all of them bottlenecks. Each RFP meant wrangling experts from across the company to complete their sections, on top of their real jobs.
  • Time lost to looking things up. About half the effort went into digging through old RFPs to find what the company had already said, so repeat questions got consistent, approved wording.
  • AI that said yes too easily. An earlier AI-assisted first pass over-claimed what the product could do.
  • Paid software that didn't fit. Commercial RFP response platforms exist, but they were overkill for what the company needed.

How altspark solved their problem

The technology leader built RFP Responder on altspark with Claude Code, through altspark's MCP server. It works in four steps.

Intake. The RFP package (workbooks plus context documents) comes in by email. Code, not a model, copies every question byte for byte and checks it against the customer's cell. A short brief captures the customer, their scale, their integrations and how they define "Yes." The RFP lead approves the brief before any drafting starts.

Drafting from a large bank of known good answers. Each question gets one of three outcomes:

  • an exact prior answer;
  • an adapted answer, written only from cited sources;
  • a blank, flagged "Needs SME" and assigned to the right owner.

The model can't touch the question text, can't choose a value outside the customer's dropdown, and can't add a capability its sources don't state.

Review on a shared page. Reviewers filter by section, owner or status; edit, assign and approve; and the RFP lead locks each final answer. A Gaps view shows where the bank is thin and which must-haves landed on "No."

Out the door, and back into the bank. A clean export or a copy-paste view returns the answers to the customer's format. When the RFP closes, the locked answers join the bank, so every RFP the company submits makes the next one faster.

"Our experts used to start every RFP from a blank page, then dig through old RFPs to answer repeat questions consistently, with approved wording. Now they start with a nearly complete RFP they simply review, tweak, and get back to their real jobs."

Technology leader, hospitality tech company

The impact (expected, in beta)

  • 70% reduction in time spent answering RFPs.
  • Experts review and fill gaps instead of drafting hundreds of answers from scratch, which matters most on short deadlines.
  • No more digging through old RFPs: approved wording for repeat questions is already in the draft.
  • Answers can't claim more than the company has already said in reviewed RFPs.
  • Every submitted RFP adds its final answers to the bank, so future RFPs take even less time.