From failed AI project to 95% time savings
A life sciences client came to Signal Point on their third attempt at an AI solution for RFP intake. Two vendors had already tried and failed, budget was nearly gone, and leadership had lost confidence in the entire project.
The problem
- RFP intake took more than 10 hours of human effort for a first pass.
- A basic AI prototype existed, but it hallucinated and produced answers no one trusted.
- Prior vendors treated it as a coding exercise, not a process problem — nobody deeply mapped how RFPs actually flowed through the business.
What we did
- Started with discovery, not code: interviewed stakeholders, mapped the real intake workflow, and documented constraints (on‑prem only, no external APIs, full traceability back to source documents).
- Built and demoed a new architecture on the client’s own NVIDIA hardware in three days.
- Iterated twice over two weeks and moved into production.
The solution
- A graph‑based retrieval system that understands how products, certifications, and specs relate, instead of just matching keywords.
- Every answer is backed by citations to the underlying documents, and nothing leaves the client’s environment.
- A “pinned answer” library lets sales engineers mark great responses so the system can reuse them on similar questions in future bids.
The results
- RFP intake time dropped from 10+ hours to about 30 minutes of human review — roughly a 95% time savings.
- Adoption and trust increased because every answer is sourced and reviewable.
- With each new RFP, the answer library grows, making the next response faster and more consistent.