The problem in one sentence
Public-sector RFPs and RFQs are scattered across dozens of portals, written in inconsistent formats, and mostly irrelevant to any given firm — so the real work is not finding solicitations, it’s triaging them fast enough that the good ones get a proper response.
This page describes the pipeline Node8 is designing for an award-winning landscape architecture studio: a small firm with a strong win rate whose growth is constrained by discovery and triage, not proposal quality. The design generalizes to most professional-services firms that live on public procurement.
Stage 1: Sources and monitoring
The pipeline starts with an inventory of where opportunities actually appear, built with the client rather than assumed:
- Portal aggregators — BidNet-style state and regional systems that consolidate municipal solicitations.
- Individual agency sites — parks departments, transit authorities, port authorities, and city capital-projects offices that post directly and never syndicate.
- Non-portal sources — institutions, universities, foundations, and quasi-public entities issuing RFQs for culturally significant projects; these rarely hit procurement aggregators and are where generic tools go blind.
- The human network — opportunities forwarded by collaborators. These enter the same queue so they get the same structured analysis instead of living in someone’s inbox.
Monitoring is scheduled polling plus change detection, normalized into a single stream: every new solicitation, whatever its origin, becomes one record with a source link and the original documents attached.
Stage 2: Extraction
Solicitation documents are long, boilerplate-heavy PDFs. An LLM extraction pass converts each one into structured fields:
- Issuing agency and project location
- Scope summary and project type
- Budget or fee signals, when stated
- Key dates — questions deadline, submission deadline, anticipated award
- Selection criteria and their weights
- Team and eligibility requirements — licensure, MWBE participation, insurance, prior similar projects
- Prime-versus-subconsultant structure
Two design rules keep this honest. First, every extracted field carries a citation back to the page it came from, so a human can verify in seconds. Second, “not stated” is a valid answer — hallucinated budgets or invented eligibility rules are worse than gaps.
Stage 3: Fit scoring against the firm’s portfolio
This is the stage no generic RFP platform does, and the reason the engagement exists. Keyword alerts on “landscape architecture” bury a firm focused on culturally significant public work under median plantings and drainage retrofits.
The fit profile is built from the firm’s own record — project and pursuit lists from the past three years, wins and losses included. From that history, the analyst scores each solicitation on dimensions like:
- Work type match — does the scope resemble what the firm does and wants more of?
- Significance markers — memorial, heritage, community, and public-realm signals in the scope language.
- Winnability — selection criteria the firm scores well on, incumbent signals, team requirements it can meet directly or with known partners.
- Practicality — geography, timeline versus current studio capacity, submission effort versus fee size.
Scores come with written reasoning, not just a number. A principal should be able to read why the analyst rated something 8/10 and disagree — and that disagreement is fed back into the profile, so the scoring converges on the firm’s actual judgment over time.
Stage 4: Go/no-go summary and human review
For each opportunity above threshold, the analyst drafts a one-page go/no-go brief: what it is, why it fits (or where it’s marginal), what pursuing it would cost in effort, deadlines, and open questions to resolve before committing. The decision itself stays human — the historical 18-20% win rate reflects judgment the system is meant to amplify, not replace.
Below-threshold opportunities aren’t deleted; they’re logged with their scores. That archive is how the firm audits the analyst for false negatives — the failure mode that actually matters in this domain.
Stage 5: Proposal inputs
When the firm says “go,” the extracted structure keeps paying off: a requirements checklist to track compliance, the evaluation criteria mapped against the firm’s qualifications and past projects, and suggested portfolio matches for the relevant-experience section. The client’s earlier experiments with ChatGPT-drafted proposals produced generic prose that needed heavy editing — so this design deliberately stops at structured inputs and matched evidence, leaving narrative voice to the designers.
Where the design stands
This is an in-progress engagement. The current phase — analyzing the firm’s 2024-2026 pursuit history, inventorying sources, and evaluating whether existing RFP tools cover the monitoring layer — comes before any build, and the fit-scoring and go/no-go stages will be piloted against real solicitations with the principals reviewing every output. If an off-the-shelf platform handles Stage 1 adequately, Node8 will use it and build only the stages that are genuinely firm-specific.
Work with Node8
If your firm’s pipeline depends on public solicitations and triage is eating principal time, Node8 designs and builds AI analyst systems scoped to how your firm actually wins work — starting with an honest look at your pursuit history. See our AEC practice or get in touch.