Rfp Fit Assessment

RFP Fit Assessment (Go / No-Go)

Step 2 in the RFP Agent Swarm. After rfp-intake has parsed the RFP into structured metadata and a classified task list, this skill runs a transparent, weighted qualification scorecard and assembles a one-page Go/No-Go memo so a human owner can make an informed bid decision fast.

When to Use

Invoke this skill when any of the following are true:

• A new RFP has just completed rfp-intake and a bid/no-bid decision is required before proposal work begins.
• An Account Executive (AE) or Proposal Lead asks for a "fit score", "qualification", or "should we bid" analysis on a specific opportunity.
• Leadership needs a standardised one-page memo to approve or decline pursuit.
• The answer-bank match rate needs to be estimated BEFORE committing SME hours.
• The team wants a reusable scorecard artefact to attach to CRM opportunity records for audit and retrospective analysis.

When NOT to Use

Quick Start (Worked Example)

Scenario: rfp-intake has just produced working/rfp_metadata.json and working/task_list.json for "Acme HealthCo — Patient Portal Modernisation RFP, deadline 2026-05-20, 142 questions".

User: "Run fit assessment on the Acme HealthCo RFP"

Claude:
  1. Load working/rfp_metadata.json + working/task_list.json (from rfp-intake)
  2. Pull bank_stats.json from rfp-answer-bank (historical match rates)
  3. python scripts/kb_match_estimator.py
       --task-list working/task_list.json
       --bank-stats working/bank_stats.json
       --output working/kb_match_estimate.json
     => overall_estimate_pct: 68, confidence: MEDIUM
  4. Prompt human / AE for dimension scores 0-5 (technical, commercial,
     competitive, strategic, resource, deadline) — store in scorecard.json
  5. python scripts/compute_fit_score.py
       --scorecard working/scorecard.json
       --output working/fit_result.json
     => weighted total 71.5 -> band: CONDITIONAL
  6. python scripts/generate_go_no_go_memo.py
       --fit working/fit_result.json
       --kb working/kb_match_estimate.json
       --metadata working/rfp_metadata.json
       --template assets/go-no-go-memo-template.md
       --output working/go_no_go_memo.md
  7. Render Adaptive Card via render-ui (KPI row + risk donut + recommendation)
  8. (Optional) Hand working/go_no_go_memo.md to docx skill -> exec Word memo
  9. (Optional) Hand working/fit_result.json to xlsx skill -> detailed
     scorecard workbook with per-dimension evidence

Human decision owner (default: VP Sales) signs the memo. Only after sign-off does rfp-respond begin.

Core Instructions / Workflow

Step-by-Step

1. Ingest intake outputs. Read rfp_metadata.json and task_list.json produced by rfp-intake. If missing, STOP and route the user back to rfp-intake.
2. Estimate KB match rate. Run scripts/kb_match_estimator.py against the task list and answer-bank historical stats. Record overall_estimate_pct and confidence.
3. Gather dimension scores (0-5) from the AE / Proposal Lead. Use the rubric in references/fit-scoring-rubric.md. KB-match dimension is pre-filled from step 2; the remaining six are human-supplied but rubric- anchored.
4. Run the kill-criteria check. See references/fit-scoring-rubric.md section "Kill Criteria". If ANY fire, flag on the memo — human still confirms.
5. Compute weighted score. Run scripts/compute_fit_score.py. Output band: Go (>=75), Conditional (50-74), No-Go (<50) — ADVISORY ONLY.
6. Identify top risks & mitigations. Pull from the dimension evidence fields; cross-reference references/competitive-positioning-playbook.md.
7. Generate memo. Run scripts/generate_go_no_go_memo.py with assets/go-no-go-memo-template.md.
8. Render summary card via render-ui (built-in) for quick review.
9. Produce exec artefacts (optional): one-page Word memo via docx; detailed workbook via xlsx.
10. Route for human sign-off. Named owner per references/go-no-go-decision-criteria.md RACI table.

Scorecard Dimensions & Weights

Weights MUST sum to 100; compute_fit_score.py enforces this.

Recommendation Bands

Kill Criteria (Auto-Flag, Still Human-Confirmed)

See full table in references/fit-scoring-rubric.md. Examples:

• Mandatory certification we do not hold (e.g. FedRAMP High).
• Onsite / in-country data residency we cannot provide.
• Deal value below the commercial floor defined in Deal Desk policy.
• Exclusive incumbent relationship with a hard-locked renewal.

Built-In Skills Used

No other built-ins are required. Do NOT invoke PDF, PowerPoint, Meetings, Scheduling, Daily Briefing, or M365 Search for this step unless explicitly asked.

Audit Log

This skill appends events to the shared RFP audit log at output/rfp-<rfp_id>/audit-log.xlsx. Rows are appended via the Excel built-in skill; the schema itself is owned by rfp-answer-bank.

• Schema reference: audit-log-schema.md
• Append helper (owned by rfp-answer-bank): append_audit.py
• Note: rows are appended via the Excel built-in; the schema and helper script are owned by rfp-answer-bank — do not fork or duplicate the schema here.

Adaptive Card Dashboard

After compute_fit_score.py produces working/fit_result.json, this skill renders a scorecard dashboard via the Adaptive Cards built-in. The card is the primary human-facing review surface before sign-off.

Card layout:

• KPI row — weighted_total score, recommendation badge (Go / No-Go / Conditional), confidence label from the KB-match estimator.
• Radar / bar chart — all 7 dimensions plotted with their 0-5 scores: KB Match, Technical Fit, Commercial Fit, Competitive Positioning, Strategic Alignment, Resource Availability, Deadline Feasibility.
• FactSet — top 3 strengths (dimensions scoring >=4 with evidence) and top 3 risks (dimensions scoring <=2, or any kill-criterion flag).
• Action buttons —
  • Approve — proceed to drafting (unlocks rfp-respond)
  • Decline
  • Defer for exec review

After the human clicks an action button, this skill writes a HUMAN_DECISION_LOGGED event to output/rfp-<rfp_id>/audit-log.xlsx with the decision, decision_maker, and decision_rationale captured from the card submission.

Output Deliverables

Guardrails

1. The system does not decide. Ever. Every output surfaces a named human decision owner. The score is a recommendation band, not an instruction.
2. No freeform AI memo content. generate_go_no_go_memo.py performs PURE template substitution on assets/go-no-go-memo-template.md. Do not rewrite, embellish, or omit placeholders.
3. KB match is an ESTIMATE. Always label it "estimated" on the memo and card. True match rate is only known after rfp-respond runs.
4. Weights must sum to 100. compute_fit_score.py validates and exits non-zero on mismatch. Do not silently normalise.
5. Kill criteria are flags, not auto-declines. Flag prominently on the memo; the human still signs.
6. No external API calls from scripts. Python 3 stdlib only. Any retrieval happens via built-in Enterprise Search / Deep Research in the agent loop, not inside scripts.
7. Evidence required. Any dimension score below 2 or above 4 MUST include an evidence string (1-2 sentences) referenced in the memo.
8. Audit trail. Persist working/scorecard.json and fit_result.json alongside the memo. The xlsx artefact is the archival record.

Common Issues

Related Skills

RFP chain

rfp-intake  ->  rfp-fit-assessment (THIS SKILL)  ->  rfp-respond  ->  rfp-gates  ->  rfp-review  ->  rfp-assemble
                              |
                              +-- rfp-answer-bank (substrate across all steps: KB, audit log, match stats)

• This skill consumes rfp-intake's task list (working/task_list.json) and rfp-answer-bank's KB match-rate estimate (bank_stats.json + kb_match_estimator.py).
• Its approval unlocks rfp-respond — no drafting begins until the human signs the Go/No-Go memo and HUMAN_DECISION_LOGGED is written with decision = Go.
• rfp-answer-bank is the shared substrate: it owns the knowledge base, the audit-log schema, and the append_audit.py helper consumed by every step in the chain.

Cowork built-ins leveraged

References

• references/fit-scoring-rubric.md — dimension-by-dimension 0-5 anchors, weighting formula, worked example, kill criteria.
• references/go-no-go-decision-criteria.md — decision matrix, stakeholder RACI, escalation timing, common patterns, AE question templates.
• references/competitive-positioning-playbook.md — incumbent/competitor inference, win-theme mapping, enrichment via Enterprise Search & Deep Research.

Scripts

• scripts/compute_fit_score.py — weighted scorecard computation; emits band + per-dimension contributions.
• scripts/kb_match_estimator.py — estimates % of questions answerable from the answer bank.
• scripts/generate_go_no_go_memo.py — deterministic template substitution to produce the Go/No-Go memo.

Assets

• assets/go-no-go-memo-template.md — exec-brief one-page memo template with [PLACEHOLDER] tokens.