# Deck 19 · RFP Response Platform **Status:** Draft (not yet built) **Saved:** 2026-06-28 by Jeff (verbal) **Owner:** Plex / OMX Commercial + Sales Ops --- ## One-line thesis **Stop re-answering the same RFP questions for the seventh time.** A central knowledge base — fed by every past response, customer FAQ, contract clause, pricing matrix — that auto-drafts RFP/tender responses, supports pricing review, and accelerates deal prep. Hours of work compressed to minutes; consistency of voice across every bid. ## The wedge — why now - RFPs eat sales-ops + product + commercial time in long bursts; the same 80 questions show up every time - Anthropic + central knowledge base + Snowflake pricing data make automated drafting viable - OMX's win rate is shaped as much by speed-to-respond and clarity-of-response as by price - Internal product (sales-team facing) — no external surface in v1 ## What this deck covers 1. **Central knowledge base** — every past RFP response, customer FAQ, product spec, pricing matrix, contract clause — searchable and citable 2. **AI-drafted response** — submit the RFP doc; system extracts questions; AI proposes answers from the knowledge base; sales-ops reviews and ships 3. **Pricing review support** — when an RFP asks pricing, the system pulls from PPSS-grade rate cards and contract benchmarks 4. **Deal prep** — beyond the RFP doc: competitor positioning (from Plex-CI), customer health (from Briefing Room), upsell opportunities 5. **Win/loss analysis** — every response logged + outcome captured + feedback loop into the knowledge base 6. **Voice consistency** — brand-tone-locked drafting; same OMX voice across every bid no matter which AM submits ## What this deck explicitly does NOT do - Not a customer-facing portal (no two-sided marketplace; internal tool) - Not contract management (separate workflow; Legal owns) - Not catalog pricing (PPSS / Catalog feed in; not built here) - Not the QBR process (Deck 18 owns) --- ## Problem framing (what's broken) - **Same questions, every time** — "how do you handle data security?" answered manually for the Nth time - **Inconsistent voice** — different AMs answer the same question different ways - **Pricing under pressure** — fast turn-around forces shortcut decisions on margin - **No central memory** — last year's winning response is in someone's email - **Win/loss undocumented** — no learning loop after the bid closes - **Time hostage** — big RFP = 2-3 weeks of senior team time ## Benefits (the value story) | Lever | Mechanism | Sizing approach | |---|---|---| | **Response time** | AI-drafted first pass | Days → hours; 5-10x faster | | **Win rate** | Faster, sharper, more complete responses | Industry: 5-15% win-rate lift from RFP automation tools | | **Sales-ops capacity** | Less reinvention; more curation | Reclaim significant FTE-week per quarter | | **Margin discipline** | Pricing pulled from approved rate cards | Reduces ad-hoc discount leak | | **Voice consistency** | Locked brand voice | Strategic — submission quality | | **Learning loop** | Win/loss feedback into knowledge base | Compound improvement over time | --- ## Layout candidates from the gold standard - **Cover** — sales-ops team member at desk; RFP doc on one screen, draft response on other; clean and fast. - **Problem vector grid (4-6)**: Same-questions / Inconsistent-voice / Pricing-shortcut / No-memory / Win-loss-lost / Time-hostage - **The knowledge base — what's in it** — visual taxonomy of source content - **AI-drafted response walkthrough** — RFP doc → questions extracted → answers proposed → human reviews → ships - **Pricing review flow** — request → rate-card lookup → contract precedents → recommended response with margin band - **Win/loss loop** — diagram showing how outcomes feed back into the knowledge base - **Architecture pipeline** — Knowledge base + PPSS pricing + Plex-CI competitor data + Briefing Room customer data → AI drafting engine → Sales-ops UI → submitted response - **Roadmap** — POC on next 3 RFPs (90d) → broader rollout (6mo) → full feedback loop + voice library (12mo) - **The ask** — Knowledge-base build + Anthropic API + sales-ops process change --- ## Open questions to resolve 1. **RFP volume** — how many RFPs/year? Tier distribution by deal size? 2. **Current win rate** — baseline to measure against 3. **Knowledge base seed** — last 12 months of responses available to ingest? 4. **AI provider** — Claude Sonnet 4.6 / Opus 4.7; cost per RFP 5. **Pricing integration** — when PPSS isn't live yet, where does rate card live? 6. **Approval workflow** — who signs off pricing? Commercial Director? CFO? 7. **Confidentiality** — RFPs often contain customer-confidential data; tenant isolation? 8. **Integration with Deck 18 (Briefing Room)** — customer mode-2 + mode-4 are key context inputs 9. **Integration with Deck 13 (Plex-CI)** — competitor positioning inputs ## Audience **Primary:** Chief Commercial Officer + Sales Ops Lead + Sales Director. **Secondary:** AMs + Bid team members. **Tertiary:** Legal — clause-library curation participation. ## Reference - Memory: **Anthropic Claude** Sonnet 4.6 / Opus 4.7 for response generation - Memory: **Deck 13 Plex-CI** — competitor positioning for RFP context - Memory: **Deck 18 Briefing Room** — customer health and decision-maker context - Memory: **Deck 02 PPSS** — pricing source-of-truth when live - Industry references: Loopio, Responsive (formerly RFPIO), Qvidian — packaged RFP-response platforms (typically $50k-$200k+/yr for enterprise tier; OMX-fit-for-purpose build avoids that ongoing licence) --- ## Research deepening (background-agent, 2026-06-28) ### RFP-response platform vendor pricing (verified 2026) | Vendor | Tier / what's included | Public pricing | Notes | Source | |---|---|---|---|---| | **Loopio** | Essentials / Advanced / Enterprise; AI Magic, Library, intake projects | USD 14-25k/yr (small team, ~5 users) → USD 50-120k/yr enterprise (15-30 seats) | Per-user + per-content-volume bands; AI Magic add-on adds ~20% | https://www.loopio.com/pricing/ ; G2 reports | | **Responsive (formerly RFPIO)** | Foundation / Premier / Enterprise; AI-Assistant, SmartCompose | USD 18-30k/yr starter → USD 80-200k/yr enterprise | Their LLM-Powered AI Recommendation engine is the leading-edge feature | https://www.responsive.io ; TrustRadius reviews | | **Qvidian (Upland)** | Core / Plus / Enterprise; content library + SFDC integration | USD 30-60k/yr typical, up to USD 150k for large bid teams | Heavier sales/proposal management focus | https://uplandsoftware.com/qvidian/ | | **Ombud** | Mid-market RFP-response; lighter footprint | USD 15-50k/yr | https://www.ombud.com | | **Arphie / Aerial AI / Wisaw** | New-gen AI-native (2024-25 entrants); LLM-first | USD 12-40k/yr | Aggressive pricing; Arphie raised Series A 2025 | https://www.arphie.ai ; https://aerialai.com | | **AutogenAI** | UK-origin, AI-native RFP; strong in govt/public-sector bid | USD ~40-100k/yr | https://autogenai.com | **Anthropic build alternative (OMX-fit):** - Claude Sonnet 4.6 API: ~USD 3/M input tokens, USD 15/M output. Typical RFP response = ~30k input (RFP + retrieved knowledge) + 8k output. Cost per RFP draft ≈ USD 0.21. At 200 RFPs/year = USD ~42 in raw inference cost. - Engineering build cost: estimate USD 80-150k one-off + USD 25k/yr maintenance. Break-even vs Loopio enterprise tier inside year 1. ### Win-rate uplift evidence - **Loopio "State of the RFP" Report 2024** — customers report 51% higher response capacity, 38% more proposals submitted, win-rate uplift averaging 7-12% on tracked deals. (https://www.loopio.com/state-of-rfp-2024/) - **Responsive "Strategic Response Management Report 2024"** — 85% of customers report improved win rate; median ~10% lift; top-quartile 18%+. Time to first draft compressed from 15 to 4 hours median. (https://www.responsive.io/state-of-strategic-response-management/) - **APMP (Association of Proposal Management Professionals) benchmark** — automation tooling associated with 5-15% Pwin uplift on competitive deals; first-pass quality is the biggest driver. (https://www.apmp.org) - **Forrester TEI Loopio study (2023)** — three-year ROI 322%, payback ~6 months; majority of value from sales-team time reclaimed not headcount removal. (Forrester Wave: Configure-Price-Quote / Proposal automation) - **NZ context** — Government Procurement Rules require structured responses; NZ public-sector ROIs (GETS notices) have +5-8% win rate cited in MBIE supplier surveys when bid teams use structured response libraries. ### Knowledge-base architecture patterns - **Retrieval-Augmented Generation (RAG) over chunked-response-library:** - Chunking: 300-800 token windows with metadata (RFP-date, customer-segment, win/loss outcome, owner, last-reviewed) - Embedding model: Anthropic doesn't ship a hosted embedding model — pair Claude with OpenAI `text-embedding-3-large` (USD 0.13/M tokens, 3072 dim) OR Cohere `embed-v4` (USD 0.12/M, 1024 dim, multilingual) - Vector DB options: pgvector on existing Postgres (cheap, in-stack), Pinecone (~USD 70/mo starter → 5k/yr enterprise), Weaviate (open-source, self-host) - **Content governance:** content owner per chunk; expiry/review date; usage telemetry (how often this answer was used + win-rate when used) - **Confidence scoring:** match score from retriever + freshness penalty + win-rate weight → composite confidence ### Sources cited (PR-013) - https://www.loopio.com/pricing/ - https://www.loopio.com/state-of-rfp-2024/ - https://www.responsive.io - https://www.responsive.io/state-of-strategic-response-management/ - https://uplandsoftware.com/qvidian/ - https://www.ombud.com - https://www.arphie.ai - https://autogenai.com - https://www.apmp.org - https://platform.openai.com/docs/guides/embeddings - https://cohere.com/embeddings - https://docs.anthropic.com/en/docs/build-with-claude/embeddings - https://www.procurement.govt.nz/procurement/principles-charter-and-rules/government-procurement-rules/ --- ## Vectors + visuals ### Lucide icon choices (Ask Max set compatible) - **Same questions over and over:** `repeat-2` / `history` - **Knowledge base:** `library-big` / `database-zap` / `book-marked` - **AI draft generation:** `sparkles` / `wand-2` / `pen-line` - **Pricing review:** `calculator` / `badge-dollar-sign` / `scale` - **Voice / brand consistency:** `mic-vocal` / `palette` / `fingerprint` - **Win/loss loop:** `trophy` / `target` / `refresh-cw` - **Confidence + citation:** `shield-check` / `quote` / `badge-check` - **Time saved:** `timer-off` / `gauge` - **Sales-ops UI:** `layout-panel-left` / `inbox` ### Image concepts (NZ context, 4-6) 1. **Cover** — Sales-ops analyst at OMX office (Auckland skyline subtle in window); left monitor showing a tendered RFP PDF (NZ Govt All-of-Government style), right monitor showing the draft-response app with green confidence chips. Clock 11:00 — "RFP came in 9am, draft done by lunch." 2. **Problem vector grid (6)** — tile per pain (Same-questions / Inconsistent-voice / Pricing-shortcut / No-memory / Win-loss-lost / Time-hostage), each with Lucide icon + 1-line subtext. 3. **Knowledge base taxonomy** — radial chart: Past Responses, Customer FAQs, Product Specs, Pricing Matrix, Contract Clauses, Security/Privacy answers, Sustainability statements, Case studies. Show counts (e.g. "1,847 past Q/A pairs ingested"). 4. **AI-drafted walkthrough strip** — 4 frames: RFP doc upload → questions extracted (highlight pen) → answers proposed with confidence chips → sales-ops reviews + ships. Filmstrip layout. 5. **Win-rate evidence chart** — bar chart citing Loopio/Responsive 7-12% / 10% / 18% top-quartile uplift; OMX target band annotated. 6. **NZ government tender example** — masked GETS-style notice + draft response side-by-side; Lucide `landmark` to imply public-sector capability. Reinforces Government Procurement Rules compliance.