AI-Native Reverse Auction Market

01 Problem

In fragmented B2B markets (construction materials, auto parts, agriculture, wellness, etc.), the purchasing process is broken:

Friction

Buyers spend hours requesting quotes manually via WhatsApp, phone calls, and emails — one supplier at a time.

Opacity

No reliable reference prices. Pricing varies by region, stock, urgency, and who's asking. Information flows through word-of-mouth, not markets.

Inefficiency

Sellers waste time responding to inquiries that never convert. No qualification or intent signal.

No Market Infrastructure

No good marketplaces exist for most verticals. No good markets means no good information. Everything runs on relationships, not data.

The result: both sides lose. Buyers overpay or under-compare. Sellers compete blindly.

02 Solution

A reverse auction platform powered by AI agents that flips the traditional marketplace model:

1

Buyer posts a need (material list, parts request, purchase order) — the system normalizes it automatically.

2

Seller Agents compete in real-time — each representing a real vendor — to offer the best consolidated bid (price + shipping + delivery time).

3

Buyer receives a comparative table and selects the best offer.

What makes this different

Traditional Marketplace	This Platform
Seller uploads catalog manually	AI agent connects to seller's existing inventory
Buyer searches and compares	Buyer posts need, agents compete
Static pricing	Dynamic, real-time bidding
Disputes handled manually	Cryptographic verification of fulfillment
Identity required	Anonymity layer available
Price data is public list prices	Closing prices are the real dataset

03 How It Works

Buyer                    Platform                     Sellers
  |                         |                            |
  |-- Upload list/photo --> |                            |
  |                         |-- AI normalizes demand --> |
  |                         |-- Broadcast to Seller   -->|
  |                         |   Agents                   |
  |                         |                            |-- Agent checks
  |                         |                            |   inventory
  |                         |                            |-- Agent generates
  |                         |                            |   bid (price +
  |                         |                            |   shipping + ETA)
  |                         |<-- Bids collected ---------|
  |<-- Comparative table ---|                            |
  |                         |                            |
  |-- Select winner ------->|                            |
  |                         |-- Confirm order ---------->|
  |                         |--- Dispute resolution -----|

Three Agent Types

Phase 1

Buyer Agent

Receives raw input (text, photo, PDF, voice)
AI normalizes into structured product list
Broadcasts to suppliers via WhatsApp/email
Generates comparative table

Phase 2

Seller Agent

Connects to existing inventory (Excel, API, photo)
Auto-responds with structured bids
Negotiation algorithm (volume, proximity, urgency)
Learns from win/loss data

Phase 2+

Dispute Agent

Monitors fulfillment
Cryptographic verification of events
Reputation system with trust scores
Future on-chain integration

04 Vertical Selection Framework

Rather than committing to a single vertical upfront, we use this framework to evaluate the best initial niche:

Evaluation Criteria

1Buyer recruitment ease — Can we easily recruit initial buyers? Do we personally know people who buy in this niche?

2Feedback loop quality — Frequent opinions, fast delivery cycle, recurrent orders?

3.1Inefficiency from opacity — Major lack of transparency and automation?

3.2Personalized orders — Benefit from customized/tailored orders?

3.3Anonymity benefit — Would buyers prefer to purchase anonymously?

3.4Cash payment benefit — Meaningful discount or incentive for paying in cash?

Candidate Verticals

Criterion	Construction	Auto Parts	Wellness	Discreet Products
Buyer recruitment	High	Medium	Medium	Medium
Feedback loop	Medium	High	High	High
3.1 Opacity	Very High	High	Medium	High
3.2 Personalization	High	High	Medium	Low
3.3 Anonymity	Low	Low	Medium	Very High
3.4 Cash benefit	Very High	High	Low	Medium
Overall	Strong: 1, 3.1, 3.2, 3.4	Strong: 2, 3.1, 3.2	Strong: 2	Strong: 2, 3.3

Decision to be made after initial user interviews. The framework guides the conversation, not a predetermined answer.

05 Business Model

Revenue Stream	Description	Phase
Transaction commission	% of each closed transaction	Phase 1+
Seller Agent SaaS	Subscription for sellers to deploy their own AI agent — an app that helps sellers show up, compete, and win negotiations	Phase 2+
Qualified leads	Charge sellers for high-intent buyer leads (pre-normalized, ready to buy)	Phase 1+
Seller tooling	Analytics dashboard, pricing intelligence, competitive insights	Phase 2+
Closing price data	Anonymized market data sold to industry players	Phase 3+

06 Moat & Defensibility

Layer 1: Network Effects

More agents negotiating → more transactions → more data → better prices for buyers → more buyers join → more sellers deploy agents.

Layer 2: Closing Price Data

The most valuable asset is real closing prices — what was actually paid, with what conditions, for what volume, in which region. In an inflationary economy, this data is especially powerful.

Layer 3: Dispute Resolution Infrastructure (Core Moat)

The strongest long-term defensibility: cryptographic verification of fulfillment, camera integration for proof at delivery, tamper-proof reputation systems, on-chain records for critical transactions.

"The greatest defensibility comes from dispute resolution. I envision cameras, reputation systems, and cryptographic integration to 'prove' something actually happened — fulfilling what was agreed with the consumer. That verification structure is a strong defense." — N.S.

Layer 4: Anonymity

For certain verticals, buying anonymously while still having dispute resolution and trust guarantees is a unique value proposition no traditional marketplace offers.

07 Roadmap

Months 1-3

Phase 1: Buyer Agent MVP

Upload list → AI normalizes → auto-sends to suppliers → comparative table. No seller onboarding required.

Parallel

Phase 2: Seller Agent Integration

Seller Agent connects to inventory, auto-responds, negotiation algorithms, SaaS dashboard.

Parallel

Phase 3: Ecosystem Expansion

Open agent deployment, new verticals, protocol-level value exchange, dispute resolution as product.

"I expect horizontal expansion (Phase 2) and vertical expansion (Phase 3) will run in parallel rather than sequentially. Priority will depend entirely on user feedback." — N.S.

08 Tech Stack

Layer	Technology	Purpose
AI / LLM	GPT-4o / Gemini 1.5 Pro	Agent reasoning, normalization, negotiation
Agent Framework	LangGraph	Multi-step workflows, state management
Channels	WhatsApp Business API (Twilio)	Primary communication with sellers
Database	Supabase (Postgres)	Transactional data, user data, agent state
Search	Pinecone	Semantic product matching
Automation	Python + Browserbase	Web scraping, price capture, outreach
Auth	Supabase Auth	User authentication
Hosting	Vercel / Railway	Deployment

Stack may shift depending on chosen vertical. — N.S.

09 MVP Spec — Buyer Agent

As a buyer (e.g., a contractor), I want to upload my material list and receive comparative quotes from multiple suppliers without having to contact each one individually.

Input Methods

Text list — free-form (e.g., "50 bolsas de cemento, 200 ladrillos huecos")
Photo — handwritten or printed material list
PDF — bill of quantities
Voice note — describing what's needed (future)

Key Decisions

No seller app — reach sellers on existing channels
No seller login — they respond via WhatsApp
Simple buyer web UI — upload list → see comparison
Curated suppliers — manually onboard 10-20 initially

Processing Pipeline

Step 1

Parse input

OCR / transcription

→

Step 2

Normalize

AI standardization

→

Step 3

Match sellers

Category + location

→

Step 4

Broadcast

WhatsApp / email

→

Step 5

Collect bids

Parse responses

→

Step 6

Compare

Generate table

Success Metrics (Phase 1)

100

Quote requests
first month

>40%

Supplier
response rate

>50

Buyer NPS

>2h

Time saved
per request

>15%

Conversion to
transaction

10 Open Questions

Dispute resolution design: How exactly would crypto-based verification work? Simplest version for MVP?
Naming: "Market de contratos", "AI-native market", "Agent-focused market"? Test with users.
Vertical selection: Run user interviews using the framework before committing.
Anonymity layer: How to implement anonymous purchasing while maintaining dispute resolution?
Regulatory: Licensing or compliance requirements depending on vertical?
WhatsApp Business API: Rate limits, template approvals, multi-device support at scale.
Pricing intelligence: Can we build a price index from closing data? How valuable?