AI-Powered Quotation Automation & Approval System
Designed and implemented an end-to-end quotation workflow that transforms incoming email requests into structured quote records, supports AI-assisted revisions and human approval, synchronizes approved quotations with the business system, and delivers finalized quotes back to customers.
Role
AI Automation Engineer & Systems Architect
Tech Stack
▸High-level orchestration canvas: incoming mailbox trigger, validation, multi-modal Gemini document processing, persistent state registration, approval gating, and downstream delivery.
The Challenge
Quotation requests arrived through email in inconsistent formats and often included forwarded messages, attachments, incomplete specifications, or information embedded in free-form conversational text.
Handling this manually forced commercial operators to spend hours deciphering specs, re-typing line items into business software, and coordinating adjustments back and forth. Crucially, fully autonomous quote dispatch was unacceptable: commercial commitments demand human accountability.
- ✕Details had to be interpreted manually across fragmented email threads
- ✕Attachments required separate inspection and measurement extraction
- ✕Quote status was difficult to track centrally without dedicated database state
- ✕Corrections required repetitive manual re-editing of pricing line items
- ✕Approved quotations required double-entry into the commercial system
Unstructured Requests
Critical project details arrived inside free-form email conversations and forwarded threads with inconsistent formats.
Attachment Dependency
Crucial specifications, measurements, and requirements existed inside attached documents and images rather than the email body.
Fragmented Quote State
Review, approval, revision history, and delivery status lacked a single unified, reviewable operational record.
Human Approval Requirement
Commercial quotations have direct financial and contractual commitments and could not be transmitted autonomously.
System Handoffs
Approved quotation parameters needed to be synchronized reliably into the downstream quotation environment without manual data entry.
Operational Reliability
Duplicate mailbox processing, webhook drops, and transient workflow interruptions needed to be controlled deterministically.
End-to-End System Architecture
An eleven-stage architecture spanning autonomous ingestion, structured state persistence, human review with AI assistance, and auditable downstream execution.
Quotation Orchestration Topology
Ingests raw emails, analyzes attachments, generates draft JSON schema, and persists state.
Commercial manager inspects quote, triggers conversational revisions, and signs off.
Synchronizes quote with external business systems, generates PDF, and delivers email.
The system intentionally isolates automated interpretation from commercial commitment. External APIs and client delivery only fire after deliberate human approval.
▸Multi-stage quotation orchestration workflow in n8n showing routing nodes, Gemini API calling, conditional branches, error handling, and transactional database updates.
Intelligent Request Intake
The workflow begins by preparing incoming quotation requests for reliable downstream processing. Rather than blindly executing code on every unread message, the system monitors dedicated mailbox channels via IMAP and applies intent classification.
Inquiries with forwarded communication histories, multiple recipient threads, or mixed attachment types (PDFs, CAD drawings, spec sheets, PNG renders) are isolated. The intelligence layer parses both body copy and attachments to synthesize a structured quotation context.
Isolates the actual customer inquiry from email client headers, prior signatures, and internal correspondence.
Inspects attached documents and images to extract item dimensions, quantities, and specific material codes.
“Automation begins with interpretation, not execution.”
The system first determines what the incoming communication represents before allowing downstream quote-processing steps to continue. Inquiries failing qualification are tagged for manual sorting rather than triggering erroneous calculations.
Persistent Quote Operations
Once extracted, each processed quotation receives a structured operational record. Rather than relying solely on ephemeral workflow execution memory, state is persisted in an external database where every proposal remains reviewable, filterable, and trackable over time.
Visible data points indexed for each transaction include unique quote references, client profile data, project summaries, computed itemized totals, generation timestamps, approval status flags, and final delivery verification.
▸Centralized operational quotation registry: tracking unique quote references, client records, project scopes, totals, generation dates, and real-time approval states.
Human-Controlled Approval
The system automates preparation and orchestration but keeps the final commercial decision with a human operator. Quotations represent binding financial agreements that directly affect profit margins, contractual liabilities, and client goodwill.
The mobile-friendly review interface gives managers immediate visibility into quotation states: All, Pending, Approved, and Refused. Operators can inspect line item breakdowns, unit prices, and attached source drawings before allowing any downstream execution.
▸ AI prepares.
▸ Automation coordinates.
▸ Humans authorize.
▸Operator review interface: filtering quotes by Pending, Approved, and Refused with itemized pricing inspection.
AI-Assisted Quote Revision
▸Conversational editing assistant: manager inputs plain-text revision prompt; AI updates schema without manual spreadsheet math.
Quotations frequently require adjustments before authorization: adding line items, recalculating sizing options, altering material types, or offering special volume discounts. Instead of forcing managers to manually recalculate complex pricing equations, the system provides an AI Revision Assistant.
Managers describe requested changes through natural-language instructions (e.g. “add a 15% discount to the second item and update delivery date to Friday”). The assistant reinterprets the request, adjusts line item math, and presents the revised draft for verification.
Controlled Business-System Synchronization
Once a quotation passes affirmative human approval, structured quotation details synchronize with the downstream quotation platform. This eliminates double-entry friction while keeping the core commercial database as the single source of truth for business operations.
The integration carries operationally relevant information including quotation line items, quantities, pricing tiers, material descriptions, optional add-ons, grand totals, and internal cross-references.
▸API synchronization canvas: validating approved quotation payload, authenticating against downstream quotation system, creating line items, and updating synchronization state.
Automated Customer Delivery
Following human approval and business system synchronization, the workflow generates a consistent client-facing delivery email and attaches the formalized quotation document when ready.
The dispatch engine ensures clean recipient resolution, merges branded HTML email templates with itemized pricing summaries, attaches generated PDF quotations, and records delivery timestamps into persistent state for audit trail compliance.
✓ Delivery status updated to 'Sent' in operational registry
▸Final customer delivery interface: formalized quotation dispatch with clean summary, dynamic client recipient handling, and official PDF quotation attachment.
Built for Operational Reliability
Production environments encounter network timeouts, malformed emails, and concurrent triggers. The architecture incorporates enterprise reliability patterns to guarantee deterministic execution.
Persistent State
Quote records survive individual workflow runs. Transient execution memory is backed into persistent operational storage.
Duplicate Protection
Idempotency controls and message-ID deduplication ensure repeated mailbox polling never spawns redundant quote items.
Attachment Preservation
Multi-modal specification files, drawings, and documents are ingested, cataloged, and linked throughout the quote lifecycle.
Controlled Execution
Downstream commercial actions occur only at their designated workflow stage upon affirmative human clearance.
Payload Recovery
Structured quotation payloads maintain strict schema contracts, allowing graceful reprocessing on transient connection drops.
Status Tracking
The system logs explicit operational transitions across pending review, revision, approval, sync, and client dispatch.
Human Decision Boundary
AI draft generation is strictly separated from commercial authorization. The LLM cannot independently dispatch quotes.
Modular Orchestration
Intake, multi-modal extraction, revision assistant, sync, and delivery run as decoupled, independently maintainable modules.
Engineering Decisions
Key architectural trade-offs that separate robust production systems from fragile automation scripts.
Human-Controlled AI
AI drastically reduces interpretation and drafting overhead, but commercially sensitive commitments require human oversight. The system positions AI strictly as an intake and revision copilot.
LLM outputs draft quotation JSON schema → stored with status 'Pending Review' → held until human reviewer signs off.
Persistent State Over Workflow Memory
Relying on transient workflow execution context makes systems fragile to timeouts and impossible to audit. Quotation records live in external persistent storage.
Airtable operational registry holds quote reference, line items, pricing tiers, and client context independent of n8n runs.
Stage-Based Orchestration
Monolithic automation pipelines fail completely when any single node breaks. The workflow decouples intake, extraction, review, sync, and delivery into discrete stages.
Sub-workflows communicate via webhooks and state events; failure in downstream dispatch does not lose parsed quote data.
Approval Before Execution
Commercial actions—such as customer delivery and database synchronization—must never occur preemptively or on draft estimates.
Downstream synchronization and client email triggers require an explicit 'Approved' status flag toggled by authorized personnel.
Structured Data Between Systems
Free-form email inquiries cannot be passed reliably into billing or ERP software. Raw inquiries are converted into deterministic, validated JSON objects.
Strict schema validation maps quantities, units, material descriptions, and pricing calculations into typed payloads.
Centralized Operational Visibility
Operators and executives need to monitor quote pipelines without parsing through complex automation backend logs.
A unified dashboard indexes every quote with current operational status, customer details, timestamps, and total value.
Business-System Integration
Automation creates actual operational value only when it seamlessly updates the company's core commercial system of record rather than maintaining a disconnected silo.
Approved quotation data is pushed via authenticated API directly into the downstream quotation environment.
Safe Customer Communication
Customer delivery requires professional formatting, verified recipient handling, and document attachments to prevent customer friction.
Pre-formatted HTML email templates merge dynamic quote summaries and attach generated PDF documents only after clearance.
What the System Changed
A qualitative shift from high-friction administrative overhead to structured, human-governed automation.
Before: Fragmented & Manual
- ✕Manual email interpretation & scattered inquiries
- ✕Separate inspection of PDF/image attachments
- ✕Fragmented quote tracking across personal inboxes
- ✕Repeated, error-prone manual line item editing
- ✕Manual re-entry into business quotation system
- ✕Manual customer delivery & follow-up logging
- ✕Limited operational visibility into pending pipeline
After: Automated & Supervised
- ✓Automated request intake & classification
- ✓AI-assisted multi-modal information extraction
- ✓Persistent, centralized quotation records
- ✓Dedicated human review & approval dashboard
- ✓Natural-language conversational revision assistance
- ✓Controlled downstream business-system synchronization
- ✓Automated final email dispatch with PDF attachment
- ✓End-to-end trackable quotation lifecycle and audit log
Tech Stack
Technologies orchestrated together to form a robust, fault-tolerant quotation engine.
Coordinates multi-stage pipelines, triggers, webhook gates, and error handlers.
Interprets unstructured email text, forwarded context, and attached documents.
Stores quotation records, line items, status flags, and revision histories.
Monitors mailbox accounts, filters spam, and routes quotation threads.
Pushes validated quotes, customer data, and line items to quotation software.
Ensures no commercial commitments are dispatched without managerial approval.
“This system demonstrates how AI automation can support real operational workflows without removing the controls businesses need. The result is not an autonomous black box, but a coordinated system where AI handles interpretation, automation manages execution, persistent state provides visibility, and humans retain authority over critical commercial decisions.”
Interested in a similar system for your operation?
Discuss a Project