ProcurePilot

AI-driven procure-to-pay for the enterprise: structured demand, governed approvals, then PayPal Checkout to capture payment and record ledger events end-to-end.

Comment

Inspiration

ProcurePilot started with a problem we encountered while developing and operating our own ERP system.

Our ERP already handles procurement workflows in production, including purchase requisitions, multi-level approvals, purchase orders, goods receipts, and master data connected to sites and inventory.

However, we noticed a recurring gap in everyday business operations. Smaller purchases, such as office equipment, software subscriptions, or spare parts, often bypass formal procurement processes. Employees resort to emails, spreadsheets, or personal payment methods because traditional procurement workflows can feel unnecessarily cumbersome for relatively straightforward purchases.

Yet, the need for financial accountability doesn't disappear just because a purchase is small. Businesses still need spending limits, budget controls, authorization, and proper audit trails.

We saw an opportunity to bridge this gap: What if employees could simply describe what they needed, while the system handled the procurement intelligence and controls behind the scenes?

That question led to ProcurePilot.

What It Does

ProcurePilot is a standalone procurement application built from an existing ERP procurement module, extended with an AI-assisted purchasing experience and PayPal integration.

It combines traditional procure-to-pay workflows (PR → PO → GRN), role-based access control, approval workflows, procurement policies, and ERP-style master data with a more intuitive purchasing interface.

Its AI Buy workflow allows users to:

  • Describe purchasing requirements in natural language.
  • Convert requests into structured procurement requirements.
  • Compare available supplier offers.
  • Understand purchasing trade-offs through AI-generated explanations.
  • Proceed with purchases through PayPal Sandbox when policy conditions are satisfied.

ProcurePilot follows a fundamental principle:

AI prepares and explains. Application logic validates. Humans authorize. PayPal executes.

AI recommendations never independently authorize spending. Deterministic application logic validates applicable spending limits and available budgets before checkout is enabled. Human approval remains an essential part of the purchasing process.

How We Built It

Rather than creating a procurement application from scratch, we extracted the existing procurement module from our ERP into a standalone application.

This gave us an established foundation consisting of authentication, role-based access control, sites, items, suppliers, purchase requisitions, purchase order workflows, and goods receipt closure.

We consolidated the existing database migrations into a standalone schema, preserving the core procurement architecture while removing dependencies on the larger ERP application.

On top of this foundation, we developed the hackathon-specific components: AI Buy, procurement policy integrations, spend auditing, and a PayPal payment ledger.

Technology Stack

  • Frontend: React, Vite, AG Grid for supplier offer comparison.
  • Backend: Node.js and Express.
  • Database: PostgreSQL.
  • AI: Ollama with an OpenAI-compatible API, supported by a heuristic fallback.
  • Supplier Offers: Channel3 integration when enabled, with seeded demonstration data as a fallback.
  • Payments: PayPal JavaScript SDK with server-side order creation and capture, linked to procurement purchase records.

The architecture keeps AI-assisted decision-making separate from deterministic business rules and payment execution. This allows us to introduce natural-language interactions without compromising the underlying procurement controls.

Challenges We Faced

One of our biggest challenges was extracting an established ERP module rather than building a new application from a blank slate.

The procurement module was part of a larger system, so separating it required identifying its core dependencies, consolidating database migrations, re-establishing access controls, and preserving the integrity of existing procurement workflows.

Another challenge was balancing two different product experiences: a comprehensive ERP-style procurement application and a focused demonstration of the AI-assisted purchasing journey.

We wanted to showcase the complete product without overwhelming users with enterprise-level complexity.

Reliability was equally important. We didn't want the demonstration to depend entirely on locally hosted AI models, external API availability, or live supplier integrations. We therefore introduced heuristic extraction, seeded supplier offers, and PayPal Sandbox support to keep the primary workflow demonstrable even when external dependencies were unavailable.

Accomplishments We're Proud Of

  • A real procurement foundation: We extracted an existing ERP procurement module into a runnable standalone application instead of building a collection of mock interfaces.
  • AI integrated into business workflows: AI Buy connects to actual procurement records and approval processes rather than functioning as an isolated chatbot.
  • An end-to-end purchasing journey: We connected natural-language intake, supplier comparison, policy validation, human approval, and PayPal payment execution.
  • Business controls preserved: Budget enforcement, authorization, and transaction traceability remain central to the application.

Our biggest accomplishment is bringing together an existing enterprise workflow and a new AI-assisted experience without treating them as two disconnected systems.

What We Learned

Building ProcurePilot reinforced an important lesson: AI is most valuable in enterprise applications when it reduces complexity without weakening the systems of control underneath.

Extracting an existing ERP module also taught us to distinguish between foundational capabilities and unnecessary complexity. Approvals, authorization, master data, and auditability are essential to procurement, but not every feature of a larger ERP needs to be carried into a focused standalone product.

We also gained a deeper appreciation for the relationship between purchasing decisions and financial transactions. Order creation, authorization, payment capture, and procurement state must remain consistent if businesses are expected to trust automated purchasing workflows.

Ultimately, we learned that introducing AI into an established business process is less about replacing existing systems and more about making them significantly easier to use.

What's Next

ProcurePilot is currently a hackathon prototype built on top of an extracted, established procurement module. Our next step is to move beyond the demonstration and strengthen it into a more production-ready product.

Our priorities include improving authentication and operational reliability, simplifying the default experience around guided purchasing, implementing production PayPal integration with webhooks, and strengthening transaction reconciliation.

We are also exploring whether ProcurePilot should evolve as an independent procurement product or become a complementary purchasing interface within our larger ERP ecosystem.

Our long-term vision is straightforward:

Make business purchasing as simple as describing what you need, while maintaining complete control over how the money is spent.

Built With

Share this project:

Updates

Submission history