Get in touch

Automating Bookkeeping with AI: Building Parpro’s Scalable Accounting Solution

Parpro
Startup
Prototype
AI/ML
2025
by Do Well Physio

An AI-powered bookkeeping tool that automates journal entries, built to scale accounting workflows.

proximahq.io
boonmovement.com
The Opportunity
Parpro came to us with a clear problem: bookkeeping is slow, manual, and difficult to scale.
Creating journal entries from receipts and bank statements required extensive data entry, categorization, and verification. Existing workflows were repetitive and time-consuming, limiting how efficiently accountants could operate.

They needed something focused: a way to automate document processing and generate accurate journal entries, while maintaining control and reliability.

The Opportunity

The Opportunity
Parpro sought to simplify the tedious, manual processes involved in bookkeeping, specifically the creation of journal entries from financial documents such as receipts and bank statements. Traditional methods involved extensive manual data entry, categorization, and verification, significantly limiting the scalability and efficiency of bookkeeping operations. Parpro approached us with the ambitious goal of leveraging Large Language Models (LLMs) to automate these tasks and reduce the manual workload for bookkeepers and accountants.
Go Oil employee doing a tire change.
Go Oil employee doing an oil change.
Go Oil employee pose in front of go oil's service Van
Improving and rebuilding GoOil's tech infrastructure without disrupting business operations

Defining our MVP

We developed a streamlined solution using Gemini, integrated with AWS S3 for secure file storage.

Automated journal entry generation
Use LLMs to process receipts and bank statements, turning raw financial documents into structured journal entries.
Batch processing at scale
Allow users to upload and process up to 100 documents at once, reducing repetitive manual input.
Customizable accounting logic
Enable users to define categorization rules and prompts tailored to vendors and transaction types.
Built-in validation
Layer automated checks and AI self-correction to ensure entries are balanced, accurate, and reliable.
What we built
We developed a streamlined solution using Gemini, integrated with AWS S3 for secure file storage. The solution included a user-friendly web interface allowing users to upload batches of up to 100 financial documents simultaneously and input custom categorization rules tailored to specific vendors and transaction types. Once uploaded, Gemini processed each receipt or bank statement to generate accurate journal entries.

The system employed multiple layers of validation:
audit icon
Initial journal entry generation by Gemini
coding icon
Automated manual checks to ensure entries balance (debits equal credits) and referenced accounts exist
backend icon
Gemini-powered error identification and correction based on the initial automated checks
backend icon
Self-assessment by Gemini to gauge confidence in entry accuracy
Prototype Architecture Diagram
Diagram illustrating the new backend infrastructure developed for Go Oil.​
1. The original infrastructure of Go Oil utilizes an entire suite of 3rd-party offerings, which lacked flexibility and was costly.
2. Our first step was rebuilding a new Front-End UI, which solved the immediate UX issues with the customer booking flow.
3. The next step was to deprecate the Back-End of their 3rd party vendor, while GoOil continued operating their business without disruptions.
4. Once all the components were built, we were able to switch over to a completely custom solution and sever all dependencies to the 3rd party vendor.
Improving and rebuilding Go Oil's tech infrastructure without business interruptions

The Outcome (So Far)

01

Immediate Efficiency Gains

Reduced the manual workload of bookkeepers and accountants by automating repetitive data entry and initial categorization tasks.

02

Improved Accuracy

Achieved a 65% accuracy rate in automated journal entry generation, significantly improved from an initial 25-30%.

03

Robust Validation System

Created an integrated, multi-step validation approach combining AI self-diagnosis with manual checks, effectively balancing automation with accuracy.

04

Built to Scale

Identified future enhancements such as further prompt refinement and integration of traditional coding methods, ensuring continuous improvement in accuracy and scalability for large-scale operations.

“I worked with the Invoke team at a critical juncture in our business. They helped us scope and build an exploratory proof of concept that now serves as the foundation of our product. The entire team was professional, collaborative, and brought real value through their expertise. I'm looking forward to continuing our partnership."

John Marshall , Co-Founder of Parpro

Outcome
Improving and rebuilding GoOil's tech infrastructure without disrupting business operations
A Growing Product Ready to Scale - Go Oil were able to pivot to a subscription model for their franchisees, because of the tech foundations we had implemented. We were able to add a subscription model within a few weeks
Identified Market Clarity - Successfully pivoting from a residential customer base to a commercial, fleet-based enterprises
Rapid Growth & Investment  - From our first engagements with GoOil, they have grown across Canada and US markets, successfully scaling their business and securing funding from Investors including, the League of Innovators (LOI)
Long Term Partnership with Invoke - Without a technical team to start, Go Oil engaged Invoke as their main development partner, and only recently hired their first developer

  1. Significant Efficiency Gains - Reduced the manual workload of bookkeepers and accountants by automating repetitive data entry and initial categorization tasks.

  2. Customizable Accuracy - Provided mechanisms for users to customize prompts and categorization rules, allowing tailored accounting processes for different client needs.

  3. Robust Validation System - Created an integrated, multi-step validation approach combining AI self-diagnosis with manual checks, effectively balancing automation with accuracy.

  4. Strategic Roadmap for Growth - Identified future enhancements such as further prompt refinement and integration of traditional coding methods, ensuring continuous improvement in accuracy and scalability for large-scale operations.

Measurable Impact

Improved Accuracy

Achieved a 65% accuracy rate in automated journal entry generation, significantly improved from an initial 25-30%.

High Value Processing

Enabled simultaneous processing of batches up to 100 receipts, dramatically reducing manual input requirements.

Scalable Solution

Demonstrated potential to scale further by refining prompts and strategically offloading specific tasks from the LLM to traditional coding methods, moving towards the client’s targeted 99% accuracy.

I worked with Dominic and Wes at a critical juncture in our business. They helped us scope and build an exploratory proof of concept that now serves as the foundation of our product. The entire team was professional, collaborative, and brought real value through their expertise. I'm looking forward to continuing our partnership.

John Marshall

Co-Founder

More case studies

Operator Technologies

Prototype, MVP, AI/ML

Putting life-saving gas leak detection data in the hands of Critical Facility Technicians

Toyota Signal

Design Concept, Prototype

We co-created and launched a Customer Insights platform with Toyota Dealers across the United States

Proxima HQ

Design Concept, Prototype, MVP, Optimization

Building a digital sales centre experience that raised $1 million in seed funding

* Invoke is a digital innovation studio, working with founders, entrepreneurs and established brands to design and build digital products and solutions.
Home
Work
Blog
Invoke Media Inc.
333 Seymour Street, 8th Fl.
Vancouver BC
V6B 5A6
Canada
604.398.3600