Categories: FAANG

How FibroGen achieved 40x ROI by automating invoice processing

1 Fibrogen Invoice Automation Blog D.max 1000x1000 1

The small accounts payable (AP) team at FibroGen, a biotech focused on developing therapies for chronic and life-threatening conditions, has been inundated with vendor invoices. Despite a recent SAP implementation that streamlined many operations, invoice processing has remained stubbornly manual and time-consuming.

FibroGen receives about 1,000 invoices per month. With a two-person AP team and an average of five minutes to process an invoice, roughly 500 hours are spent per year per person – or about 25% of total working hours – processing invoices.

Yet at its core invoice processing is just data entry with some important but straightforward business rules, like differentiating between invoices and credit memos. Invoices are received by email, typically as a PDF attachment, downloaded to a shared folder, then manually entered into SAP.

Needless to say, the process is ripe for automation with significant potential benefits to productivity, error prevention, and employee satisfaction.

Discovering Document AI

Fibrogen’s VP of IT Parag Govil is laser-focused on strategic cost cutting through innovation, including AI. After learning about Google Cloud Document AI and its Invoice Parser, Parag decided to pilot the service in collaboration with the AP team. However, FibroGen was new to Google Cloud and lacked a dedicated technical resource to implement the solution.

To fill the gap, Parag enlisted Fibrogen’s SAP partner, Deloitte, and Google Cloud to help design and implement a solution.

Driving business outcomes with AI can be daunting. The key is to start somewhere and prioritize ROI. By automating invoice processing with Document AI, we’re targeting a meaningful but finite efficiency gap. More importantly, we’re opening the door to business transformation using AI.

Parag Govil, VP IT, FibroGen

Building the right solution for FibroGen

Given the lean FibroGen team and the novelty of relying on AI for a critical business function, the solution had to meet several key requirements:

Simplicity: To allow FibroGen to deploy and eventually manage the solution, the approach needed to be easy to understand and adapt, with as few moving parts as possible.
Oversight: Developing trust in the solution, including its ability to handle the inevitable edge cases, is critical. To achieve this over time, the solution needed to support and eventually minimize human-in-the-loop.
Cost: The solution needed to provide significant ROI to justify the upfront implementation investment.

Additionally, the solution needed to integrate with the existing systems: email (Outlook) and SAP.

With these requirements, the team designed a solution with Document AI at the core, but also relying on:

The AP team to download invoices from email to a Google Cloud Storage bucket (with a phase 2 plan to automate)
Triggered Cloud Functions to invoke Document AI and handle its output
SAP Cloud Integration to retrieve the output, enforce key business rules, and upload to SAP R/4

Additionally, FibroGen decided to integrate a human review step, routing invoices for AP team approval when a configurable threshold isn’t met.

Solution components

The below diagram captures the key solution components, including:

An input GCS bucket where the AP team uploads invoices
An “invoker” Cloud Function, triggered by invoice upload, that calls the Document AI API to process an invoice
A Document AI processor, based on the Invoice Parser (another option could have been the gen AI-powered Document AI Custom Extractor)
A “handler” Cloud Function, triggered by JSON-formatted processor output, that forwards the output to SAP Cloud Integration
An SAP Cloud Integration flow that listens for the output, applies validations and other business rules, and uploads to SAP R/4

Results and what’s next

The entire process, from design through production deployment, was completed in under three months. Though initially the solution will require oversight from the AP team, over time, the parser will be further uptrained on FibroGen vendor examples, eventually reducing if not eliminating the need for human involvement.

The total run cost of the solution is estimated at $150 per month, including Google Cloud support fees. With the potential to free up 25% of the AP team’s bandwidth, the solution provides an extraordinary estimated 40x return on investment. Furthermore, the AP team will be able to redirect this time toward higher value, and more interesting, activities.

Given the success of the invoice pilot, FibroGen is planning to expand Document AI to other procurement use cases.

To learn more about Document AI and how it can help your business, start here.