Uber advances invoice document processing using GenAI with the TextSense platform
The problem
Uber's invoice processing relied on manual data entry and RPA that could not scale to diverse invoice formats, invoices arriving in over 25 languages, and high volumes, leading to high average handling time, errors, and rising operational costs.
First attempt
Existing Rule-Based Systems and RPA could not adapt to new invoice formats without manual rule-setting, failed to scale as Uber onboarded new suppliers and document formats, and required continual maintenance and manual error correction.
Workflow diagram · grounded in source
1
Invoice submission trigger
Trigger
Suppliers submit invoices via a self-service platform or by email.
▾ source quote
“There are two methods for suppliers to submit invoices”
2
Document ingestion
Integration
Documents from emails, PDFs, and the ticketing system are ingested and saved to object storage.
▾ source quote
“Integrates documents from sources like emails, PDFs, and ticketing systems, and saves files in an object storage platform. Supports structured and unstructured data formats.”
3
Pre-processing
Ai action
Image augmentation handles low-resolution scans and handwritten text, and documents are converted to a standard format.
▾ source quote
“Includes image augmentation to handle low-resolution scans and handwritten texts.Converts document formats (PDFs, Word documents, images) into a standard format suitable for processing.”
4
OCR text extraction
Ai action
Uber's Vision Gateway CV platform uses optical character recognition to extract text from document images.
▾ source quote
“Uses Uber's Vision Gateway CV platform for optical character recognition to extract text from document images.”
5
LLM data extraction
Ai action
LLM models extract specific data elements such as invoice numbers, dates, and amounts from the document.
▾ source quote
“Leverages trained or pre-trained LLM models for extracting specific data elements like invoice numbers, dates, and amounts.Continuously improves through periodic re-training and feedback loops to address accuracy issues and adapt to new document formats.”
6
Post-processing and validation
Validation
Business rules and post-processing steps refine extracted data, cross-referencing with databases for data quality.
▾ source quote
“Applies business rules and user-defined post-processing steps to refine extracted data before final use.Ensures data quality by cross-referencing with existing databases or predefined rules.”
7
HITL review
Human review
Users perform Human-in-the-Loop review using a side-by-side comparison UI of PDF data versus model-extracted data.
▾ source quote
“We designed a UI to enable users performing the HITL (Human in the Loop) review to do a side-by-side comparison of the PDF data versus the data extracted from the models”
8
ERP integration and payment
Integration
Approved invoices are sent to the ERP system for final approval and vendor payment.
▾ source quote
“the documents are processed as invoices and sent to the ERP system for approval and vendor payments”
9
Metrics and feedback loop
Feedback loop
Key performance indicators like processing speed and accuracy rates are captured and used to drive continuous model improvements.
▾ source quote
“Captures key performance indicators like processing speed, accuracy rates, and cost efficiency.Uses these metrics to drive continuous improvements.”
Reported outcome
The GenAI-powered TextSense system achieved a 2x reduction in manual invoicing, an overall accuracy rate of 90%, a 70% reduction in average handling time, and a 25-30% cost saving compared to the manual process.
TextSenseVision GatewayOCRNLPGPT-4CadenceRPALlama 2Flan T5ERP system
◆ Does this fit your context?
Compare to your context
Tell us your scale, team, and constraints. We'll show what changes at your size, what fails at your scale, and whether this case is a fit, needs adaptation, or won't scale to you. Free demo, no signup.
The GenAI-powered TextSense system achieved a 2x reduction in manual invoicing, an overall accuracy rate of 90%, a 70% reduction in average handling time, and a 25-30% cost saving compared to the manual process.
Existing Rule-Based Systems and RPA could not adapt to new invoice formats without manual rule-setting, failed to scale as Uber onboarded new suppliers and document formats, and required continual maintenance and manu…
How is this invoice processing AI workflow structured?
Invoice submission trigger → Document ingestion → Pre-processing → OCR text extraction → LLM data extraction → Post-processing and validation → HITL review → ERP integration and payment → Metrics and feedback loop.
This case is one data point. Whether its pattern fits you depends on your volumes, your stack, and your exception load — that comparison is the step no case study can do for you.