Concentrix needed a better way to manage rapidly growing invoice volumes from more than 100 utility providers, each using different invoice formats and layouts.
Concentrix built an intelligent document processing solution on Microsoft Power Platform and AI Hub to automate invoice extraction and processing at scale.
The solution now processes 100,000 invoices per month, improves extraction accuracy to as high as 99%, and reduces development effort significantly.
For more than 20 years, Concentrix has delivered fully integrated technology solutions to companies around the world. Today, the organization is trusted by more than 2,000 clients across major industries.
To find a better way to process and manage an increasing number of customer invoices, Concentrix built an intelligent document processing platform based on Microsoft Power Platform and AI Hub. The solution automatically processes 100,000 invoices each month, improves data extraction accuracy to as high as 99% and significantly reduces development time.
“The impact of this program has been transformative for our operations. By modernizing invoice processing with intelligent automation, we’ve unlocked new levels of efficiency, accuracy, and scalability across our utility partners.”
Kelly Desmond, Managing Director, EMS Business, Concentrix
20,000 unplanned service visits avoided
As a global services organization supporting clients across nearly every industry, Concentrix has built its reputation on helping customers operate efficiently. From banking and financial services to retail, telecommunications, healthcare, and education, the company delivers technology-enabled services designed to improve business performance and customer experience. Concentrix operates through a solution-focused, technology-powered, and intelligence-fueled business model that is designed to deliver end-to-end customer experience and business transformation services. The company supports this model by deploying solutions at a massive scale across more than 70 markets.
This breadth of responsibility means Concentrix routinely manages high-volume, business-critical processes on behalf of its customers. One of the most complex of these processes is invoice and utility bill processing for large utility providers, a domain characterized by extreme variation, regulatory nuance, and relentless volume growth. For Concentrix, modernizing this process was not simply an internal efficiency initiative—it was a strategic imperative to continue delivering value at enterprise scale.
“We work across all verticals—banking, retail, telecom, education, healthcare—and automation is a key part of how we help our clients operate better and faster,” says Syed Rasheed, Principal Architect at Concentrix.
Before launching its automation initiative, Concentrix relied heavily on manual and semi-automated processes to manage utility invoices for clients across the United States. These invoices arrived through multiple channels including shared network drives, email, SharePoint, and Microsoft Teams and originated from more than 100 different utility providers, each with its own invoice design, layout, and data structure.
In the early stages, invoice volumes were manageable. Concentrix processed only a few hundred invoices per month, allowing operational teams to download files, manually validate invoice content, and enter data into downstream systems. As client demand accelerated, volumes grew dramatically, eventually exceeding 100,000 invoices per month, with 6,000 to 8,000 invoices arriving daily.
“At that scale, handling invoices manually becomes a nightmare,” Rasheed says. “The requirement itself—extract the data and load it—is simple. The real challenge is handling the volume, the variation, and the speed the business demands.”
The diversity of invoice layouts further complicated the process. Each state, utility provider, and vendor used different formats for presenting charges, meter readings, and billing details. Concentrix encountered more than 300 distinct invoice variations, or “patterns,” across its client base. “Some vendors place charge tables on page one, others on page three. Some invoices look nothing like invoices at all,” says Rasheed.
To address these challenges, Concentrix initially adopted traditional intelligent document processing approaches based on trained models. The organization evaluated several automation platforms, including UiPath and Automation Anywhere, but Microsoft Power Platform stood out.
“We ran the same proofs of concept on multiple platforms,” Rasheed explains. “Power Platform was significantly more cost-effective—easily one-tenth the cost in some cases—and performance was better as well.”
The team began using built‑in AI tools to pull key details from invoices they already recognized. This worked well when volumes were low and formats were predictable, but as invoice volumes grew and layouts became more varied, the approach became hard to maintain and performance suffered. “To support 300 different invoice layouts using our original approach, we would have needed a large development team and months of work for each release,” Rasheed explains. “That just wasn’t realistic for our clients.”
Accuracy became another challenge. Many invoices arrived as scanned PDFs or used inconsistent layouts, and by early 2025, accuracy had dropped to around 65–70%. This meant more manual review, higher costs, and lower customer satisfaction.
To overcome these issues, Concentrix shifted to a more flexible, AI‑driven approach on Microsoft Power Platform, using generative AI models based on the GPT-5 multimodal large language model to interpret entire invoices in a single step. This eliminated many manual processes, improved performance and accuracy, and enabled a fully automated, scalable invoice solution.
With this new setup, Concentrix could send entire invoices—either as text or as full PDF files—directly into prompts to tell the AI what information to look for. The AI handled interpretation and organization of the data in one step, eliminating many of the manual checks and workarounds used before.
Beyond invoicing, the same approach is well suited to other document‑heavy industries, including financial services and healthcare, where high‑volume, rules‑based document processing is common.
The new automated solution is built around the following areas:
Sources
Customer invoices enter the solution via email, SharePoint, shared drives, or as PDFs in a Teams intake channel. Users securely access customer-managed shared drives through a gateway.
AI processing
Every 15 minutes, the system uses Power Automate to collect new invoices and process them in batches. It uses AI to read documents, recognize text, and extract key details, combining document reading, trained AI models, and large language models to accurately handle many invoice formats.
Data enrichment
The solution standardizes extracted data using cloud flow transformations and Dataverse reference tables. Business rules are applied through Power Automate, while Dataverse maps cities and ZIP codes and enriches vendor-specific fields before loading the final dataset into SQL Server.
User experience
The operations team uses a Power Apps mobile app to review exceptions. Users can view the original PDF alongside extracted data, enabling fast validation and keeping the automated pipeline flowing smoothly.
“We ran the same proofs of concept on multiple platforms. Power Platform was significantly more cost-effective—easily one-tenth the cost in some cases—and performance was better as well.”
Syed Rasheed, Principal Architect, Concentrix
With prompt-based AI, Concentrix can now scale more easily. “The flow becomes very simple with prompts,” Rasheed says. “One action now does what previously required 30 or 40 actions. That alone removed one of our biggest scaling constraints.”
Today, approximately 90% of invoices are processed using AI prompts rather than AI Hub custom models. A general prompt handles simpler layouts, while specialized prompts address more complex patterns.
The solution currently processes 100,000 invoices per month, with plans to double capacity to 200,000 invoices per month.
Since implementing the new solution, Concentrix has increased data extraction accuracy from 65–70% to over 96%, reaching 99% in January 2026. Higher accuracy has reduced downstream correction effort and improved customer satisfaction. The solution has also transformed agility for the Concentrix development team. Now, new invoice types no longer require months of retraining. “With GPT-based models, we didn’t need three months of training anymore,” Rasheed explains. “The model already understands documents. We just had to define the right prompts and extraction rules. When a new invoice pattern comes in, QA validates it and we move straight to production. There is no development cycle anymore.”
As a result, operational headcount requirements have dropped significantly. Teams managing invoice patterns and extraction workflows decreased from approximately 40 people to 11, with staff redeployed to higher-value initiatives. Automating invoices reduced manual processing roles by 50 to 60 people, with those team members placed on other projects within the company.
"The impact of this program has been transformative for our operations. By modernizing invoice processing with intelligent automation, we’ve unlocked new levels of efficiency, accuracy, and scalability across our utility partners,” says Kelly Desmond, Managing Director, EMS Business at Concentrix. “The significant cost savings, improved data quality, and measurable margin growth demonstrate the power of our collaboration with Microsoft and the strength of our AI‑driven approach. This success is setting a new benchmark for enterprise automation.”
“With GPT-based models, we didn’t need three months of training anymore. The model already understands documents. We just had to define the right prompts and extraction rules.”
Syed Rasheed, Principal Architect, Concentrix
With the intelligent document processing platform proven at scale, Concentrix is expanding both volume and capability. In 2026, the company plans to support multilingual invoices, beginning with French, Spanish, and Arabic, as it extends the solution beyond US-based utility providers.
The next major milestone is fully automating all 800,000 invoices processed monthly across clients. Concentrix plans to onboard an additional 50,000 invoices at a time to ensure stability and performance.
The company is also exploring deeper cloud integration, including routing all invoice intake through Azure Blob Storage and migrating on-premises SQL databases to Azure SQL. These changes would further streamline operations and simplify access. “AI is not the solution for everything,” Rasheed says. “You have to start with business need and ROI. But when you do, and when you adapt quickly, AI can transform how you operate at scale.”
Follow Microsoft