This content has been archived, and while it was correct at time of publication, it may no longer be accurate or reflect the current situation at Microsoft.
Microsoft Digital has created a new approach to verify payments and reimbursements. By combining artificial intelligence and best practices, payment validation can be completed in near real-time with greater accuracy, freeing up human agents to put more time and energy into complex claims.
If you’ve ever walked through a reimbursement request, you know how cumbersome validation can be. Proof of Execution (POE) validation reconciles payment requests with work done, promotional offers, and incentives programs, similar to how insurance claims are resolved. It’s a vital, but cumbersome, component of partner payment compliance at Microsoft. Proper validation compares evidentiary documents, images, and other artifacts against reimbursement requirements before issuing payment. Validating reimbursement requests is a lengthy and manual process, requiring requests be validated against a series of compliance requirements to ensure each request was completed, valid, and adhered to the stated guidelines. Manual validation places a heavy burden on those involved, including account managers, validators, and partners.
POE validation is used throughout Microsoft’s business operations, including expense reporting and partner incentive claims. The stakes are high—POE validation is the gatekeeper for catching improper payments and fraud. Mistakes cost money and potentially create vulnerabilities to bad actors.
Our POE validation process is getting an overhaul thanks to one of Microsoft’s Digital teams. To help expedite POE validation, improve accuracy, reduce costs, and boost compliance, the Microsoft Digital team has identified best practices and a new automated approach. In doing so, they’ve improved reimbursement payout within the partner incentives space, creating a solution that’s extensible across Microsoft. By applying Azure AI offerings for POE validation, Microsoft Digital dramatically transforms the speed and accuracy of reimbursements, improves scalability, and sets a new standard for what the industry should expect from future validation models.
Revamping POE for an automated world
Microsoft Business Operations (MBO) specializes in making it easy to do business with Microsoft, engaged Microsoft’s validators and account managers to map out the traditional model for validation. Examining validation steps with stakeholders allowed the team to understand all aspects of POE, including strengths, challenges, and inefficiencies. Working with users and stakeholders allowed MBO to align objectives—a critical step in creating a solution and building trust in a new system.
The traditional POE validation model involves multiple hand-offs of numerous images of invoices, receipts, forms, catalogs, brochures, flyers, statements of work, metrics data, and artifacts that need to be cross-referenced against a set of business rules. Each piece of evidence needs to be scanned for accuracy, including dates, product names, and dollar amount. For example:
- Microsoft program designers establish POE requirements—the documentation and evidence the partner must supply for reimbursement. This is different across programs, both in the types of materials to be supplied, what needs to be verified, and how the POE is delivered.
- When it’s time to seek reimbursement, the partner delivers the pertinent documents. In some instances, the POE needs to be extracted from emails and spreadsheets prior to validation, creating additional steps and complications.
- A validator compares the supplied materials to POE requirements, checking for errors, omissions, inconsistencies, or outliers. If anything is incorrect, the reimbursement request can’t be fulfilled. Similarly, if mistakes aren’t caught, it’s possible that an incorrect payment is issued.
- After review, and assuming there are no issues with the provided documents supporting the reimbursement request, the validator passes the documents to an Incentive Capability Manager (ICM), who ensures that partners are paid accurately and on time.
It wasn’t hard for MBO to identify opportunities across the process.
Three key recommendations emerged from this interaction:
- Simplify POE requirements. The types of acceptable documentation were overly complicated, creating circumstances where partners might submit incorrect, duplicated, or incomplete POE. Where one program might require an advertisement clipping, another might need a receipt. Furthermore, complex requirements take additional time to accurately review. Simplifying the types of documents needed to validate reimbursement requests makes it easier to process requests with higher accuracy and speed.
- Standardize POE templates. Each program designed POE requirements on their own, which created a decentralized and siloed process across Microsoft. The variety of non-uniform reimbursement documents meant validators and partners could easily confuse program requirements or make mistakes. Standardizing requirements takes some of the complexity out of the process, creating a “one size fits all” centralized template for POE validation.
- Automate POE validation. A large volume of complex requests is manageable until it grows. When validators are pushed to capacity, you create the risk of not processing requests on time, accurately, or compliantly. Introducing automation allows a bulk of the work to be handled by technology, freeing up validators to focus on difficult edge cases. Doing so increases both speed and accuracy.
These three design concepts formed the basis for a robust change to MBO’s POE process. Team members saw change-management fundamentals as a way to address opportunities to simplify and standardize. Not only would simplification and standardization make the process less complex, it would also support automation needs. At the same time, MBO quickly capitalized on existing Microsoft AI services to bring automation to this highly manual and error-prone process. Automation reduced manual inefficiencies and allowed human validators to prioritize high-risk outliers over routine reimbursement requests.
The tech solution
In designing POE automation, the MBO team used Microsoft Azure’s Cognitive Services to build a microservices architecture, or a series of modular applications that can be plugged in to perform tasks, resulting in a scalable and extensible environment. The architecture offers domain services that specifically cater to business operations and help orchestrate the microservices. These microservices are agnostic of MBO’s POE domain, making them good candidates for re-use.
The domain services, dubbed Retail Marketing POE, validates the POE of retail offers and promotions.
If the POE validation clears all the checks defined within the business rules, it’s considered successful. Any failure during the POE validation is marked and reported in the architecture. The resulting rejection reasons are cataloged and detailed through Microsoft PowerBI, making it easy for validators to examine the reimbursement request, proactively detect issues, and take necessary actions. Because each of these rules are implemented as an independent microservice, agnostic of MBO’s specific requirements, a different domain service, such as expense reports and payment reconciliation, could use the validation process.
The solution’s architecture takes advantage of shared microservices, such as logo detection, entity extraction, product detection, and duplicate detection. The framework is based on Microsoft Azure App Services, Microsoft Cognitive Services, Microsoft Azure Cosmos DB, Microsoft Azure Event Grid, and Microsoft Power BI. The framework can support a near-real-time pass or fail.
Read API, Optical Character Recognition (OCR), Microsoft Azure Content Moderator, Microsoft Azure Text Analytics, object detection, and Microsoft Azure Cognitive Search are the primary services used to capture relevant information for business rule processing. They were utilized in the following ways:
Product detection. All the images are staged in Azure blob containers and ingested as a source to Microsoft Azure Cognitive Search, which creates an index of the images and their content. This index is refreshed periodically and on-demand when new images are added to the blob. The OCR skillset is used to recognize the printed and handwritten text from the images. With this setup, the solution can quickly recognize the search input from the image and provide results. The microservice has the flexibility to match content in a “Full String,” “Any Word,” or “All Word” searches based on need, to enable an effective search. The Exclusion List capability of the service helps remove noise from the search. It also maintains synonyms for search keywords, enabling better accuracy.
Entity extraction. Capable of supporting 60 languages, this microservice helps identify “Date” entities within POE. However, this functionality can potentially be extended to include other entities, including “Organization,” “Zip Code,” “Phone Number,” and more. OCR and Read API are used to extract lines from the image, and Microsoft Azure Text Analytics is used to extract entities from these lines (in this case, the “Date” entity). Additionally, a Regex-based handler is implemented to extract any entities Microsoft Azure Text Analytics has failed. Microsoft Azure’s Text Translator service translates any input language to English, making it easy for validation.
Logo detection. To detect logos, this microservice uses object detection and OCR. The benefit of object detection is that you can use it to train and identify logos with an accuracy percentage. With the datasets that MBO has been working on, a 70 percent threshold seemed appropriate for detecting a logo with a high degree of accuracy. The threshold is a configuration that you can set to detect entity clarity. OCR is used to detect logos with free-flowing text, not necessarily official logos, and to serve as a fallback option if object detection fails. Utilizing both APIs has improved accuracy. Though not currently a function of the Retail Marketing POE domain service, you can use this microservice independently for various scenarios, including validating the clarity of a logo before submitting a reimbursement request.
But it’s not just extraction. The architecture can also perform duplicate detection, which helps identify anomalies caused by either manual error or fraudulent activity.
Fraud detection. By using Microsoft Azure Content Moderator’s Match API, the solution can report a percentage of match accuracy across all input images. Based on the type of input sets being tested, you can set an appropriate threshold to flag for duplicate submissions. For Retail Marketing POE, anything above a 98 percent match threshold is flagged as a potential duplicate.
MBO built the Retail Marketing POE validator and duplicate detection on Microsoft Azure App Services exposing GraphQL endpoints, a flexible way to interact with APIs and its schemas, making it easy for client integration having multiple scenarios run under a single gateway. Because validation data can be quickly parsed, users like account and capabilities managers can easily review a reimbursement request.
Shared microservices, like logo detection, have REST API endpoints, simplifying and standardizing how information is communicated to users. In addition to program details, managers can now see the number of requests within a program that have been processed, the number of rejections, and the reason for those rejections. The ability to quickly call and examine this information in real time strengthens audits because reviews can identify which programs or regions are at high risk of being noncompliant.
Testing the hypothesis
With an automation solution in hand, MBO selected Microsoft’s Retail Incentives Program as a pilot for their new approach to validation. The Retail Incentives Program demonstrated a major need to move away from manual validation due to the volume and variety of reimbursement requests being processed. Human validators couldn’t scale up efforts to match a large influx of requests. Furthermore, the wide spectrum of programs and requirements increased the likelihood of human error when processing the POE for validation.
The benefits of the MBO solution to the Retail Incentives Program pilot include:
- Increasing scalability. With automation, any number of submissions could be validated without adding more time or resources to the process.
- Reducing human touch. Moving human validators onto edge cases—special circumstances that require additional attention—is a better use of their expertise. Validators only focus on requests rejected by the AI services instead of processing routine POE validation.
- Improving validation quality. Automation catches more errors and issues than a human validator. It’s also better at identifying fraudulent anomalies, like altered POE requests or duplicated documents.
- Reducing validation time. Real-time validation creates a positive experience for partners. AI reduces validation time to minutes. As the cycle time decreases, payments go out faster.
- Improving compliance. With improved accuracy and reduced validation time, compliance audits—the ability to review altered or duplicate requests—occur in real time, not retroactively.
In the end, all objectives for the pilot program were met. The microservices solution could easily scale up and handle an increased number of reimbursement requests, all without an increase in resources or a reduction in accuracy. Routine validation occurs almost instantly, speeding up the payment cycle and creating a better experience for Microsoft’s partners. Focusing human validators on only rejected and edge cases improved anomaly detection rates. Altered and duplicate requests are now caught with greater accuracy.
The Retail Incentives Program scenario has shown that automated POE validation can have a positive impact on cost, quality, compliance, and timeliness.
What’s next for POE?
It’s still early in the transformational journey for the MBO team, but they now see opportunities to expand this solution and principles throughout the company, with the automation offering eventually being an enterprise product. Due to the extensible nature of the microservices architecture, concepts introduced in this initial effort have the potential to improve reimbursement request validation in every industry and vertical.
Automated validation was a big first step. The transformation of POE is still underway, but it’s only going to get better as new practices develop. The next stage includes a major change management push to fulfill the simplification and standardization opportunities initially identified by MBO. This includes working with program designers to build “minimum only” POE requirements to simplify the validation process. Defining a minimum list of POE artifacts that still adhere to your compliance requirements helps both account managers and partners. More so, it has the added benefit of creating greater compatibility with AI validation technology. The MBO team is also leading standardization efforts, creating templates, and defining the types of documents that work best for POE validation and easier AI recognition.
The team’s success can be partially attributed to early engagement with stakeholders, which afforded them buy-in from users as MBO developed and tested their solution. By aligning interests, listening to challenges, and involving program designers, account managers, and validators in the process, MBO learned what would create the biggest impact on user experience.
Reimbursement validation isn’t just a Microsoft problem. You can use the microservices architecture and refined process anywhere automated document validation is required, for any company. New opportunities, like invoice validation and a web crawler bot, are being considered for future iterations of the automated validator. The extensible nature of the architecture allows new microservices and functions to be easily incorporated, making the validation solution stronger with subsequent microservice development. The Retail Incentives Program pilot was a first step towards an eventual enterprise product; an extensible, scalable POE validation service will be the keystone for a new method of reimbursement.
