chevron-right Created with Sketch. back to thoughts

How we helped DAS automate the decision-making process of disputing invoices

The decision-making process regarding whether or not to dispute supplier invoices in claim management is long and requires a certain expertise. Sagacify therefore deployed Machine Learning to automate this process. It was the second project Sagacify initiated in close collaboration with DAS, the Belgian market leader in legal protection insurance, in order to increase the efficiency of their business processes.

Problem to be solved

AI can be a tool to reduce the workload of cumbersome and repetitive tasks without replacing the person responsible for this, but to give them the opportunity to focus on more difficult tasks requiring their personal knowledge and experience.

The problem to be handled is rooted in the decision-making process of disputing invoices. This process works as follows:

  • Firstly, the responsible manager at DAS encodes the incoming invoices of suppliers into the system. Here, suppliers can be doctors, lawyers, experts, etc. contributing in solving a particular case.
  • Afterwards, this invoice is analyzed on the existence of possible anomalies and if so, the manager decides on disputing the invoice or not. For instance, an anomaly can refer to an absurdly large amount regarding a particular mission.
  • Finally, the dispute implies the necessity of writing a contestation letter to the parties linked to those anomalies.

At first sight, this system might seem solid, nonetheless, it can take up to one hour per invoice. Therefore it is perceived as an exhausting and inefficient procedure. Above that, this procedure is subjective which might cause people to make misjudgements or increase the risk of wrongfully rejecting certain disputes. Consequently, this raises the possibility of missing out on certain cost-saving opportunities.

Sagacify has the answer...

Sagacify handled according to a standard machine learning project approach, starting with precisely understanding the business question and processes of the client. We understood what legal experts were looking at during their analysis of invoices, which helped us deriving distinct features, such as the payment history, parties involved... After gaining a clear understanding about the procedure and the problem to be solved, Sagacify worked on a Proof-Of-Concept (POC) where we showed and convinced the client of the advantages our AI services have to offer while retaining an objective view on the situation. This POC is an early ML model that proved its capacity to determine whether to dispute an invoice or not. The results achieved showed a significant positive impact on the efficiency of the procedure:

  • We reduced the workload with 90% enabling the employees responsible for this task to focus on their core competencies. This means that only 10% of the invoices that our model processes, need to be verified by the manager itself.
  • On top of that, more than 25% of the invoices that still require the expertise of the responsible employees should be disputed or are at least labeled as abnormal. Previously, this number was much lower (only 2%), so remarkable progress was achieved.
  • DAS noted an increase in the motivation of their employees to dispute incoming invoices.
  • It allowed them to select a subset of the team trained for this renewed procedure leading to a more specialised team and an improved chance of success in each dispute.

Moreover, this POC allowed the client to build a business case, with a clear ROI and integration strategy. After finishing this POC, Sagacify deployed it in the cloud on Amazon AWS and the client integrated it through an Application Programming Interface (API) in their production system.

The model mainly allowed DAS to focus more on other tasks, because only a subset of the invoices required their attention. Also, the tendency to sometimes neglect this procedure was imminent with the result that less invoices were disputed and therefore losses were inevitable. The system that was introduced by Sagacify discouraged that tendency.

… but the answer is not always clear

The need to consult the client and collaborate closely with them is crucial for such a project to succeed. Thanks to the previous projects with DAS and our customer-focused approach, we were able to overcome the difficulties encountered during the project.

The biggest obstacle was the creation of features that make up the dispute of the invoice. This was highly dependent on the specific business process of DAS. A fruitful cooperation with the client helped to gain the insights necessary to finish the model.

A fruitful cooperation

Here at Sagacify, we help our customer transition towards a future-proofed AI-ready organisation. To achieve this, the quality of our relationship with the client is one of the most important factors in the success of our projects. Sagacify ought to promise less and deliver more rather than to promise more and fail to deliver on those promises.

Our purpose is to help the client to be successful because we believe their success is our success. The only way to fulfill this purpose is to establish a clear communication with the client. Therefore, we organise so-called sprints or quick sessions with the client where we report the ongoing progress of the project, ask for - if necessary - additional information and feedback from the client. We also appointed a success manager that is responsible for maximizing the value for the customer.

When the model is implemented, we continue to create value for the client by continuously applying maintenance and improvements to the model.

What comes next?

In the pursuit of continuous growth, we seek to initiate new projects with new and current clients. With DAS we believe other opportunities are possible, such as extending our current model to automate their whole invoice process. This can be interesting, not only for DAS, but also for other companies seeking to improve the efficiency of similar processes.

Does your company can benefit from this technology as well?

G e t i n t o u c h search Created with Sketch.