TechnologyCompliance with machine learning resource contract

Compliance with machine learning resource contract

In an effort to promote an efficient and compliant approach to responsible sourcing across the company, Ericsson’s leading AI experts and data scientists have recently begun implementing a new machine learning-based solution that has proven to be effective in:

  • Reducing the manual effort of audits by a large number of hours
  • Full coverage guarantee for all contracts
  • Provide insights and suggest proactive actions
  • Reduce the risk of non-compliant contracts

Below, we’ll take you through these benefits from a responsible sourcing perspective and delve deeper into our new machine learning-based solution.

Benefits of automating contract compliance

Let’s start with context: What are the benefits of solutions like machine learning-based signature verification? To answer that, it helps quantify the challenges posed by the organization’s traditional operations, such as monitoring compliance with supply contracts in this case.

Today, Ericsson provides software, services and related hardware from supplier partners around the world to cover the needs of all Ericsson units to meet business needs. The sourcing team and specifically the sourcing contract managers sign contracts with different suppliers over a diverse operational scope, diverse geography, and a wide range of legal terms and conditions. To ensure legal status, these contracts must be agreed upon bilaterally and signed by both the supplier and Ericsson’s power of attorney to make them comply with the law.

Ericsson’s supply team works with more than 40,000 suppliers culminating in 180,000 contracts worldwide in 170 countries.

While managing supply contracts of this sheer volume and variety, they can sometimes have unintended gaps which in turn pose a significant risk to Ericsson. If signatures are lost, the contracts become void and have no legal status. The impact of undetected infractions can result in breach of ethics, high financial penalties, and deterioration of Ericsson’s brand reputation globally. To deal with the risk of non-compliant contracts, the sourcing team is very proactive in conducting periodic audits. Since audits are manual and resource-intensive effort, automation can help reduce the hours required to perform these audits drastically (see Figure 1).

It is also essential to ensure comprehensive coverage and to give the sourcing contract manager information about potential non-compliance so that they can take early preventive action. This is another advantage that our machine learning-based solution offers, as it not only helps to significantly reduce reliance on manual efforts, but can also provide complete and standardized coverage and provide proactive insights into potential non-compliance.

This is clearly in line with Ericsson’s focus area Responsible Business and Digital Inclusionwhich includes a major strategic focus on Responsible sources.

The Responsible Sourcing Strategy states: “Managing the social, ethical, environmental and human rights impacts in our supply chain is part of our value chain approach to embed corporate responsibility throughout our business. Building capacity for our suppliers to meet high standards in all of these areas is an essential part of our approach.” Compliance with the supply contract is one of the building blocks that go along with making the strategic vision a reality.

Figure 1: Manual steps behind the Ericsson sourcing compliance process

Leverage machine learning models to ensure continuous improvements in compliance

Determining the most appropriate machine learning techniques requires breaking down the business problem into the following components.

  1. Disclosure: Does the document contain any signature page?
  2. Verification: Is the document signed?
  3. Definition: Is the document signed by Ericsson?

This is also shown in Figure 2 where the problem components are shown.

The different steps of a business challenge

Figure 2: The different steps of a business challenge

Figure 3 shows the overall automated solution. It includes multiple models for achieving outcomes, and we will now delve deeper into both the models and outcomes.

Overview of the source signature detection solution

Figure 3: Overview of the source signature detection solution

Step 1: Detection – Define signature pages

To cover all terms and conditions, contracts can consist of three to 150 pages. While analyzing the data, it was realized that the signature page usually contains some common words that can be used to identify it as a page containing a signature.

Common words on signature pages

Figure 4: Common words on signature pages

Through Natural Language Processing (NLP), a text classification model based on machine learning techniques was used to identify the page as a signature page. Figure 4 shows common keywords that appear on signature pages.

Step 2: Verify – reveal signatures

The next step was to find the number of signatures and locate each signature on the page. To do this, we used YOLO, a family of pre-tested object detection models, which give us the exact location of the signature(s).

Input image sample

Figure 5: Input image form

Figure 5 shows the three signatures on the contract page. The contract contains three signatures, one manual and two digital. The result of the YOLO model is shown in the lower right part of Figure 5. It indicates the coordinates and signature types (0 – manual signatures and 1 – digital signatures).

Step 3: Identification – Verify the authenticity of offline signatures

A person’s signature shows a high level of consistency and does not change much from time to time. For this reason, we have used a machine learning model that detects irregularities and at the same time can be able to catch very similar forged signatures in case of skilled forgery. A Siamese neural network was used to train it to approximate the similarity function which produces a score between 0 (similar) and 1 (different), see Figure 6.

Example of pairs of similar (label = 0) and different (label = 1) signatures used to train the model

Figure 6: Example of pairs of similar (label = 0) and different (label = 1) signatures used to train the model.

Notes based on training machine learning models

We’ve shown that we can use machine learning to identify valid signatures in contracts and have made a scalable solution that can be used anywhere we have sample power of attorney signatures. As with machine learning models for training, the more high-quality data we have, the more accurate the solution. So it is not surprising that it was easier to find signature pages in contracts written in English and in scanned contracts in high quality.

The most difficult of the three tasks was to verify signatures, and the highest success rate was in locating signatures. Validation of the most common signatures was successful, but also showed satisfactory results for less common signatures. The complete model as shown in Figure 3 showed, with all three steps, an accuracy of 85 percent. This is a result that would not have been achieved a few years ago, but thanks to the development of neural networks and deep learning and their applications, we can achieve these results today.

bottom line

We believe that applying machine learning models, along with a robust deployment strategy, can provide a much-needed foundation to enable a larger set of automation-based use cases while improving and standardizing the process.

As we do this for sourcing, there is a high potential for harnessing the benefits of the solution across other enterprise units as well – resulting in minimal human intervention with stronger oversight across many business-critical processes.

towards the future

In light of the increasing volume of contract data consumed by the solution, we are further optimizing the models to provide a reliable and comprehensive contract compliance solution. Given the sensitivity of contractual data and subsequent decision making, the next steps are exploration Interpretable AI Solutions to meet an understandable, transparent, interpretable and trustworthy system.

Want to know more?

Other areas of application of Ericsson machine learning include predictive network planning, defect detection, ticket classification and management, bill of materials generation, node failure forecasting, transportation management freight forecasting, inventory optimization, supply planning, and many more. Learn more about these areas and the other opportunities that lie ahead for Ericsson Artificial intelligence page.

.

Source

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Exclusive content

Latest article

More article