Driving Efficiencies in Processing and Analytics of Investment Portfolio Documents with AI

Automating Document Classification with AI for More Accurate Predictions, Faster Decision-Making

Dynamo Software Inc. has been revolutionizing the alternative investments industry and optimizing processes for over 1,000 clients, including fund managers, institutional investors, and service providers, since 1998. With a diverse portfolio of configurable, cloud-based, automated and versatile software, Dynamo helps its clients to identify and solve key challenges in the alternative investments ecosystem, thus improving the productivity of fundraising, deal, research, investor relationships, and portfolio management teams worldwide.

Data collection and processing is one of many business-critical pipelines that ensure the success of Dynamo’s clients. The faster the documents are collected, classified, and transferred to appropriate investments, the higher the return on investment that Dynamo’s clients can expect over time.

The Dynamo teams themselves can benefit from a rapid, more accurate and efficient processing of documents. Benefits include:

• Minimized amount of repetitive manual work
• Reduced operational expenses
• Increased performance and productivity
• Rapid decision-making on client investment portfolios

With this understanding, the leaders of Dynamo were looking to enhance and modernize their document classification platform through AI/ML and automation.

The existing platform received thousands of various types of documents every month. Some of them were sent directly to the platform, while others were emailed, to be manually added by managers. Once added, the documents would be properly stored and classified, either manually or by a machine learning tool.

The leaders of Dynamo wanted to significantly improve the accuracy of their existing ML tool, and to automate a manual portion of the data processing pipeline. Their expectation was that a new AI-powered document classification solution would be at least 85% accurate on new data. On top of that, they wanted to gain the ability to make predictions, based on a document’s contents.

The document collection and processing pipeline is only a part of the Dynamo platform that enables them to track contacts and transactions, view account balances, and make decisions on client portfolios with advanced analytics. In light of the scope of their activities, Dynamo was also considering other opportunities for adopting AI & ML.

Provectus, an AWS Premier Consulting Partner with competencies in Machine Learning and Data & Analytics, was selected to join the Dynamo project through the AWS Fast Start Program. Our proven track record of developing intelligent document processing (IDP) solutions in the cloud positioned us favorably, and helped us to move forward into the next stage of cooperation with Dynamo.

Developing a Document Classification Model and a Training Pipeline for Dynamo

Provectus approached the opportunity, bearing in mind that Dynamo was ready to deploy the enhanced document classification platform only after the model had exceeded the required accuracy threshold of 85% on new data. Thus, the team’s primary focus was on developing a highly accurate and precise model, and a robust training pipeline, to justify further cooperation.

Overall, our work consisted of a series of steps that encompassed:

• Setting up an infrastructure for the development and management environments
• Building and implementing an experimentation infrastructure for document classification
• Conducting an exploratory data analysis (EDA) of various datasets provided by Dynamo
• Developing a testing dataset, and datasets for data extraction
• Training a baseline model based on unstructured data
• Developing a training pipeline for the document classification model
  Training and tuning the document classification model for the extraction of data from PDF documents (four labels)

The document classification model designed and built by Provectus exceeded the expectations of Dynamo, returning an f1-score of 95% on a test dataset.

The training pipeline was built using AWS best practices for data security and privacy, to ensure data confidentiality. The suite of Amazon SageMaker services was used to develop the model building pipeline. In the meantime, the inference pipeline, built on AWS Step Functions, was built and improved by the Provectus IDP team.

Note: The solution developed by Provectus for Dynamo is an essential and customizable part of the Intelligent Document Processing (IDP) platform. This enabled us to easily integrate the solution into our platform, and to provide Dynamo with state-of-the-art accuracy and quality, and the best cost per document on the market from day one.

Realizing the Benefits of Automated AI-powered Document Classification and Next Steps

The first milestone of the Provectus team was to develop a document classification model that could confidently and correctly classify at least 85% of the documents, to incentivize the leaders of Dynamo to proceed to the next steps. We were able to build and train the model, and to develop its training pipeline, in less than six weeks.

When run on a test dataset, the model returned an f1-score of 95%, which exceeded the requested threshold of 85%. The initial implementation handled PDF documents using four labels: capital calls, distributions, capital account statements, and tax documents.

The model’s success demonstrated the potential of AI and automation to Dynamo. Not only could they automatically classify over 90% of various documents with AI; they would also be able to make predictions based on extracted document data.

The leaders of Dynamo saw the opportunity to continually enhance their document classification platform with Provectus, to reduce the amount of repetitive manual work performed by their data team, lower operational costs, increase performance and productivity, and minimize the time needed for decision-making on client investment portfolios.

Dynamo is now ready to move forward into the next stage of cooperation with Provectus and AWS.

Moving Forward

1. Learn more about the Provectus Intelligent Document Processing (IDP) solution
2. Watch the webinar on Choosing the right document processing solution
3. Apply for Intelligent Document Processing Solution Discovery Program to get started