Expediting Last Mile Process Automation

Innovative data-driven AI and ML technologies from Magic FinServ

Originally published on 06 June 2022

Front, middle and back-office processes, for asset managers, banks, and related service providers, such as fund administrators, custodians, prime brokers, can all be expedited by technology partly or fully replacing more laborious and repetitive human input, particularly in support functions such as financial control, planning, analysis, and treasury, and even for some parts of customer relationships and engagement, where smart servicing using AI comes into play.

Technology not only scrapes, identifies, and extracts relevant data, but also enhances its value by interactively cleaning, structuring, converting, enriching, amalgamating, formatting, and validating from multiple sources, including APIs, excel and unstructured offline data such as documents, emails, tableaus, social media, and web downloads. The objective is initially “last mile” process automation and ultimately to strive towards the holy grail of complete and holistic end to end solutions. 

The objective is initially ‘last mile’ process automation and ultimately to strive towards the holy grail of complete and holistic end to end solutions.

Versatile technology capabilities 

Technologies used by Magic FinServ include AI, ML, RPA, API, cloud, and a mix of open-source libraries, proprietary and customized algorithms, which continuously learn and improve. Applications include DeepSightTM, which can automate downloads, and extract critical data elements (CDEs) from multiple sources, clean them, transform them, and feed them into downstream applications. “It can extract data, using a mix of Template Based Algorithms based on spatial information and Heuristics Based Algorithms driven by rules specific to the business and underlying data. It can classify documents and organize them into subcategories. DeepSightTM can handle images as well as text content: for instance, it can parse diverse formats into a preprocessing pipeline to correct and enhance images for better OCR (optical character recognition) accuracy. Or we might download a concatenated part of a security description, and then enrich it with static data to get it moving through a workflow and ingested into a target platform,” says Parag Samarth, President at Magic FinServ. The system can corroborate and validate data even if it does not always fully verify its accuracy or authenticity.

It becomes possible to analyze more data, more often – and with greater accuracy: the systems can identify both known exceptions and unknown patterns that require further investigation, which can feed back into new codified error alerts, iteratively improving the system.

Improving data quality and quantity paves the way for personalized, real time and scalable solutions, and automation can produce cost savings.

Ruchi Aggarwal, Marketing Manager, Magic FinServ

Customization can allow for personalized solutions, labelling and presentation, and labelling, which may use visualization tools such as bar charts.

“Improving data quality and quantity paves the way for personalized, real time and scalable solutions, and automation can produce cost savings,” says Ruchi Aggarwal, Marketing Manager at Magic FinServ.

We highlight some case studies of use cases:

Front office
Regulatory filings such as 10k reports can be interrogated for positives and negatives. ESG is increasingly in focus as investors and regulators alike are vigilant about “greenwashing”. “We assist with ESG related research by extracting items from the sustainability section or directors’ report,” says Samarth.

Fundamental company research is a relatively free form use case, where the data can be configured to an event-driven strategy. In the front office, a $14 billion New York-headquartered discretionary event-driven asset manager was using a very labor-intensive research process involving Excel models, annotating PDF documents, and sharing research via data rooms. The manual tasks entailed were not fully auditable and traceable. Magic DeepSight™ streamlined the process, drawing data from sources such as S&P Capital IQ for sell side research, forecasts and corporate transcripts, and SEC EDGAR for public forms and filings. Magic DeepSight™ is trained to identify and retrieve the relevant data points providing a full data lineage. “The system has also analyzed risk disclosures in SEC filings, covering macro risks, loss-making investments in the past and M&A. It has identified human errors and even errors in SEC filings, which helped to win new mandates,” says Samarth.

All of this increases the volume of data that can be gathered and analysed, and improves the frequency of analysis, by surmounting human time constraints. Some quantifiable benefits for the asset manager above were clear within 120 days: front office analysts spent 40% less time on these tasks while processing more than nine times as many data points. Other indirect benefits included a reduction in subjective human judgement errors.

Compliance – KYC
Compliance routines, such as KYC, AML, and broader monitoring of communications for suspicious or anomalous text, across multiple electronic and hard copy channels – such as email, instant messaging, Slack messages, printed documents, and faxes – are areas that can be partly automated through AI, which allows a larger volume of data to be processed much faster, at lower cost and avoiding some forms of human errors: “A compliance team has saved over 78% of the time spent on manual monitoring and at the same time covering over 4 times as much data as was monitored by manual scanning,” says Aggarwal. 

KYC automation gathered data from unstructured documents, data and fields, parsing formats, and classifying document types. “KYC routines can to some degree be automated through AI and ML, which can extract and process data from multiple unstructured documents and data sources, and then place the data into appropriate fields in third party and internal applications,” says Samarth. 

Compliance – AML and sanctions
Banks, administrators, and funds are regularly found to be remiss in their AML routines, even if actual cases of money laundering are much rarer. One culprit of weak AML procedures is outdated systems and manual processes, which can be improved using technology. “DeepSightTM can read KYC documents and extract specific data based on rules, use Cognitive RPA to update case tools or customer applications, check for some sanctions data in documents and web downloads that can be “read” by DeepSightTM. It also reads transactions, creates a client profile, and looks for patterns that could meet the money laundering rules. Output is scanned to reduce false positives and operational effort,” says Samarth.

Beneficial ownership validation was another recent project, which can be helpful for applying some sanctions. 

Alternative and direct lending
For lending strategies, including digital loans, technology makes it possible to trawl through 200- 300-page contracts, and automate manual review and classification of documents and emails to speed up loan processing. “AI and Machine Learning (ML) are powerful tools for reviewing data and flagging up compliance failures, errors, and potential errors, based on rules and reverse lookup mappings. The AI, ML and automation can also be coordinated with OCR and NLP. All of this can greatly accelerate loan processing and scalability,” says Satadeep Mitra, COO at Magic FinServ.

Back office – expense management 
For expenses management, it is possible to automate manual processes for vendor invoices. Cloud-based expense reporting can automate manual processes, using AI and ML to manage real time data. A global US investment firm found invoice management was manual, complex, time-consuming, and error-prone even after adopting platforms such as Coupa and Concur. “Magic used AI, ML and RPA to drive straight through processing, including parsing diverse formats, classifying invoices, and extracting information from invoices of multiple formats. RPA was also used to enrich the data and enable workflow for approvals and authorizations. The solution learns through actions and feedback loops and applies complex mappings to help with automatically populating the data. It enhances accuracy, efficiency, and compliance with matters such as correct withholding tax allocations,” says Samarth. 

“In summary, Magic FinServ’s data driven AI ML innovative tech solutions, can meet customer demands, provide 24/7 surveillance, deploy advanced cyber forensics, and save money,” says Samarth.