Client Case Studies
Following Case Studies are brief details of our Product / project Implementations for some of our clients:
1. Counterparty Credit Risk Software Implementation
Counter-party Credit Risk Solution for a large Oil & Energy Company
- Client is a large SE Asia based company trading in Oil and Oil products (including Gas Oil, Fuel Oil and Petrochemicals) and is a part of Forbes Global 2000 companies list. It has a huge sourcing requirement from thousands of Counterparties through multi-year contracts, which exposed the Client to Market & Credit Risks.
- Counterparty defaults caused operational disruptions and reputation loss, besides losses in terms of replacement cost.
- Client was using an Excel based model to compute Potential Future Exposure (PFE) for managing Counterparty Defaults.
- Running the model in excel for each deal was practically impossible and was also error prone, thus limiting the use of the model
- Correlation across counterparty defaults due to correlation between various commodities) was not being taken into consideration in the model
- Client wanted a custom GBM (Geometric Brownian Motion) based system that could provide correlated estimates of their commodity prices
- There was no integration with the existing Trading system in the company.
- Oil & Energy companies work with a lot of Basis contracts and complex OTC contracts structured around them posed a challenge to model development
- Unorganized data, including Price data with different frequencies (daily, weekly, monthly) required changes to the standard approaches
- Monte-Carlo Simulations when run on thousands of contracts, computed billions of intermediate values, thus posing a computational challenge
- Risk Edge delivered a Robust and scalable Counterparty Credit Risk Solution for the entire Portfolio and integrated it with their trading and back-office systems.
- The Client now sees each Counterparty’s Potential Future Exposure for multiple horizons on a daily basis. The solution also points when the Exposure is max / min and whether there is any concentration of risk among the counterparties
- Module integrates with key systems in the organization to give an accurate and holistic picture of the Credit Risk
- Performance optimization was done using multi-threading and parallel processing to calculate PFE very efficiently and quickly.
2. Risk Consulting – Developing a New Hedging Model
Risk Consulting Project for a Large Oil Refining company
- Client is an India based company with business primarily in Crude Oil Refining and Marketing of products.
- With Crude Oil prices showing high volatility, Client’s profitability was becoming very uncertain and client was losing money on both the buy and sell side.
- Client realized that Existing Hedging Models were not working for them and the whole hedging strategy was being questioned at the board level
- Client approached Risk Edge to evaluate a possible development of a new Hedge Model based on its unique business requirements.
- The objective was to deliver a new model that was flexible and cost effective compared to their existing hedging models.
- Unorganized data was one of the issues faced in the project since client’s entire position data was stored in many excels, some of which weren’t reconciling.
- Risk Edge developed a model after studying client’s business process, but back-testing past several years data on that model was a challenge. However, Risk Edge software was utilized in this project for generating results on the past data and the entire process was automated with it.
- Risk Edge’s consultants studied the Client’s business processes for a few weeks along with their existing hedging models.
- A new proprietary model was built by Risk Edge’s Statistical and Business Analysis teams that we thought would suit client’s requirements better.
- A temporary, quick customization of Risk Edge software was done to allow our team to back-test the model on historical data correctly and quickly.
- Huge sets of past data were created and run through Risk Edge Software, back-testing results were collected and passed on team for further analysis.
- Risk Edge delivered the new model along with the back-test results, which showed significant savings in Hedging Costs for the Client.
- Client was able to save over USD 30 mn in just hedging costs in a year using the new model. Also, the hedging model allowed client to incorporate qualitative perspectives into hedging decisions.
3. Machine Learning – Yield Prediction for Palm Crops
Yield Prediction for a Fortune 500 Agri Client
- Client is a SE Asia based company who is one of the world’s largest producers of Oil Palm.
- To manage their business operations and sales planning, the management wanted a fairly accurate picture of expected yield, which was fluctuating quite a bit recently due to weather patterns.
- Client invited us to help them predict Yield on Palm crops.
- Client wanted to use AI to predict yield on their with good accuracy, so that they could use those numbers to plan their business operations and sales targets better.
- Their internal data analytics team had been trying to predict yield for their palm crops but were unable to get a good accuracy.
- Unorganized data, including several different data formats (maps, csv, xml, excels, etc.) were difficult to reconcile into one large dataframe.
- Several of the parameters viewed as critical for predicting Yield had missing values in between.
- Quite a few critical parameters like fertilizer application, etc. were missing from the data. A few others were based on random sampling rather than accurate measurement.
- Risk Edge brought in very well-researched algorithms and created an ensemble of those algorithms to deliver results. The accuracy was around 82%, which was deemed as acceptable by the management team.
- We used Hierarchical models, along with several other Machine Learning models to predict Yield. Different algorithms were used to interpolate missing values and extract seasonality and trends from critical parameters.
4. Risk Consulting – Automating Market Risk Management and Mark-to-Market P&L Calculations
Market Risks & P&L Calculations – For a large Agri-business
- Client is a leading Global Agri-business operating in several countries.
- They need to maintain and update their Mark-to-Market (M2M) reports for all transactions in each commodity segment periodically and accurately.
- Client has a myriad of systems and processes to calculate their M2M P&L for each commodity. These systems and processes have been developed over time and hence have become too complex to operate and change.
- Client had bought some commodity trading businesses from its competitors and along with it came new systems and processes, which added further to the complexity of integrating existing and new systems and arrive at a common way of calculating and analyzing M2M P&L and market risks.
- Client needed Risk Edge to understand the existing systems (including spreadsheets) and processes for multiple commodities and provide an execution strategy to change existing systems and integrate new systems such that Managing market risks and M2M Calculations are completely automated.
- Coordination with Large, globally dispersed teams owning different applications. Some applications were new even to the teams, while some others were in transition.
- Unique nature of transactions in different commodities for which automation of risk and P&L calculation had to be done.
- The time provided for the project was very short, as the project had been going on internally for many months already, with little success.
- Risk Edge, given its team’s prior experience with large Agri companies, was able to understand the specificity of trading across different commodities very quickly, which helped shorten the time to execution significantly.
- For one of the commodities, Risk Edge understood and tweaked existing, in-house systems to ensure that the automation was done quickly. The teams were trained on rules and formats for inputs, and also on how to extract the daily / weekly reports.
- For another commodity, Risk Edge brought in a specific software that could quickly deal with integrations across multiple systems and geographies, and helped reach automation goals well in time.
- Client has since been able to take daily, weekly and monthly reports on Market Risks and M2M P&L with very little manual efforts. The entire process is now transparent and rule-based, giving accurate numbers which will save hundreds of person-hours for the teams each year.
5. Machine Learning Solution – Predicting Yield and P&L Simulation
Machine Learning Software for Predicting Yield and P&L Simulation – For a large Agri-business
- Client is a leading Singapore based Agri-business operating in over 70 countries, and supplies food and industrial raw materials to over 22,000 customers worldwide.
- The client has presence in many commodity segments, including edible oils, spices, nuts, confectionery and beverage ingredients, etc.
- They need to prepare and present budgets to their top management for which they need to simulate their P&L to see how various input parameters can affect their P&L. They also needed to predict Yield of one of their commodities to ensure that budgeted numbers are as close to reality as possible.
- For a particular commodity, client wanted to improve their budgeting process. The existing process of simulating various input parameters in spreadsheets was very cumbersome and fraught with inaccuracies.
- Client also wanted deeper understanding of relations between various input and output parameters, for which 3D charts built using latest technologies was required.
- Client was using spreadsheet to forecast yield for upcoming months, but the results weren’t great. They wanted to use advanced machine learning algorithms that could predict yields with greater accuracy, and hence could make the entire budgeting process more accurate.
- The time provided for the project was very short, the client needed the solution for an upcoming budgeting process.
- Understanding how inputs like ratios, prices, and currencies affected outputs like PBT, EBITDA, etc.
- Risk Edge was able to, in record 1 month time, deliver a web-based solution that enabled client to not just simulate their P&L, but also predict yields on their crop based on their actual historical data.
- The solution was web-based, and was designed for all devices – laptops, mobile, tablets, etc. The solution used latest technologies to enable 3D charting and graphs, which were appreciated by the client.
- All the calculations for this project, running into thousands, were done in-memory, without storing them in database. As a result, users were able to change inputs parameters and at the same time, see the impact of those changes on their P&L immediately.
- Advanced machine learning algorithms were developed and implemented to ensure a more scientific way of predicting yield.
6. Machine Learning – Using NLP to power a Robo-Advisor
NLP based algorithms to power a Robo-advisor
- Client is a leading Global Forbes 100 Investment Bank with more than USD 700 bn in Assets under Management.
- The Client has major presence in the US and handles investments for several high net worth clients.
- They needed an intelligent, automated system that can work within a permissioned framework and suggest Goals of its clients based on their social media activity to its Advisors.
- In the fast paced world of social media, Advisors usually find it tough to keep up with the changes in the lives of their clients. They need to pick up the context from each client in every meeting, and then suggest appropriate solutions, thereby wasting their and the client’s precious time.
- Client needed an intelligent system that could connect with multiple sources of data and then make sense out of all the data in its entirety and present it in a summarized format.
- Client also needed complete compliance with the GDPR requirements and the legal terms and conditions dictated by each data source.
- While connecting to multiple data sources and extracting data was not an issue, connecting information coming from various sources – social, financial, health, etc. was a challenge as there was no such system in place.
- Global GDPR regulations started tightening in the middle of this project which created additional tasks for the team to study and comply with those requirements.
- Handling the data volume from multiple data sources that their client base was connected to was another Big-Data challenge that we faced.
- Risk Edge was part of the vendor consortium which was selected to execute this project. Risk Edge’s responsibilities were to design and develop the Machine Learning Algorithms, and manage the offshore project management activities.
- Risk Edge was able to build and deliver a robust and scalable solution which could ingest large volumes of data and process those elements through machine learning algorithms and finally gather and store all the results in a No-SQL database.
- With the solution, client was able to extract value in terms of saving time for its advisors and clients, and also get better context of their clients overall financial position and thereby offer better products that met their needs.
7. Machine Learning – Predicting Machine Failures
Predicting Machine Failures
- Client is a large Renewable Energy Company with more than 4 GW of production capacity
- The client was operating and maintaining hundreds of Windmills and solar panels which would break-down / become inefficient occasionally thereby impacting the revenues.
- The Client had access to telemetry data which was collated at a per second interval using multiple sensors on each machine.
- This telemetry data was, however, rendered unusable due to its sheer size. It was stored in proprietary database using advanced compression algorithms.
- Client needed to not just analyze the data that was being stored for various reporting purposes, but also wanted to see if that data could be used for predicting machine failures so that the down-time could be minimized.
- The Supervisory control and data acquisition (SCADA) based data was only part of client data. Their breakdown related data was maintained in their ERP. Combining these two diverse data sources was a major challenge.
- While we deployed Big Data techniques to handle large volumes of high velocity data, we also had to ensure that we first clean the data so that the average size of datasets could be reduced. For this, dissecting the data, cleaning it of all unwanted information and then putting it back together was another major challenge we had to overcome.
- Handling the data volume was another Big-Data challenge that we faced.
- Risk Edge was entrusted with the task of providing meaningful insights into the data, which Risk Edge was able to do by analyzing Terabytes of data, processing it and rendering a few MBs of data for reporting purposes, which Client’s internal systems were able to handle.
- Risk Edge provided high accuracy of prediction of Machine Failures by using Recurrent Neural Networks algorithm.
- Client was able to not only reduce the down-time due to breakdowns by 60%, but also able to increase their management reporting scope to include several newer parameters which were not available to them earlier.
8. Machine Learning – Journal Entry Anomaly Detection
Predicting Anomalies in Journal Entries
- Client is a Global Metals & Mining Company
- The client’s Internal Audit Team was spending an inordinate amount of time going through millions of Journal Entries to find anomalies. They were using a simple Rules based framework which was giving them a lot of False positives.
- The Client’s Internal Audit team had buit some rules to filter anomalous Journal Entries but the result contained thousands of false positives.
- Due to its sheer size, it was not possible to sift through every possible Journal Entry looking for anomalies, so random sampling was done. However, it still left the client exposed to possible fraud / error in Accounting.
- Client needed to use a better framework than rules to not only detect and score anomalous transactions, but also reduce false positives significantly.
- The entire dataset was unlabeled (i.e. no Journal Entry was marked as anomalous / normal), thus making training of any AI algorithm difficult.
- The dataset was quite large, running into tens of GBs, and applying Deep Learning algorithms needed us to deploy special hardware that could handle that kind of volume.
- The high level of confidentiality of data meant that we had to adhere to extremely strict IT guidelines and sometimes even work with data inside client’s cloud environment, which was deployed on Microsoft Azure.
- Risk Edge reduced the data size by understanding the domain and using feature engineering. We also reduced class imbalance (the proportion of good v/s bad entries) by removing definitely-normal entries from the dataset.
- Auto Encoder algorithm, one of the Deep Learning alogs, was used to detect anomalies in Journal Entries.
- A Statistical filter was also built along with the AI algorithm to sift through the transactions, and a final list of transactions was made using rules, AI, and statistical filters which significantly reduced the False positives in Anomalous transactions.