Problem

• Customer struggled with detecting anomalous financial transactions – the current process is labour intense and error prone due to mass volume of the transactions and amount of false positive alerts

Solution

• Unsupervised machine learning system to automatically flag anomalous transactions together with prioritisation based on severity and an explanation why a given transaction can be an anomaly

Impact

• Lower need for manual interaction
• More accurate transactions monitoring
• Quicker time to act on an alert

Challenge

• Difficulties maintaining code and monitoring data health
• Making assumptions leading to incorrect outputs
• Need to reimplement existing pieces of software
• Big overhead to integrate new data sources for each use-case

Solution

• Trusted, opinionated and reliable code assets that define guidelines and enforce compliance
• Code modules that allow downstream users to consume and interpret data assets

Impact

• Reduced time to onboard new data assets
• Reduced human errors thanks to guideline enforcement
• Increased transparency about availability and content of data assets
• Ensured common understanding of the data model, allowing to scale data asset development efforts

Identifying clients that are most likely to have financial problems through an automatic model based on warning indicators to alert credit officers earlier

Challenge

• Credit officers need to assess and manage financial risk for their respective portfolios
• The amount of data collected each day by the risk department and its velocity make it hard to properly monitor and manage risks for every client

Solution

• Help the credit officers to identify and focus on clients that are more likely to have financial problems in the future
• Leveraged Foundry as a data platform and defined early warning indicators – rule base alerts triggered by specific events
• Unit8 implemented those alerts on parameters such as late leasing payments or overdrafts

Business Impact

• Significant simplification of the assessment and monitoring that credit officers need to conduct
• Clear email based alerts that can be easily and quickly reviewed

Identifying clients that are most likely to have financial problems through an automatic model based on warning indicators to alert credit officers earlier

Challenge

• Regulations require the client to setup surveillance for money laundering and illegal activities
• Disparate data sources and formats, and varying regulations between countries
• Inability to monitor client behaviour to spot illegal activities

Solution

• Leveraging a previous project Unit8 conducted (Single Client View), built a generic framework enabling data scientists to easily define surveillance scenarios to monitor potential high-risk client activities
• Pipeline specifically designed to scale to millions of customers tracking billions of equity positions globally

Business Impact

• Greatly reduced amount of time needed to define surveillance scenarios, from weeks to minutes
• Reduced resources required and build-time of pipelines by more than 50%
• Improved reliability of scenarios
• Decreased manpower necessary to conduct surveillance audit by a factor of 20

Creating a single client view for an international Swiss bank allowing them to track and compute total risk exposure, and meet AML compliance regulations

Challenge

• Siloed data across multiple systems
• Inability to calculate risk exposure
• Lack of unified view of their clients
• Limited ability to meet compliance regulations
• Sub-optimal conditions to apply proper AML

Solution

• Aggregated, cross-referenced and combined data from hundreds of disparate data sources
• Used ML to automatically surface high-risk clients
• Implemented a scalable solution to account for future needs

Business Impact

• Achieved single client view, increasing visibility
• Able to calculate overall risk exposure
• Compliance with AML regulations
• Already stopped previously undetected terrorism-financing activities

Creating a new data pipeline for collateral risk calculations to improve the quality of data information and reporting for end users

Challenge

• Risk calculations on collaterals provided daily to end-users through dashboard
• Need to leverage this risk calculation data model for reporting purposes
• Difficult to identify which data would be required for reporting, and to accurately provide it through previously-built pipeline that wasn’t sustainable

Solution

• Initial discussion to fully understand data sources and existing data flows
• Created new pipeline to extend the already existing data model by including specific information
• Identifying- and providing solutions to issues in the current codebase which introduce data quality issues
• Provide guidelines to prevent similar issues from reoccurring

Business Impact

• Improved data quality of the reports by onboarding the data on the main data processing platform through the newly built pipeline
• Enabled client team to improve and learn by giving tips and providing guidelines
• Reduced disk spillage per TB of data from 50GB to less than 100MB

Refactoring an entire codebase making the pipeline run time 7x faster, allowing quicker new feature development, and increasing data quality for analysis

Challenge

• The “Portfolio Monitoring and Reporting” project provides multiple dashboard for risk managers to assess their risk
• Over several years of development, the codebase has become entangled and slow which was blocking the delivery of new features

Solution

• Analysed, simplified and refactored the entire codebase to better future-proof the project
• Solution leveraged was the Unified Foundry Ontology (UFO) that we had previously developed
• Pure foundry solution combined with Unit8’s programming expertise and best practices

Business Impact

• Increased code quality and resolved bugs that accounted for millions of CHF in exposure reporting deviations
• Reduction of pipeline execution time from 10h to 90 min, resulting in fresher data available for analysis
• Allowed smoother onboarding of 200+ users
• Established scrum and coding best practices

Aggregating all marketing data into a single view matching leads to customers to gain a better understanding of marketing campaigns performance and costs

Challenge

• CRM data spread across entities with little to no aggregation at a group, vertical or customer level
• Lack of data integration meant no overall visibility on factors influencing consumer behaviour
• Inability to check marketing investment effectiveness in customer acquisition and retention

Solution

• Created a data pipeline in AWS to aggregate al marketing campaigns, leads, and customer data into a single source of truth (Data Mart) on AWS
• Pipeline cleans data and builds a unified view matching leads with customers

Business Impact

• KPI tracking per advertising platform, campaign and ad spot (e.g. cost per lead, transformation rate, ROI, etc.)
• Transition to a data driven marketing plan rather than based on “gut-feeling” and beliefs

Assessing current custom-made data platform to make recommendations on the best approach and create valuable insights for future platform strategy

Challenge

• Client wanted to assess current capabilities and maturity of their custom-made data platform
• Need to compare current setup to potential future alternatives
• Required an assessment to decide on future platform strategy and to investigate recurring issues raised by users

Solution

• Delivered platform assessment across 4 dimensions:

1.  data platform tooling and process capabilities
2.  data analytics and science platform strategy
3.  data driven use-cases
4.  platform development

• Assessed strengths & weaknesses in depth and developed recommendations
  for Data Platform team

Business Impact

• Increased awareness on strengths and weaknesses of custom-made platform in different application scenarios
• Recommendations for a hybrid approach: custom-made ingestion and off-the-shelf platform for Analytics capabilities
• Decision basis for future platform strategy

Building up a Neo4j Graph solution to help different divisions map relationships and interactions between people, companies, accounts and assets

Challenge

• Talend ETL and SQL queries resulted in high manual overhead and could not properly maintain relationships between different parties and accounts
• Looking for a more business friendly interface for their bankers

Solution

• Built up a Neo4j environment
• Created a necessary data schema and ingested required datasets
• Leveraged the newly created Neo4j environment to create two required views:
  Party
  Account

Business Impact

• Self service interface with granular access rules to allow different departments (wealth management, risk, fraud, etc.) to perform their own queries
• Ensured regulatory compliance regarding particular sanctions

Developing an ML model to identify potential rogue traders and flag them to minimise risk of financial losses and lower need for regulatory cash reserves

Challenge

• Inability to monitor and identify traders whose behaviour differs from what is expected, resulting in potential monetary and reputational risks for the client

Solution

• Integrated trading data from several sources into a single platform
  Deployed ML techniques to identify potential rogue traders based on their trading behaviour
• Configured an alerting system to flag traders with an elevated risk of rogue trading

Impact

• Able to quickly identify risky traders based on their trading activities, which would have been impossible to do manually
• Minimised rogue trading risk and potential financial losses, lowering the need for regulatory cash reserves

Designing and integrating Tableau to enable secure access to datasets stored on the main Foundry instance

Challenge

• Need to enable a wide pool of users to have access to vast datasets stored in a Foundry instance

Solution

• Designed and integrated a Tableau server with Postgres and Postgate
• Foundry datasets can now be accessed through tableau while preserving security standards present in Foundry

Impact

• More than 300 Tableau developers and users now able to securely access and consume the datasets directly from Tableau

Optimising platform performance with a set of tools and best-practices combined with coaching, resulting in lower costs and smoother user experience

Challenge

• Platform adoption skyrocketed and rate of increase is higher than what new hardware added is capable of handling
• Usage expected to keep climbing as more projects are being onboarded, leading to further cost increase and performance degradation

Solution

• Established performance metrics baseline
• Coached client’s engineers to enable them to improve platform performance
• Created tooling and reports to pinpoint performance problems and allow self-service improvements

Impact

• Reduction in new hardware addition and maintenance costs
• Increase in platform performance resulting in smoother user experience

Creating a dashboard unifying all trading information resulting in a more holistic understanding for credit risk officers and substantial risk reduction

Challenge

• Credit risk officers struggle to understand precisely who they are trading with, what kind of goods are traded, and what are the exposure concentrations

Solution

• Created a dashboard presenting an interactive view of the CTF portfolio and its clients to mitigate risks and recognize business opportunities
• Entity resolution (based on free text sources) to provide a unified view of all external parties involved in the transactions

Impact

• More holistic understanding of CTF business across departments
• Substantial risk reduction through better understanding of exposures

Defining and implementing a surveillance framework to identify illegal client behaviour resulting in greatly improved ability to meet surveillance requirements

Challenge

• Regulations require the client to setup surveillance for money laundering and illegal activities
• Disparate data sources, and diversity in regulations between countries
• Inability to monitor client behaviour to spot illegal activities

Solution

• Leveraging a previous project Unit8 conducted (Single Client View), built a generic framework enabling data scientists to easily define surveillance scenarios to monitor potential high-risk client activities

Impact

• Greatly reduced amount of time needed to define surveillance scenarios, from weeks to minutes
• Reduced resources required and build-time of pipelines by more than 50%
• Improved reliability of scenarios

Problem

• Car manufacturer holds large global leasing portfolio that has to be periodically evaluated
• Current valuation predictions are inaccurate

Solution

• Based on a large volume of past transactions (12 years) and cars parameters (model, mileage, options), etc. predict accurately the residual value of the car

Impact

• Accuracy of the valuation improved by $100M’s (from large overall base value)

Problem

• Significant churn in existing customer base which limits overall growth ambitions in saturated Swiss market
• Large number of customer data available, which was not systematically analysed
• Client advisors often surprised if customers leave

Solution

• Early warning system was created based on a machine learning model to identify customers at risk of leaving or withdrawing a large part of their assets (attrition risk)
• System based on regular monitoring of client behaviour (e.g. transaction behaviour, the intensity of engagement) to predict the attrition risk
• A customer group specific retention approach was developed to pro-actively contact customers with high attrition risk

Impact

• It could be proven that the machine learning model identifies the right customers at risk
• Retention approach has proven to be highly effective since the churn rate of customers at risk could be significantly reduced
• Client experienced strong business benefits, since revenue outflows due to customer churn could be reduced

Developing a centralised solvency and risk assessment model to create risk exposure reports automatically and improve data quality

Challenge

• Difficulty to predict total exposure to financial losses resulting from companies at risk of becoming insolvent
• Current risk reports are generated manually and data quality is not guaranteed

Solution

• Central gathering and processing of the relevant partner data allowing for central risk modelling and data quality checks
• Automatic creation of monthly risk reports based on the curated data

Impact

• Improved risk management capabilities through information centralisation and standardisation
• Improvement in data quality and report generation
• Possibility to create reports dynamically to visualise more aspects than before

Problem

• Impared visibility of user activity and behaviour allows for malicious behaviour to be undetected for a long period of time
• Compliance group in a global private bank decides to monitor user activity on the central data platform

Solution

• Introduction of dynamic altering and anomaly detection system including enhanced signal information via integration of new sources of data (HR data, badge swipes, VPN logs, … )

Impact

• Greatly improved security, audit and monitoring capabilities of the platform team
• Decreased risk of malicious behaviour remaining undetected

Creating a single client view for an international Swiss bank allowing them to track and compute total risk exposure, and meet compliance regulations

Challenge

• Siloed data across multiple systems
• Inability to calculate risk exposure
• Limited ability to meet compliance regulations
• Sub-optimal conditions to apply proper AML

Solution

• Aggregated, cross-referenced and combined data from hundreds of disparate data sources
• Used ML to automatically surface high-risk clients
• Implemented a scalable solution to account for future needs

Impact

• Achieved single client view, increasing visibility
• Able to calculate overall risk exposure
• Compliance regulations can be met