ABOUT US
At APDS Analytics, we are passionate about helping businesses succeed through the power of analytics and data science. Our team of experts has years of experience working with retailers to develop innovative solutions that drive growth and profitability. Whether you need help with credit risk assessment, customer segmentation, or predictive modeling, we have the expertise and tools to help you achieve your goals.
Book a
30-Minute Review
Email Us
OUR SERVICES
APDS Analytics offer a range of services targeted towards the retail credit industry in South East Asia, Africa and the Middle East. Services include:-
-
Originations
-
Portfolio Management
-
Revenue Modelling
-
Independent Model Validation
-
Monitoring & Portfolio Review
-
Other Models
-
Training
OUR CLIENTS
Our clients range from big banks offering a full range of credit products, country level credit bureau to small independent start-ups and payment gateways. Clients have been sourced through partners, through contracting agencies and organically. Over the years clients have included: -
-
Xendit Payment Gateways
-
I-Score (The Egyptian Credit Bureau)
-
CIC (Vietnamese Credit Bureau)
-
The West Brom
-
NatWest (thru an agency)
-
Nordea
-
WestPac NZ (via Mastercard Advisors)
-
Riyad Bank
-
Arab Bank / Ajman Bank
-
etc
OUR PARTNERS
APDS Analytics operates with and through a range of partners from across the globe
-
CrePASS Solutions (Korea)
-
Analytix Engine (South Africa)
-
Paragon Business Solutions (UK)
-
Aptivaa (Middle East / Africa)
-
OGMA Risk & Analytics (USA)
-
Raffles Risk (Vietnam)
-
SAS (global)
-
Independent Consultants
Book a
30-Minute Review
Email Us
.jpg)
Origination Scorecards
Origination Models help banks and finance houses make the right decision when a potential customer applies for a loan.
Credit Scoring as a Service
CSAAS takes the lender for zero data (maybe a product launch) to a full bespoke models engineered from their own data with full transparency. The service draws on APDS 30+ years model development experience to build a tailored expert model, construct an appropriate monitoring framework and subsequent model enhancements using AI and human expertise. An experienced consultant is always on hand to offer sage advice and guidance.
New Application Modelling
Over the years the data used for application scoring has been extremely static, app form data and bureau data only. However with the introduction of Open Banking the sphere of data that can be used has drastically widen improving the quality of the prediction as more behaviourally based characteristics can be used. Additionally, the reliance on bureau data may be lessened as the Open Banking data can reveal commitment to regular payments behaviour, therefore potentially opening new lending markets.
Alternative Data & Financial Inclusion
A global phenomena is the over-reliance of credit reference agency data. However in most developing markets, for a variety of reasons, access to credit and therefore a bureau record is severely restricted and banks cannot or will not lend to these invisible clients. This in turn pampers country wide economic growth and participation. The solution is teh adoption of alternative data (sourced from mobile phones, utilities and other sources). APDS partners with CrePASS to collect and score the new data using AI/ML techniques.
Portfolio Management Models
Portfolio models aim to either expand the portfolio through additional lending or reduce or limit losses when customers fall on hard times.
Behavioural Scoring
Behavioural Scoring takes derived and trended information from account operation to determine the level of risk associated with existing customers. The score can be used to determine whether or not the bank should lend additional funds to the customer through limit increases, cross-sell or upsell activities. As the data is behavioural in nature and derived across range of dynamic time periods the data is extremely powerful. As banks embrace Open Banking data the power of the data will increase as it moves to be sectorial as opposed to being from one portfolio or institution.
Early Warning Models
From time to time even good customers (most likely SMEs or Micro Businesses) will struggle. However, the first thing the bank knows about this is when payments start to be missed, i.e. when it is too late. An Early Warning System is a special type of delinquency model that identifies which customers are vulnerable prior to the account going delinquent, so that the bank can initiate measures to help good customer over bumps in the road and differentiate service levels by the strength of the relationship. A particular application of these models may be when climate data is added to help predict which farmer or builder etc. in in trouble after extreme weather events.
Collections Scoring
Once an account has become delinquent the challenge for the bank is to predict which customers will roll backwards towards being up to date and which will roll to ever worsening states of delinquency. Once it is predicted management can determine which customers receive which actions and which collectors are assigned to which cases. By targeting the action and process the bank will be able to improve collections efficiency, lower NPL rates, reduce the Expected Loss numbers and improve customer satisfaction. Typically roll rate models utilise collections specific data like 'promises to pay kept or broken', number of times 30 dpd last 6 months etc. A second branch of models looks at how much can be recovered post default (90dpd), like a pseudo LGD model without the discounting.
Portfolio Review
All implemented models will decay over time, as the portfolio naturally changes with differing customer behaviour, amended customer targeting and sand-shifting economic & climatic conditions
Model Monitoring & Policy Rule Review
As we know all models will dynamically decay, however as model & portfolio managers we need to know how the model prediction changes over time and whether the changes are in the TTD population or back end model performance.
From a model performance we look to follow a five test approach that looks at score distribution & model prediction stability and drills down into characteristic level stability and alignment of odds.
At a TTD Level policy rules can affect the model prediction, so periodically these are also reviewed.
Independent Model Challenge
Whereas, model monitoring is essential, few organisation will challenge the implemented model with new data, new assumption or different modelling techniques. For independent challenge we will build a range of quick & dirty models and compare against the performance of the developed / implemented model.
Model Fine-Tune / Recalibration
If model monitoring or independent challenge throws up concerns then the model may need a full rebuild or a re-calibration of the attribute scores.
The full re-build will incorporate the elements of the independent challenge that indicated where improvements could be reached.
If a fine-tune is recommended, then the existing model is used as a base and the characteristic scores are reweighed to reflect the new operational environment.
Independent Model Validation
As part of capital, provisioning and model risk regulations independent model validation is required.
Qualitative Validation
The most basic form of model validation that involves a comprehensive review of the model documentation, critical assessment of the data used in modelling, a critique of the methodology utilised, results assessment and a set of recommendations for potential improvements, suggestions of alternative approaches etc..
Quantitative Validation
The quantitative validation approach takes all the elements of the qualitative approach and assesses model performance utilising a more recent data sample. The validation aims to determine whether the model(s) produces comparable results to the development period, the models rank order and the components tell the same story as at time of development.
Challenger & Replacement Models
Building on the the quantitative approach challenger modelling builds a new version of the model, perhaps utilising approaches to address the limitations mentioned in the qualitative report, maybe utilising alternative datasources and modelling approaches. The independent view of the challenger model will help to frame the modelling environment more comprehensively and help the development and internal validation teams to develop new perspectives.
If the model requires re-development APDS can initiate a new project to do but utilise the experience and knowledge garnered on the validation exercise to expedite the timeline.
Revenue Modelling
Provide a general description of the items below and introduce the services you offer. Click on the text box to edit the content.
Response Models
In many markets around the globe regulation requires lenders to obtain customer consent to change the terms and conditions associated with their accounts. This can include changes to limits and access to new data, such as open banking data. Response Models help to identify the customers that are likely to respond positive to such actions and can therefore ensure that the marketing spend is optimised for biggest 'bang for buck'.
Spend Models for Behavioural Change (transactor to revolver)
Most Credit Card Portfolio suffer from serious Revenue Leakage (between 15 & 40%), where most of the portfolio is either dormant or transacting with low utilisation, resulting in low revenue and higher costs. APDS will help you to identify customer groups like Transactirs, Revolvers, LT Dormants and ned build models in order to identfy which are likely to convert to interest earning revolvers (controlled by the risk behavioural models).
Spend Model can predict which customers will respond to promotion spend offer, whether it is on domestic or overseas spend, again with the object of changing behaviour into higher revenue earning segments.
All spend offers need to be governed by risk models.
Activation, Dormancy and Attrition Models
Dormancy and Attrition are major problems for credit card portfolios as it means a reduction in revenue and a potential increase in costs. For new account the dormancy problem is about encouraging the new customer to activate and use their new card so that the bank starts to collect behavioural data and generate revenue. Models based upon the application data and/or Open Banking data can determine which customers will activate and spend early in their life on book.
Patterns in transaction may show evidence of wind down to dormancy or worse towards attrition and account closure, if a model can identify likely dormant or leavers then the bank can proactively try to convert back into a user or revolver.
Transaction Scoring
We Recently developed a suite of models for a Payments Gateway in Asia and realised the potential of the data as an SME Score. A Payments Gateway could develop scores from the data to : -
-
Sell Loan Target List to Partner Banks, increasing volume for the bank - CUSTOMER NUMBERS UP
-
Risk Assessment Scores, an origination score for a bank wanting to lend to the PG's Merchants - HIGHER ACCEPT RATE
-
A Behavioural Transaction score for on-going risk, where banks wish to renew loans, upsell additional projects - REVENUE UP
-
An Early Warning Score to detect which businesses with loans are starting to struggle so that pro-active strategies to save the business can be put in place - LESS BUSINESS' IN COLLECTIONS, NPLs DOWN
-
Collection scores based on affordability calculated from business operations - MORE COLLECTED, NPLs DOWN
SELLING SCORES MONETARISES THE PAYMENT GATEWAY DATA, INCREASES REVENUE & DEEPENS RELATIONSHIPS WITH MERCHANTS AND PARTNER BANKS
Training
We provide training related to Credit Scoring Concepts, Credit Portfolio Monitoring and Model Validation. Additionally we are happy to an assisted development package and analytical mentoring sessions (described below).
Concepts of Scoring
Building models is a very structured process, on this two-day session the delegates with learn the concept behind modelling and the decision that can be made off the back of scores. The session include theory and change of war stories generated over the course of a long global career. Additionally, the attendees will build a scorecard from scratch using the Modeller package from Paragon Business Solutions.
Assisted Development
Assisted development is where APDS offers the Concepts of Scoring training but then leads a bank project whereby the analysts develop the model with close support from APDS consultants, covering key areas such as design, performance definition sampling, characteristic analysis and monitoring. The Assisted Development programme is system agnostic.
Mentoring
Those analysts new to analytical credit risk management APDS runs a series of paid mentoring programme that can be tailored to individual needs and immediate career goals whilst allowing individual to choose the number of sessions they would like. APDS Consultants will help to set goals, set tasks to ensure the individual moves towards achievement of those goals and helps to track progress. The sessions will be 1 hour conversations at a frequency to be determined.
Our Team
Matthew Freeman has thirty plus years retail financial services industry experience, predominantly within the analytical risk management sphere, developing all manner of risk and marketing models, covering diverse geographies ranging from the United Kingdom and Western Europe, throughout the Middle East and India, China, Asia (North and South East) , Australasia (Australia and New Zealand)) and within emerging markets (such as Mongolia and Cambodia).
From a regulatory perspective Matthew has led teams (and built) all components of the AIRB and IFRS9 arenas in Europe, the Middle East and Asia (South East Asia, China and South Korea).
From an innovation perspective Matthew is leading the way in the use of alternative data for credit scoring / financial inclusion.
www.linkedin.com/in/matthewwfreeman