Designing for the Mortgage Industry
Client- Argo Groups
Argo Group International Holdings, Ltd, or Argo Group is a Bermuda-based international underwriter of speciality insurance and reinsurance products in the property & casualty insurance, reinsurance and managed risk solutions. Government entities like Freddie Mac and Fannie Mae provide insurance against financial loss to mortgage lenders due to prepayment.
Objective- Design an Analytical dashboard, a mobile app which represent the mortgage loan level data and future credit risk data generated from the AI models.
Background Research
ArielRe, a member of Argo Group, is an international reinsurance underwriter for multiple industries.
Mortgage is a key line of business for ArielRe but it comes with a high degree of risk and a number of unknown variables.
They wanted to simulate and model a large, varied and unstructured data set (200 GB) to
(a) better determine mortgage delinquency in the USA and
(b) significantly improve data processing and analysis time.
© To design an analytical Dashboard
Jump360, working closely with ArielRe, carefully cleaned up and mapped huge volumes of both internal historic data and external data. We automated this process to ensure the data set was constantly current. The next step was to overlay 30 economic variables that influence the outcome of mortgage delinquencies. We subsequently developed numerous AI models which were calibrated based on actual historic outcome to allow for model tuning and bias.
A Lesson From History
We have started this project by analyzing an important event in US history. In 1997 the world of technology and Wall Street were reaching a fever pitch. While algorithmic trading had been a major driver of the crash of 1987, the use of quantitative risk management and burgeoning fields of algorithmic pattern analysis were once again viewed as catalysts which would forever change the market.
Two separate investment funds embodied this marriage of quantitative and financial engineering. Both companies had over $2B AUM in 1997. One company had two Nobel prize winning economists driving the creation of their main fund, while the other had a team of quirky mathematical modelers from Cal-Berkeley. Both companies had extremely bright futures and were already implementing the first instances of machine learning in their quantitative models by the early 1990’s. By 2005, one company would average a 71.8% return in its main fund, while the other would be bankrupt .
Company A — Long Term Capital Management
LTCM would go on to not only become insolvent, but almost crash the world’s economy due to faulty belief in the perfectibility in their models, and lack of respect for Black Swan events.
Company B — Renaissance Technologies
Renaissance Technologies would have only one year of losses , between the years 1989–2005. Since inception in March of 1998, its flagship $3.3B Medallion fund has returned an average of 35.6% compared to the S&P 500’s 17.9%. Today,Renaissance Technologies has one of the most advanced Machine Learning driven funds in the marketplace and continues to use technology to play a dominant force in the Hedge Fund sector.
We have highlighted these two companies because they are the foundation of the project,and vision we have undertaken. What was originally a goal to answer the future of risk origination, transformation, and securitization, has become a far larger vision on how Argo Group can become a dominant fixture in the future intersection of User Experience, technology and reinsurance. But this vision was built on a foundation of hubris, and understanding that those who do not study the past are destined to repeat it. Over the past eight months we have not built a model which will predict the future, a tool to make decisions for humans, or somehow replaced the infinitely complex expertise of a risk expert. What we have built , is one of the most powerful pattern recognition and insight generation engines in the Reinsurance Market. While our technology will not replace the need for the world’s best talent, it will empower the leaders of Argo as well as our clients, to fundamentally change the way risk is understood, analyzed, and transferred to third parties.
Below are the findings from 2018 Argo MID-YEAR REPORT-Future of Insurance, which foster the idea that AI and automation will make Argo businesses more productive.
My Role in this project-
To Collaborate with Vice President, Data scientist and development team to understand and visualise the mortgage loan-level data and future credit risk data generated from the predictive AI model, and to show them on the dashboard.
Hats which I wore during this project
Hat of User experience designer, I have to craft the overall experience of dashboard.
As a Visual Designer, I have created mood boards and visual language for the project.
As a Usability Analyst, I have tested the system.
What Job this Product does
It takes massive sets of data and patterns from the past to deliver insights about predicting loan default and prepayment probabilities.
Design Process I followed
The fundamental design process has three main parts — ‘Empathise’, ‘Ideate’ ,‘Deliver’ and ‘Test’ keeping the user at the heart of the process.
Brainstorming and Ideation
During the ideation phase, I actively collaborated with Data scientist Mayur Dangar to understand the data and how does the predictive model work. Brainstorming, user personas, interview are the building blocks to create task flow and Information Architecture.
Current
Interview Highlights
“There are three separate things we want to accomplish:
1) To upload deals
2) Run the model
3) Represent data
we need to be able to view how these parameters interact with the default rates of a portfolio. As well is how they interact with the Economic variables of the time that it was written.So I would want to be able to see over time, what factors cause our losses (either bad portfolio or bad economy)”
Initial Flow Diagram
After another round of discussion with the users, we brainstorm, combined and organised the data in a more coherent and absorbable way.
- Task flow diagram-
Task analysis is the tools which we use during the “define” stage of the Design Thinking process. We have defined all the task which the user needs to follow
- Login and settings
- Upload and view deals
- Upload new simulations
- View Deal Statistics and reinsurance results
2. Data Visualization -
The main challenge for this project, is to visualise the loan-level data and future credit risk data generated from the predictive models. Below are the steps which I have followed
Step1- Understanding Data
Step2- Divided the essential information into pockets that are easy to consume, with clear statistical and verbal descriptions.
Step3- Visualization
User needs to select a deal from the List to view the performance of a particular deal. After selecting a deal, the task is to represent the organised data. The top left-hand corner is usually where the viewer first directs their attention, so the overall summary of the dashboard, i.e. Total Unpaid Balance, Total no of loans, Avg. The loan balance is represented at the top left corner.
Total Unpaid Balance
Geo-Spatial map technique is used to represent the Total Unpaid Balance for all the states of USA. Blue colour hues are used so that all efforts should be made to enable the user to keep their attention on the map, without the need for constant checks to determine the colour-value associations in the legend.
FICO, DTI & LTV
3D bar charts are used to represent FICO, DTI and LTV because they used to promote an immediate understanding of categorical data; the contrasting heights of the bars are instantly perceptible and as such can be compared quickly between highest and lowest FICO,DTI or LTV.
Reinsurance share, List of deals & Reinsurance structure
There are times when we want to provide the user with a list of data, as opposed to graphical representation but it is still essential to arrange the data in a way that affords instant identification of a specific value and comparison between two or more of the values in such case we use tables.
Designing Better Data Tables
Tables are used to display the Reinsurance share, List of deals & Reinsurance structure.
Below is a list of design structures, interaction patterns, and techniques I used to design better data tables
- The row style method for scanning data quickly.Reducing visual noise by using stripes.Pagination works by presenting a set number of rows in a view, with the ability to navigate to another set.
3) Fixing the row header as a user scrolls provides context on what column the user is on.
4) A visual data summary provides an overview of the accompanying table. It allows the user to spot patterns and issues in aggregate before actioning summary insights.
5) Much like expandable rows, quick view enables a user to view additional information while staying in context.
6) Expandable rows allow the user to evaluate additional information without losing their context.
7) Modals allow the user to stay within the table view but provides more focus on the additional information and actions.
2. Visual Design
Argo Mobile Application
Designing the dashboard information on mobile is not a simple task. The Mobile application is designed to have a quick look at the performance of the dashboard. The main hurdle was to represent table data in a digestible and usable form.
Design Approach
How do users consume this information?
The first step was to understand how consultants used and interacted with this data in its native web-based form. The different ways that users consumed the data determined the design of its mobile counterpart. Insight from the client highlighted two different approaches:
Use Case #1: Comparison of a reinsurance results (i.e. vertical scanning, column-specific)
Use Case #2: Comparison of reinsurance results within a single location (i.e. horizontal scanning, row-specific)
Satisfying Use Cases
Wireframing
Invision Prototype-
