Category: Fraud Innovations


How We Help Auto Lenders Comply with Dealer Monitoring Regulations

I read an interesting article from Christine Pratt at AITE which was very interesting. You can read the article here – Are Indirect Auto Lenders the Real Target of the FDIC’s New Third-Party Guidance?

Something which we have known for years and now happening, is more regulation for Indirect Auto Lenders and the dealers that they work with.  The government has required mortgage lenders monitor their third party lenders and brokers for years.  We should know.  Many of the staff in PointPredictive were responsible for launching the broker and third party lenders monitoring tools in the mortgage industry.

 The FDIC is Honing in 3rd Party Monitoring

The Consumer Financial Protection Bureau’s has been pushing for more 3rd Party monitoring in the auto lending industry.  As Christine Pratt indicates, “The new FDIC requirements are heavily weighted toward the complete investigation of any third parties’ financial performance as well as ongoing monitoring requirements and will place serious constraints on lenders’ resources. It is incumbent on indirect auto FI lenders to very carefully examine this new guidance and be mindful of a very tight timeline for comments. The original comment date of September 12, 2016 has been extended to October 27, 2016—and right now there are no comments on the FDIC site. This should be an immediate call to action for indirect auto lenders, auto dealers, and the groups that work closely with them (American Bankers Association, National Auto Dealers Association, and the like).”

PointPredictive Offers the Industry’s Only Auto Dealer Risk Monitoring Network

PointPredictive provides the industries first comprehensive tool to help Auto Lenders comply with their 3rd Party Monitoring and risk management.

We began collecting information on the vast network of 40,000 auto dealers in the US to build monitoring tools which can dynamically help lenders understand the fraud, credit and performance risks of dealers across the industry.

From our experience, less than 10% of auto dealers represent the majority of the fraud and early payment default risks for auto lenders. Our Dealer Monitoring Solution provides auto lenders with an early warning system to identify those very few bad apples in the bunch so they can improve their overall loan quality.”

Dynamic Monitoring Updated Each and Every Transaction

As part of the Dealer Monitoring Solution PointPredictive leverages proprietary technology which update s for each and every transaction that a dealer submits to the system.     The dynamic monitoring ensures that good dealers are not penalized, and dealers that are risky are identified at low false positives.

Auto Fraud Network Alerts

As part of Dealer Monitoring, the solution provides alerts from the network about prior fraud experience.  We know from our analysis of 3rd Party Monitoring, the ability for lenders to share their information on fraud risk is critical.  The mortgage industry has been doing it for years, it’s time the auto lending industry had the same types of tools and alerting programs.

Portfolio and Application Level Analysis

As part of the Dealer Monitoring Solution, PointPredictive offers the capability of retro-scoring entire portfolios to mine for the level of outstanding risk.

We often find that the first retrospective analysis is capable of identifying the most serious and risky dealers on the books with a lender.

The retrospective scoring can also help identify underlying loans that represent the greatest risk of default if the lender is servicing the portfolio.

Hosted Service for Rapid Setup

Lenders that are short on time or that require an immediate setup of the service can be setup relatively quickly at PointPredictive through batch hosted dealer monitoring platform.   This helps lenders that are short on time for compliance become compliant very quickly.

Thanks for Reading.

Thank you for reading and if you would like to reach us, contact us at [email protected]


Anatomy of an Auto Lending Fraud


Auto Lending Fraud cost US finance companies, banks and lenders billions annually.  But I often get the question, “How do lenders lose money when they can actually repossess the car if they find fraud later.  The answer is, it’s often not that simple and even when the car is repossessed, they can still lose money.

Early Payment Default

Banks can lose money due to bad loans.  Most lenders lose money when loans default within the first year.  This is called Early Payment Default and it can run as high as 2% of loans in certain segments of lending like subprime, and even higher in deep subprime.   PointPredictive analysis suggest that as much as 30% of these loans contain some fraudulent misrepresentation in them.  The lenders can repossess these cars but they rarely recover 100% of the value and end up taking large losses.

Fraud Rings

Banks can lose money to outright fraud on the loan file by individuals or fraud rings.   These are often very dangerous fraud schemes where luxury cars are often sent to foreign countries in shipping containers after the purchase.  The lenders cannot repossess these cars because they essentially disappear.

Dealer Fraud

The overwhelming majority of auto dealers are good businesses.  However there are a few bad apples that can give dealers a bad name.  PointPredictive analysis suggest a small minority – approximately 3% represent real risk of fraud to a lender.  Common dealer frauds are often linked to recruiting straw borrowers, inflating the collateral value of the car or faking entire loan applications to steal from lenders and banks

This is typically the type of fraud that you will read about in the newspapers because dealer fraud is pretty common and when it happens can cost lenders millions.

Anatomy of an Auto Lending Fraud

To understand Auto Lending Fraud we don’t have to look any further than a recent case at Navy Federal Credit Union.   Navy Federal Credit Union was recently scammed out of $1.1 million in loans by a shady car dealer – Andysheh  Ayatollahi, who eventually  fled to Iran after perpetrating a massive scam against lender he worked with.

Lets look at the steps that Andysheh took to perpetrate his fraud scheme

Step 1 – Buy a Dealership

The first step was to buy a dealership and Andysheh did just that by purchasing a 50% stake in Car Store, a used car dealership in Virginia Beach. He purchased the stake in the Car Store from Reza Azizkhani who would become his partner in crime.  He paid $350,000 for that 50% ownership.  That was back in 2007 and at the time the dealership was operating above board.  But all that was about to change.

Step 2 – Hire Collusive Finance Managers 

It takes a village to pull off the perfect auto lending fraud scam and  Reza and Andysheh were determined to pull it off. To make it really work they needed finance managers  and sales people that would do whatever they told them even if that meant lying to banks, , creating fake documents to back up those lies and shredding those documents if the police ever caught on.

Step 3 – Get Luxury Cars

If you’re going to lie, lie big.  And that is what the Car Store did.  To fill their lots,they didn’t want the cheap economy cars, they wanted the luxury cars that could net them big money.  So they filled their lot with luxury cars like Jaguars, BMW’s, Lincoln Navigators and Mercedes Benz.

The bigger the car, meant the bigger the loan that they could get and that meant more money that they could put in their own pockets.

Step 4 – Get Borrowers – Straw Borrowers

The 4th step in their scheme was to get borrowers that were easily fooled.   They recruited desperate or needy people and turned them into “Straw Borrowers”. These were people with very poor credit and limited financial savvy that could be easily manipulated.  These were people that would let the sales person and finance manager put whatever they wanted on the application without questioning why.

Step 5 – Find a Target

To perpetrate a fraud, they needed to find a target bank that they understood very well.  A bank where they knew the processes, procedures and policies for getting a loan approved.  They set their sights on Navy Federal Credit Union as the target.    This way they could tailor their loan applications in a way that they knew that they could get them approved.

Step 6 – Move Product, Write Fraudulent Loans

Now that everything was in place, Reza and Andysheh were ready to begin the scam.  Starting in July of 2007, they began to dramatically increase their sales. They recruited straw borrowers and started selling cars with one small catch – the loans were far higher than the purchase price and all the information in the loans was fake.

The scam was to inflate the car price

The typical scam that they pulled was to run multiple fictitious applications through with fraudulent information to find the maximum that Navy Federal Credit Union with authorize as a loan.  Then they would inflate the value of the car they were selling to match that maximum amount the bank would lend.  In exchange for the straw borrowers complicity, Car Store would split the proceeds of the excess loan amount with them.  Everyone would win.  Except the bank of course.

Inflate Borrowers Income and Create Fictitious Documents

In mid 2007, for example they sold a 2002 BMW to an employee of a barber shop.  The borrower had poor credit and made a modest salary.  That didn’t stop Car Store from helping they buyer secure a loan on that car for $60,000 dollars.    Under the direction of Andysheh, the Finance Manager falsified paystubs, income and other documentation to show that the borrower actually had two jobs with income of over 8,000 a month.

And that was just the tip of iceberg.  Andysheh and the Car Store would go to any length to get a loan approved. The would even go so far as creating fake paystubs, earning statements or calling the bank to impersonate employers to verify income of the straw borrowers.

The losses racked up quickly

By mid 2008, the Car Store had racked up over 60 loans to Navy Federal Credit Union totally over 1.1 million dollars.  Many of the early loans were starting to go default and the lender was catching on.

As the Feds caught on, Andysheh realized he was in trouble and that is when he started to try to cover his tracks.  He instructed employees to start to destroy all the evidence and had the Finance Manager stay after hours and on the weekends to shed any and all paperwork.

When it was all said and done, Andysheh fled the country to Iran to avoid facing criminal charges.  He later returned late last year to face those charges and faces 27 months in Federal prison.

The Cost of the Fraud

In the end Navy Federal Credit Union and Insurance Companies lost $867, 388.  Close to 90% of the loan value that they had loaned went into loss.  The cost of auto lending fraud doesn’t stop there.  Consumers ultimately bear the burden of these losses in the form of paying higher interest on future loans.

Auto Lending fraud is not a small problem but a growing one here in the US.  The auto lending industry is booming and lending more to more risky customers than anytime in our history.  This is often a precursor of future risk that will be exposed.

New Behavioral Analytics Track Suspicious Behavior

To curtail fraud, lenders should have systems in place to monitor the behaviors of all dealers and the applications that those lenders submit.  PointPredictive offers the industries first Pattern Recognition algorithms to help lenders stop dealer frauds.

Contact [email protected] for more information.


Estimating Auto Fraud Lending Losses in the United States


Auto Lending fraud and risk losses are on the rise in the US.   The numbers are out and they are not good.  Last quarter uncollectible auto loans soared to over $1.1 Billion dollars and delinquencies on subprime loans hit their highest level since the worst recession in US history.  More concerning is that the rapid growth in auto lending is being fueled in part by deep subprime lending to borrowers with credit scores less than 500.

With auto lending origination levels soaring to some of their highest levels in history the downstream impacts are beginning to reveal itself with concerning levels of fraud.

As former experts in Mortgage Fraud Analytics,  PointPredictive is beginning to see history repeat itself – but this time in the Auto Industry.  In 2004, Mortgage lending volumes were hitting new heights thanks to an increase in subprime and non-direct lending through brokers.  Fast forward 12 years later and the Auto industry is experiencing the same types of rapid growth which is fueling the same sparks of fraud risk that erupted in the mortgage industry.

The Hidden Cost of Auto Lending Fraud are in 3 Areas

There is no central reporting agency for US Auto Lending Fraud so assessing the absolute levels of fraud losses in the US is difficult. The cost of Auto Lending Fraud does hit the bottom line of banks and lenders but it may not always be categorized or recognized as a fraud loss.

Auto lending fraud losses are typically seen in the following areas:

1) Early Payment Default –  Loans that default within the first 6 months have a much higher probability of containing material misrepresentations in the original loan application than loans that default much later.  PointPredictive believes that between 30% to 40% of early payment defaults in the auto lending industry can be directly linked back to some type of fraud misrepresentation.

2) Dealer Losses –  Auto Lenders, particularly in subprime lending experience high levels of risk based losses that can be tied back to individual dealers.  Some dealers have extremely high levels of early payment default, known fraud and bad loan quality that leads to losses.

PointPredictive analysis has determined that based on the lender, that almost 100% of their fraud losses may be coming from less than 3% of their dealers and close to 100% of their early payment default losses may be coming from just 10% of their dealers.

Dealer losses may not always be categorized as fraud, however through careful analysis lenders often determine that many of the losses they take are due to intentional misrepresentation at an industrial scale.

3) Fraud Losses – Most auto lenders do have some tracking in place for their fraud losses however it may not always reflect all of their losses since so much fraud is often hidden.  Identity Theft, Straw Borrower, Collateral and Dealer Fraud often top the categories of losses that lenders take on for fraud.

US Auto Lending Fraud Losses are in the Billions

Based on data analysis and industry studies, PointPredictive believes that auto lending fraud loan originations are between $2 billion to $3 billion annually based on the current origination volumes of $600 billion dollars.

auto fraud cost lenders billions

This conservative estimate assumes that approximately 30% of early payment default losses in the US contain material misrepresentation and that identified fraud is running at approximately 20 basis points of origination volume.

By applying conservative estimates to the origination volumes, it is clear that auto lending fraud is not a small problem but one that cost the auto lending industry potentially billions  a year.     Additionally, with loan quality (particularly in subprime) eroding quickly PointPredictive believes that auto lending fraud losses and rates could rise dramatically in the next 18 months.

Applying Analytic Models to Assess Risk

PointPredictive applies analytic pattern recognition models which helps auto lenders determine the latent levels of fraud and risk sitting in their servicing portfolios.  By analyzing application and loan level information, PointPredictive estimates which applications and loans are most likely to contain fraud and misrepresentation which helps lenders and banks understand how much risks is sitting on their portfolios

Additionally PointPredictive deploys these same models with auto lenders to identify risky loans prior to approval so underwriters can take the necessary actions to stop the fraud and risk before it is approved.

For questions, contact [email protected]


PointPredictive is growing and looking for data scientists to join the team

PointPredictive Inc. is looking for data scientists to join our rapidly growing fraud and risk analytics team in sunny San Diego California.

At PointPredictive you will work with us solving complex fraud and risk problems for the financial services industry on emerging and growing areas of fraud risk.

What our Data Scientist Do At PointPredictive

Create predictive models based on big data sets using a variety of tools and algorithms in the cloud

Innovate in fraud and risk modeling by testing new statistical modeling techniques on data to boost predictive performance

Collaborate with our customers to analyze, mine, and aggregate their data into our fraud and risk data consortiums

Creatively solve fraud and risk problems in auto lending, mortgage, check, mobile, and electronic payments using internal and external data sources

Evaluate new data sets to determine their ability to boost detection performance

Collaborate with Fraud Consultants and others in the company to develop and launch new, innovative predictive analytic products to the market

Deliver recommendations to clients and the industry on how to best apply advanced analytics

Join our Team in San Diego

Our offices are open plan and modern and San Diego offers the best quality of life of any city in America.

Successful candidates for the position will have a Ph.D. (or a master’s degree with statistical modeling experience) in a technical field, preferably at least 2-4 years of experience working with data in industry, facility with modern programming languages (e.g. Python) in a Linux environment, familiarity with machine learning techniques (such as neural networks, random forests, deep learning), fluency with statistical and modeling packages such as R and ADAPA, and the proven ability to quickly build and deliver production ready statistical models to clients.

Successful candidates for the position will have a Ph.D. (or a master’s degree with statistical modeling experience) in a technical field, preferably at least 2-4 years of experience working with data in industry, facility with modern programming languages (e.g. Python) in a Linux environment, familiarity with machine learning techniques (such as neural networks, random forests, deep learning), fluency with statistical and modeling packages such as R and ADAPA, and the proven ability to quickly build and deliver production ready statistical models to clients.

We move fast at PointPredictive, so if this position sounds like an attractive or potential match for you, please email your CV or profile link today to [email protected]


The Hidden Risk of Auto Lending Fraud Exposed


Auto lending fraud doesn’t get a lot of attention.  You don’t read about it daily.  It doesn’t make front page news.  In fact most people probably don’t even know what it means.  Well, the same was true of the mortgage industry 10 years ago and we learned some valuable lessons.  Auto Lending fraud will become news someday and it’s important we get ahead of the curve and do things to prevent that.

Fraud occurs when there is too much trust and not enough vigilance

There is a tendency for lenders across industries to rely too much on FICO scores and to trust information that is supplied by borrowers and third parties too much.  When this happens, the lenders can be exposed to more fraud and default risk than they were expecting.

What the Mortgage Industry Learned about fraud

In 2004, the mortgage industry was in full swing. The housing market was booming and mortgage originations were closing in on close to $3 trillion dollars annually.  Everything was seemingly going right, or was it.

Fundamentally things were about get very bad for lenders but they didn’t realize what was brewing.  In attempt to gain market share and keep up with the seemingly endless supply of people that wanted houses, lenders were expanding loan programs and layering risk in new ways that had never been tested.

In 2006, the cracks began to show – primarily in subprime lending and by 2008 the mortgage industry had completely melted down.  Loans were not performing and borrowers were defaulting at record levels.   So what happened to cause such a massive collapse.

1)  FICO scores were trusted too much – Lenders relied on FICO scores too much, however many borrowers with great credit scores were defaulting because FICO scores were primarily built on credit not mortgage data.

2) Lenders trusted the information from borrowers and brokers too much – The second problem that emerged was fraud.  Lenders trusted the information supplied by borrowers and mortgage brokers too much when in fact fraud was commonplace.   Studies indicated that 3% of mortgage brokers accounted for most of the fraud accounts and were systematically conning lenders into making bad loans.  The credit rating agency Fitch even did a study which indicated that 25% of the subprime mortgage loans that defaulted had fraud in the application.  Fraud hurt the mortgage industry dramatically.

3) Warnings of Fraud were Ignored – Lenders were in a race to get loan volume and ignored industry experts that warned of fraud.  It was commonly known that loan programs called Stated Income Programs were also called “Liar Loans” but lenders loaned on them anyway since they assumed they would perform.

4)  Over-reliance on the Collateral –  As housing prices soared property values were skyrocketing creating a market where homes became like ATM Cash machines.  When borrowers needed money they could simply refinance their houses and take the equity out.  Borrowers that could not afford their mortgage payments could simply “refinance” their way out of the problem.  Lenders were lulled into a false sense of security since the default rates were artificially low because of this natural re-aging process.

5) Hard and Fast Rules – Lenders attempted to stop the fraud problem by putting data checks and validations into place.  For example, lenders implemented processes to check pay stubs and borrowers social security numbers against public records databases.  There was a feeling that fraudsters would keep doing the same things that they had always done and that they could stop the fraud.  But this wasn’t the case, lenders relied too much on non-dynamic fraud tools and processes which were easily overcome by the clever fraudsters.

6) The Belief that Fraud Doesn’t Exist – Mortgage lenders rarely detected their fraud so they never reported it.  Because they never reported it, the believed that it never actually existed.  But they were wrong.  Fraud did exist and it was hidden in their early payment default losses.  Studies indicated while reported fraud losses were low, hidden fraud losses were in fact high.  Studies conducted in 2004 indicated that between 30% and 70% of early payment default losses had fraud in the original loan file.  Fraud was simply hidden in the lenders early payment default populations and they did not know about it.

The mortgage industry took these learnings and applied some really great technologies and processes to reduce fraud.  They began adopting fraud scores, pattern recognition technologies and fraud reporting.  They became experts in detecting fraud and they were successful.  As a result mortgage fraud losses dropped by 50% between 2008 and 2010 based on industry studies.

So What is Auto Lending Fraud?

Auto Lending fraud is unique but in many ways similar to industries such as mortgage.  There are about 8-10 primary types of auto lending fraud that impact lenders.

Income Fraud – Income Fraud is one of the most common types of auto lending fraud and it occurs when borrowers or borrowers coached by dealers inflate or outright lie about their income.

Employment Fraud – Employment fraud is commonplace in auto lending.  When borrowers lie about about their employer or employment status this is considered employment fraud and it impacts the performance of the loan.

Identity Theft – When borrowers use a social security number or identity that is not their own this results in identity theft and it has historically been a pretty sizable problem for auto lenders.  Since 2009, lenders have achieve far few losses by using external tools such as Lexis Nexis to verify identity and social security numbers.

Straw Borrower Fraud– The third most common type of fraud is straw borrower fraud.  This occurs when a borrower is either recruited by an unscrupulous buyer or broker to act as the purchaser of the car so that the real party can be hidden.  The straw borrower can be a family member or in some cases someone unknown to the actual borrower.  Straw borrowing is often confused with identity theft since the results are often the same.

Dealer Fraud  –  Dealer fraud may be less common, however the impacts are far greater.  Lenders report to PointPredictive that used car auto dealers often represent their highest risk factor when determining a loan.   Dealer fraud occurs when a dealer or finance manager systematically manipulates or coaches borrowers to misrepresent information on their applications.  Lenders report that less than 10% of their dealers represent an overwhelmingly majority of all of their fraud and early payment default risk.

Collateral “Stuffing” Fraud – Manipulation of the car value through a variety of methods including add-ons that never existed is fairly common with instances of auto lending fraud.  Since the collateral value dictates the amount of money that is leant on an automobile, collateral fraud is often one of the most damaging fraud types to lenders.

Excessive Dealer Markups – When dealers systematically markup loans against certain borrowers such as subprime or elderly borrowers this can be a big risk to lenders.  While it is not necessarily considered fraud it is oftentimes closely associated with fraudulent dealers.  Dealers with excessive markups may be more likely to be engaged in dealer related fraud.

Odometer Fraud – Odometer fraud which is has been declining is another example of fraud against lenders.  If odometers are rolled back the lenders are likely to value the collateral too highly and loan too much money on a vehicle.

Yo Yo Fraud – Yo Yo fraud occurs when a consumer is given temporary registration for a car and drives the car off the lot but the dealer later changes the terms of the sales contract forcing the buyer to accept the new terms.

What are the Cost of Auto Lending Fraud?

There are currently no industry reported fraud losses due to auto lending fraud.  But PointPredictive is analyzing the cost of fraud with lenders that are contributing to the auto fraud consortium.

There are several things that we are learning in our analysis.  We believe that auto lending fraud is running anywhere between 10 to 50 basis points of origination volume in fraud based on the portfolio and business practices.

Early Pay Default Losses are Linked to Fraud – There is some evidence to suggest that approximately 30% of early payment default losses may be related to fraud in the original loan file.  Loans that default within the first 90 days after origination and never cure have a very high correlation with fraud.

Less than 10% of Dealers Account for Most Risk –  Based on lenders surveyed and extensive data analysis, PointPredictive believe that between 3% and 10% of dealers represent an overwhelming majority of that lenders fraud and early pay default risk.  This finding is important because it follows closely with the experience of the mortgage lending industry with brokers.

Fraud Losses are Increasing with Subprime Boom – As the subprime lending industry booms, fraud losses are on the rise.  Experian reports that early pay default losses are rising to their highest levels since 2008.  PointPredictive believes that the eroding loan quality is partly due to increasing fraud on the application.

Exposing the Hidden Risk of Auto Lending Fraud

PointPredictive performs retroactive analysis to help lenders determine their fraud losses.  By using pattern recognition models built by trusted fraud scientist, PointPredictive is able to provide a lender with an independent view into their fraud risk exposure.  There are 3 primary ways PointPredictive provides the service

1) Application Fraud Scoring Model – Using millions of historic applications across lenders, we have created a sophisticated scoring algorithm that determines the probability that a loan contains material misrepresentation.  Each application in a portfolio can be scored and ranked according to risk.  PointPredictive than benchmarks the portfolio against other lenders in the consortium

 2) Default Probability Scoring Model – PointPredictive has also created a Default Probability Scoring Model which can determine the likelihood that an auto loan will default within the first 6 months.  By analyzing the early payment default population, a lender can determine if their fraud risk is also elevated.

3) Dealer Scorecard and Benchmark – Based on historical data from thousands of dealers nationwide, PointPredictive has created a predictive dealer scorecard which ranks each dealer according to their risk level.  The risk model scores each dealer every time a new loan is submitted so lenders can determine how to handle the dealer and the loans that they are submitting.

Best Practice Implementation

PointPredictive Fraud Consultants have worked with over 150 financial institutions and lenders worldwide.  By working hand in hand with lenders consultants have been able to reduce fraud losses and exposure significantly and in some cases have been able to save lenders in excess of $80 million dollars in fraud loss annually.

To reach us, please feel free to contact Frank McKenna at [email protected]

Thanks for reading!


PointPredictive Launches DealerTrace


PointPredictive, Inc. today launched DealerTrace™, a comprehensive analytic solution designed to address auto-lending fraud and compliance risks. DealerTrace helps auto finance lenders manage applicant risks, including early payment default and fraud, and the risk of their dealer relationships.

Auto lending fraud occurs when information on an auto loan application is intentionally misrepresented by the borrower, a sophisticated fraud ring, or in some cases an unscrupulous dealer. When application information is manipulated, the lender may unknowingly underwrite a risky loan. Fraud presents a problem for auto lenders because loans that have misrepresentation are more likely to result in early payment default, a term lenders use to indicate when no payments are ever made on the loan.

“Our analysis and experience suggest that less than 10% of auto dealers represent the majority of the fraud and early payment default risks for auto lenders,” said Frank McKenna, Chief Fraud Strategist at PointPredictive. “DealerTrace provides auto lenders with an early warning system to identify those very few bad apples in the bunch so they can improve their overall loan quality.”

DealerTrace uses pattern recognition, a complex statistical technique that has been perfected to detect fraud based on historical data mining. The solution analyzes historical patterns of fraud, early payment default and risky dealer activity and scores each application as it comes in to the lender from a dealer. Lenders are automatically alerted when an application has a significant number of application anomalies or presents known fraud patterns. The lender can then review the application and take action before it is approved. Over time, if a particular dealer submits many applications with similar fraud patterns, the solution will alert lenders to take appropriate dealer action.

With the launch of DealerTrace, PointPredictive is forming an Automotive Lending Fraud Consortium and is encouraging lenders to share their fraud data. By pooling data at an industry level, PointPredictive can help lenders aggregate their fraud knowledge, identify new fraud patterns more quickly and in turn reduce their risk by getting in front of the fraud.

“Creating and managing a fraud consortium is a core competence for PointPredictive,” stated Tim Grace, Chairman of PointPredictive. “Members of our team used the same approach for mortgage lenders a decade ago, which resulted in reductions of 50% or more of their fraud and default losses.”

To join the Automotive Lending Fraud Consortium or to request additional information, contact [email protected]


Modeling Updates from PointPredictive


We’ve Been Busy With Some Cool New Analytic Products

We’ve been heads down busy at PointPredictive the last couple of months launching new solutions for new markets and things have been progressing nicely.

Our focus has always been to launch new analytic solutions to serve markets that need it.  In particular we’re focused on delivering analytics to both the analytics and alternative credit markets and we’ve made some significant progress on those fronts.

Automotive Lending – We’re Launching a Data Sharing Consortium and Fraud Analytic Suite

We’ve spent the last 6 months working with fraud experts, data partners and auto lenders to finalize some groundbreaking new solutions for the auto lending industry.

In particular,  PointPredictive will be launching the first ever fraud consortium for auto lenders in combination with some groundbreaking analytic products.  We’re looking to solve the problem of risk with a combination of application models and dealer risk monitoring solutions for fraud, risk and compliance.

Our release is scheduled for March 2015 so keep your eyes open for the big announcement.  Contact [email protected]


Making it Smart and Making Customers Happy in the Process


Having just completed another successful project in record time we were thrilled when our client described our work, exactly as we would have hoped they would when we started the project. – “Point Predictive more than met our expectations..”,  “their team was available, responsive and really did an exceptional job of bringing their significant expertise to bear on our business”.  These kind words were literally music to our ears and that’s why we started PointPredictive.

Our business model has always been to partner with our customers. We bring them our expertise in data, risk and modeling and help them solve business problems quickly in a way that they can understand and use.  Through our most recent project we did just that by focusing on the business issues and working on a tactical and practical solution using smarter science.

There were many things that made this engagement successful but I felt there were 5 reasons that made it really work.

1. We started by understanding their business before we gathered the data.

There is intelligence you can derive from data but it first requires a deep understanding of the business issues to put that data into context.  That’s why we spent an entire week, working hand in hand with our customers onsite with them.  We needed to understand their business problems that they faced everyday before we could help them solve them.  After the week was over, we were able to gather years of historic data and make sense of it when we got back to the PointPredictive lab to get to work.

 2. We didn’t lock ourselves into a single analytic technique.

We realized pretty quickly on that to solve this particular issue, we were going to need to be creative and think outside of the box.  So, we decided very early on to be analytic-agnostic.  That is to say we would use any analytic technique that delivered the best predictive value while being simple enough to put into production quickly and be useable by the business.  We tried scorecards, logistic regression and even experimented with deep learning techniques.  In the end we found that a combination of techniques worked the best.


3.  We scheduled regular meetings to discuss progress and raise issues.

We are big fans of iterating a solution until we get it right.  So we scheduled weekly calls with our customers to report on results of the analytics and to get their feedback through the entire process as quickly as possible.  We didn’t develop the model in a vacuum, we created the model in an iterative process hand in hand with our customer.

4. We measured model performance for them in many different ways and focused on performance.

We took great effort to deliver a full set of performance charts for our customer.  Not only did we measure the models ability to detect risk overall, but as well its ability to reduce financial risk over time.   A model that does not help reduce financial risk has a very hard time of adoption since their is no financial incentive.  We made sure that our performance charts clearly demonstrated the benefit to improve adoption of the model.

5.  We helped to operationalize the model.

The model was extremely predictive.  We were able to show that with our performance charts. That only solves half the problem however and we needed to help our customers understand how to use that model in various aspects of their business.  That’s were our business consulting backgrounds really came into play.  Having worked with over 100 lenders, banks and financial companies on analytics and risk we were able to bring our significant expertise to bear.  Operationalizing the model helped bring it to life in a way that made sense for the business to use.

It’s so great to be able to help our customers and enable them to achieve their business goals.  That makes all the hard work we do everyday worth the effort.

For questions, please contact [email protected]


Deep Learning Presents Exciting Opportunities to Advance Analytics for Business

One of the most exciting recent developments in the area of predictive analytics is called deep learning.  Having already advanced the state of the art in many areas such as image recognition, deep learning is now poised to deliver cutting edge results for business application clients in the real estate, mortgage, and auto lending industries.

When building a statistical model, access to quality data is key.  But few data sets in industry are fed directly into a machine learning algorithm.  Instead, the raw fields are combined and transformed to produce features, which highlight important aspects of the data and problem being addressed to a model.  Features are also a way to incorporate human business expertise into a system.

One of the challenges with features is that their creation can be time consuming, and nowadays delivering a high performing, robust system quickly is especially important.  It is also difficult to be sure if a hand created feature set fully captures the data, and using more brute-force approaches to feature creation leads to a dependence on feature selection algorithms.

Enter deep learning.  By training and stacking multiple layers of Restricted Boltzmann Machines or auto-encoder neural networks, a system can now construct its own complex internal representations or features automatically.  And these layers can be built without the need for data tags or labels, which are frequently unavailable or incomplete in big, real world data sets.  Upon the newly learned features a more typical modeling technique such as logistic regression is then used to produce the final business score.

Here at Point Predictive Inc., deep learning is one of the approaches we use to build predictive features and models more quickly.  Not only faster, the transformations learned are not typically present in hand crafted feature sets, leading to more complete models.  When coupled with solution delivery in the cloud, deep learning is helping to deliver strong model performance within short project timelines.  Can deep learning play a role in your next analytic business application?


5 Things that Make a Real Difference in Risk Modeling


Having recently completed another risk model this week, I was reflecting on some of the things that factor into building a model.  And not just a good model, but a great model that can transform the way our customers do business.

I realized that 5 things were really important.

Understand the Data – I mean REALLY understand the Data.

The biggest mistake you can make in risk modeling is not understanding each and every field in a dataset and what drives the data into it.  So many models are built in a vacuum where a scientist is not given access to the business. They review datasets without understanding how that data is used in the business operations.  When this happens, assumptions are made about the data and sometimes those assumptions are wrong.  When we build a model at PointPredictive, our scientists spend an inordinate amount of time with the client, understanding each and every data element.  It’s a painstaking process but the end result is a model that works.

Focus on the REAL problem.

In risk modeling, it is quite easy to get side-tracked and end up building a model that focuses on the wrong problem.  The advice we always give ourselves is – “follow the losses”.  If you build a risk model that reduces risk and that in turn results in reducing losses then you are solving the real problem at hand.  Focus on the REAL problem when building a model and you’ll be successful.

Iterate early and often, getting feedback from the business along the way.

Years ago, a customer conveyed to me a story of a failed project that ended up causing their business a lot of frustration.  They waited months and months for model to be delivered to them. When the final model and results were delivered it didn’t address the business problem at all and as a result was never deployed.  The fact is, great models are typically the product of lots and lots of small, incremental iterations.  After each iteration, the business provides feedback and adjustments and improvements are made.   Not only do the models improve from the frequent feedback over time but the business will end up understanding the models and that in turn will improve the odds that the model will gain acceptance.

Measure Performance of the model from every perspective that makes sense

I am not sure you can ever focus too much on model performance measurements.  As a company, we spend a lot of time working with our customers to show them how our models perform.  How much risk does the model detect?  What is the false positive of the model at each score band?  What is the number of cases created?  How will the model reduce losses over time?  These are all questions that we try to answer for customers and it makes a real difference.

Be there for Go-Live.

The first day a model is deployed is a critical one.  It’s the day that everything comes together and the end users will start to use your model.   It’s also when a lot could go wrong.  Data feeds, systems, or network connections can cause problems that impact your model. Or, in some cases,  the end users may have a problem understanding how to use the model or scores in their workflow.   Being there for Go-Live to make sure everything works according to plan is a very critical phase in the model development process.

If you look at the things that make a real difference , in risk modeling,  they are all about the same thing – bringing the business and science together.  A great model is a collaboration between operations and analytics.  The more you can bring the two together and collaborate throughout the entire modeling process, the better the end solution will be.  That’s what we believe and that’s our focus.

For questions, please contact [email protected]