Contact Us

Category: Articles

The Big Data waste: Missed opportunities in mortgage servicing

The term “Big Data” has been buzzing about the technology industry for several years now and has crept itself into the vocabulary of business managers and corporate execs to mean the new must-have for the modern enterprise. Even though there may be great promise in the use of Big Data to solve complex business problems, I find that the concept itself is largely misunderstood. Making matters worse, Big Data has also become somewhat synonymous for “Big Company” and “Big Budget” – a luxury reserved for the Fortune 500 with large slush funds to spend on consultants to help figure out whatever alchemy Big Data is intended to accomplish. There certainly is truth to that, but the landscape is rapidly changing. With an increase in accessibility and simplicity of tools designed to wield Big Data into something meaningful, just about any company that produces data can use it to its advantage.

 As a software and service provider in the mortgage industry, I can confidently say that the volumes of data produced by companies in this space are astronomical, to put it mildly. I also dare to say that, for most of these companies, the wealth of information that could be extracted from this data is completely wasted. After speaking to a couple executives and frontline managers about their perceptions on Big Data and its analysis, I see a pattern contributing to the active disinterest toward exploring data sets:

  • General misinformation about Big Data, and all that comes with that, and
  • A lack of imagination about how data can be used to make a direct and meaningful impact on operations.

Perhaps it is far too ambitious to say that this article will solve both (or either) of these issues. But… maybe a brief introduction to Predictive Analytics and how these can be applied can help prompt a shift in this mentality. It is all about knowing how to ask the right questions.

What is Big Data Anyway?

Unceremoniously, Big Data is a large data set. Enormous actually. More specifically, it is a data set that is so large it requires special technologies to house, manage and analyze the information, as conventional tools prove inadequate or impractical. The term, however, has been expanded mostly thanks to marketing efforts of firms offering services in this space to also include the methods and practices available to interpret the data. Each marketing piece and slogan developed around Big Data has contributed to obfuscating its definition while brining some level of very marketable mysticism. Sales pitch aside, Big Data is the collective term for the troves of information produced across an enterprise.

What can Big Data tell us?

Well, it really depends who wants to know and for what purpose. That’s the key – identifying a specific purpose. Without clear ideas linked to measurable results, looking into Big Data is like getting a bunch of answers for which there are no questions; i.e. lots of information about nothing we care about very much. It is hard to come up with these questions. It is even harder when we don’t know how best to frame these questions to get the meaningful answers we might be expecting. I’m not a data scientist, and frankly, some of the theories and a lot of the math behind Big Data is beyond my grasp. I think in business processes and software-driven solutions. Learning about Predictive Analytics and its application has completely changed my attitude about Big Data, opening up a new playground of productivity. Here’s a little about how it works and how best to start thinking about applying Predictive Analytics so you can start phrasing your own questions.

Predictive Analytics = Forecasting the Future

Forecasting the future is not the same as seeing the future. Predictive analytics uses the sorcery of mathematical modeling and machine learning to predict the outcome of specific scenarios given some data inputs related to the process. The mathematical modeling component helps measure the likelihood of the different scenarios happening given past and fresh data inputs. Machine learning is the really exciting piece; it takes into account historical outcomes to better predict the likelihood of different scenarios, AND even predict new scenarios given patterns and nuances that only machines can identify in the data. So, the wrong way to ask Big Data a question is “Will [this scenario] happen?” – that’s seeing the future. An appropriate question for Big Data would be worded as “How likely is [insert scenario] of happening?” or “How much more likely is [this scenario] to happen instead of [this scenario]?”.

Opportunities Missed

For an industry that so heavily relies on data, it is somewhat crazy to me that this same love for data has not extended to using predictive analytics. The possibilities are virtually endless, only limited by the process owners’ imagination. Below are some quick ideas of questions I’d be asking Big Data to help tighten business processes and operations, albeit, from the high level perspective of a solution provider to the industry rather than the pointed precision expected from a business process owner:

 In Originations:

  • How likely is a candidate to close on a loan? How long with the process take? Where in the process chain can we expect hold-ups?
  • What loan products will work best for a particular group of prospective borrowers? How many post-close problems can we expect? What percent will represent buybacks?
  • How will volume be affected given a new incentive or program? Over time, does the metric hold true?

In Servicing:

  • How many loans are likely to have modifications, delinquencies or become paid in full? What factors are directly contributing to the portfolio performance?
  • What population of loans will have reconciling items? What is the expected source and resolution of these items? How many of these items hit 90 days or go to Reserve?
  • How many errors can we anticipate in a given business process? Where will these errors likely come from? If we implement a change, what might be the effect? Once implemented and over time, will these assumptions hold true?

These are a handful of questions in just two areas within the vast world of mortgage operations. Now that you have a framework for how to ask questions of Big Data, what would you like to know? How would you manage if you could predict the future? Where would you invest capital? Would you buy a lottery ticket? Wait, Big Data and predictive analytics cannot see the future.

Showing Cashbook Some Respect

Often overlooked and mostly oversimplified, the Cashbook process presents an important opportunity for reducing rework and increasing efficiency in Custodial reconciliation. During more than one occasion, I’ve heard people in Investor Accounting call it a mere formality; a means for validating the depository balance. Some have gone as far as not considering Cashbook its own process at all, but simply a data input to the real star of the show: the Test of Expected Cash (TOEC). 

Their rationale? Any outages in the account would just fall out while calculating the loan-level TOEC, so performing a full Cashbook reconciliation seems somewhat redundant.  I tend to agree, in principle. However, my experience has proven the opposite in certain situations. Any time savings gained in abbreviating the Cashbook process are more than lost when researching certain outages in TOEC.

At a basic level, the goal of Cashbook is to ensure the Custodial bank account is in balance. At a deeper level, the Cashbook process presents an optimal tool for certifying the bank statement (i.e. via performing a transactional book-to-bank reconciliation). This is good because collections, for example, recorded in the Servicing system would match deposits in the bank statement with any discrepancies falling out as reconciling outages. Yes, TOEC should catch these same discrepancies. 

How about this scenario: a wire is coded incorrectly and ends settling within the wrong P&I account? The TOEC process should also catch this, but the outage would not be linked to loan-level activity as it is an account-level item. In a sophisticated TOEC process, the outage may be caught early without missing a beat. If the process is not designed to specifically handle these scenarios, things start getting ugly. It may take analysts a lot of extra digging to identify why loan-level activity does not match up with the account balance. 

Also, consider how this outage would be recorded in TOEC. Is there an appropriate root-cause category code for it? Maybe; probably not. Lastly, consider timing (chronologically, not Reg-AB time). By the time the outage is identified in TOEC, this money may be in the incorrect Custodial account for 30 (maybe even 60) days, idle. Then, depending on the process, it may take another 30 days to initiate the transfer and move the money. Another good example for wasting time in TOEC: researching and correcting a true bank error.

From my perspective, all this could be avoided with a disciplined and well-structured Cashbook process; a proactive approach to handling account-level items that get resolved before they reach TOEC. It is time to show Cashbook some much well-deserved as past-due respect. In honor of this neglected business process, I am proposing 5 considerations for building a sound practice within your operations:

1. Clearly Define Start and End-date Parameters   

Avoid the common mistake of overlapping Cutoff start and end dates by double-checking data filtering parameters. This could get tricky as not all Cashbook reconciliations fall on month-end (think FHLMC) and processing cycles do sometimes become extended to work on a non-business day. In other words, verify that all activity for the bank statement is restricted to this range and that no book transactions enter this process ABOVE the defined range (consideration #3 below will explain why some book transactions from the previous period should be considered in the process). Not following this simple guideline will lead to a lot of transactional “noise” and a disorganized Cashbook reconciliation.    

2. Roll from a Previous Period

This may sound intuitive, but it is surprising how many times we’ve encountered companies performing their Cashbook reconciliation without considering results from the previous period. The key lesson here: it is important to live with your results (and calculations). The true power of a Cashbook Reconciliation summary is in rolling it forward; in other words, start by tying together Beginning Balance from the current period to Ending Balance of the previous period. Also, make sure to carry-forward any reconciliation discrepancies identified in the previous period to attempt resolution or continue ageing (see item #3 for more detail).

3. Track and Age Discrepancies        

The only sound method for identifying reconciliation discrepancies within the Cashbook process is to perform a transactional book-to-bank matching of bank statement items. This means bumping up collections recorded in the Servicing system, for example, and matching them with deposits on the bank statement. The benefits of this process are two-fold: (a) matching book-to-bank transactions certifies the bank statements (i.e. the backbone of the entire Custodial recon process); and (b) the process will reveal any true discrepancies /reconciling items in the Custodial account. Please remember to roll-forward any book items not matched against bank statement items (i.e. deposits in transit) for the following cycle.

Adding some additional sophistication to the process, book-to-bank reconciliation could be performed on a daily basis. Bank statement data is available daily via BAI files and there are several reports in Black Knight and other Servicing platforms that provide daily activity, such as the T690 showing daily collections (i.e. daily version of the ZZ80). Performing this reconciliation on a daily basis catches issues quickly and allows those involved to correct the issue well before this becomes an outage in TOEC.

As far as best practice – track and age any reconciliation discrepancies at the Cashbook level (even if you might be tracking certain outages “twice” if these are also identified in TOEC). Why? The majority of outages in Cashbook will fall under one of two main categories: (1) errors in movements of cash; or (2) true bank errors, such as incorrect settlement amounts. For these types of issues, communicating the discrepancy with corporate treasury, for example, will be more effective at the bank-account level. This, in turn, should reduce the turnaround time for resolution and possibly correct the item before initiating TOEC (particularly if performing this reconciliation daily).   

4. Validate ALL Balances

The clear figure to validate here is the Depository Bank Balance. The Depository Bank Balance should be composed of the ending bank balance on the bank statement PLUS any deposits (or withdrawals) in transit that are yet to settle in the account. If your process is already taking account point #3 above, this value should be simple to certify.

Another important balance to validate is the depository balance according to the Servicing system. It may sound slightly counter-intuitive, but there could be a discrepancy between the calculated Depository Bank Balance and that which is presented in the Servicing system – think adjustments not entered correctly or manual transaction activity not recorded accurately (or at all) in the Servicing platform. We’ve found that it is best practice to perform a simple daily check to make sure both these values are in synch. 

5. Track and Measure the Process.      

All the considerations leading up to this one center on ensuring a sound Cashbook reconciliation, which is fantastic; however, visibility and metrics gathering over the process as it is happening in real-time distinguishes a proactive team vs. a reactive team. What’s the difference? A reactive team sees smoke and eventually reaches the fire with whatever tools happen to be on-hand to try to extinguish the flames. A proactive team sees the spark that started the fire. This level of visibility is afforded by adopting well-defined work assignments and developing a dashboard to track the resulting metrics. 

We recommend doing what most companies already do: create a spreadsheet to assign analyst resources to specific Cashbook reconciliation, but we push it one step further by suggesting the inclusion of triggers to track the progress as it is happening. Create a spreadsheet or tool that listens for status changes in Cashbook reports (i.e. Pending, Submitted, Approved) as well as a means to collect metrics (i.e. number of items matched vs. outstanding) in an effort to get a meaningful pulse of the process as a whole. The development of the dashboard is certainly an evolutionary process; the trick is to subscribe to this mentality or management overview philosophy if the terminology is more fitting. Either way, evaluating the health of a process needs to occur as the process is happening and not after the process is completed – test this statement by applying it to a living body. Find creative metrics (and corresponding triggers) to track the process as it is unfolding to prevent a spark from becoming a forest fire.

Below is an example of real-time processing metrics as offered within Integra Recon. To get the full picture, it is not only important to see the status of current work completed (left chart), but understanding when the bulk of the work was performed (right trend analysis).

screenshot custodial reconciliation dashboard

What considerations can you share about how you manage your Cashbook business process?