The decision to build versus buy needed software functionality is a critical one that has a big business impact. Among the many factors to weigh – cost and time are the biggest. While it is typically more expensive to buy a pre-built application rather than building functionality, the latter adds an order of magnitude in deployment time and pulls staff resources away from other projects.
Let’s look at the factors comprising the decision to buy or build an application specifically for clearing stipulations in consumer lending. At Informed, we encounter this issue, and it boils down to three main options: Optical Character Recognition (OCR) technologies, Robotic Process Automation (RPA) platforms, and Informed.IQ.
Although there are many OCR technologies available, they have limitations. A major limitation is accuracy. OCR systems rely on algorithms to recognize patterns in the images of text. And they struggle to recognize distorted text, unusual fonts, or bad copies/scans of documents provided (i.e. faxes of copies) in the lending process. As a result, OCR systems can introduce errors into the digitized text, leading to incorrect decisions in the lending process without a level of validation, transformation, and standardization.
Second, OCR technology is expensive to implement and maintain. Any project using OCR needs continual training to increase the accuracy of the models that classify and extract data from documents and to minimize and test for bias. As a result, entire processes must be built to monitor and maintain the system.
Third, any OCR platform needs training on relevant documents. This includes understanding the type of documents and extracting necessary data. It assumes that the documents are separated into individual files. Otherwise you must include the logic to split/join the documents. This adds up to additional costs and time, which can be overwhelming.
RPA platforms are a popular technology that streamlines processes and reduces costs. However, process selection is just as critical. First, RPA Platforms are best at automating simple, repetitive processes. The loan origination process contains complex tasks like income calculation and fuzzy name matching. And it changes often as internal policies and external laws change, requiring changes to the RPA rules. These changes break fragile rules built into the platform, requiring additional maintenance resources.
Second, RPA platforms still require the concept of OCR to extract required data. And since many RPA platforms contain OCR capabilities, they have the same problems as OCR technology. Models must know how to split and/or join documents. And you need models to classify documents and extract data from the classified documents. This whole infrastructure must be built, tested, and maintained by a dedicated team.
Third, rule changes are challenging to implement within RPA platforms. It is difficult to identify which rules need updates, and that process is time-consuming.
The final application we’ll discuss is Informed. While Informed leverages OCR platforms for document images, it’s only one piece of the solution. Informed has been building the models to understand document classification and extraction for more than 6 years. We’ve processed over 50M documents in the consumer and auto lending space for clients such as Ally, Capital One, Drivetime, Westlake, Crescent Bank, Origence and more. New clients automatically benefit from these pre-trained models and leverage the high quality data start clearing stipulations from day one. Informed can already take a single, 50-page PDF and split it into the 15 different documents within and extract the required data to clear lender-specific rules.
When it comes to rules, this is where our experience matters. Not just simple rules like matching values across documents. Rules like matching employer names between applicant provided names (ie Popeys), and the associated franchise name (MMSC Centreville). Also, rules that understand the dealer name on an auto contract that doesn’t match the dealer name on a vehicle service contract is actually the same dealer. Rules that dynamically stip for state level documents when applicable documents appear in a deal jacket, and rules that know when the sale of a GAP contract is out of compliance exposing a lender to risk.
More Than Rules
And it’s not just rules. Informed calculates income across multiple document types that match lender rules. First 90 days of the year? Informed can use the hourly-pay method to determine someone’s current pay and extrapolate to their annual pay. Multiple employers, Military LES, SSI Income? – not a problem. Multiple sources of income are handled appropriately due to classification and extraction models providing the necessary data.
Lenders must know if customers mis-state their income or provide fraudulent paystubs. Informed’s database of 20M fraudulent paystubs grows every day allowing us to alert our customers of potential fraud before a loan has been boarded.
Integrating with existing systems is often a blocker for lenders. Rather than building separate integrations, lenders leverage Informed’s prebuilt integrations across the lending community. Informed’s integrations include F&I Sentinel, Wolters Kluwer, Finicity, Plaid, Argyle, and Pinwheel, enabling simple integrations for compliance and additional income data sources. Informed easily integrates into numerous loan origination systems (LOS) such as Fiserv, FIS, Origence’s ArcOS, Wipro, Defi, Solidify and others.
Adine Deford is the VP of Marketing at Informed.IQ. She has more than 25 years of technology marketing experience serving industry leaders, world class marketing agencies and technology start-ups.