Getting Data Certainty in Real Time (sponsored by Paradatec)
Lee Gillispie is Managing Principal at Paradatec
At best, bad data slows workflows and adds costs to correct it. At worst, bad data can affect underwriting decisions, delay closings, trigger unexpected investor pricing adjustments, or contribute to repurchase demands. Beyond its impact on per-loan profitability, bad data also costs you in lost confidence with your loan officers, brokers, borrowers, and investors.
What creates the conditions for the accidental ingestion and use of bad data? Most of the time, the challenge is establishing trust in incoming data as quickly as the business needs to use it. Documents arrive continuously, decisions are time-sensitive, and bad data can move through several process steps or even multiple systems before anyone discovers it.
Data certainty means knowing where a value came from, what checks it passed, and whether it can be considered usable. It requires accurate capture, validation, and traceability to the original source.
Previously, we shared how making existing loan and portfolio information structured, searchable, and comparable creates two strategic advantages: the ability to interrogate the collective intelligence within loan files and portfolios, and the ability to use that intelligence to support workflow automation, including agentic AI.
That article outlined the strategic benefits of accessing and leveraging more of the data you already have, and ensuring a robust and trustworthy data repository as the foundation for strategic initiatives such as proactive data analysis for risk management and opportunity harvesting, to using data for designing and supporting automated workflows.
Those advantages begin with trustworthy data. Here, we shift from the information already in your portfolio to the documents and data about to arrive—and three practical ways to establish data certainty in real time (as it’s delivered to you).
1. Capture and validate incoming data at the point of entry.
The pace of incoming documents and data varies, but the potential to go from “normal” to “swamped” is always a possibility. Borrowers send in supporting documents in multiple sporadic bursts, while brokers and correspondents send information in larger waves of data; and servicing or investor portfolio transfers can equal tsunami-level waves of incoming data to be processed.
And no matter what speed and scale you are receiving data, your customers and business partners are always measuring how fast you respond.
The most efficient lenders and servicers we’ve seen set up ways to absorb and acknowledge data at the speed it’s delivered, and then quickly review it based on intelligent prioritization, while also cleaning it before it becomes official reliable data.
Enabling other systems to quickly ingest and use it, whether you are using a data standardization framework to translate data into multiple systems or a centralized data warehouse that your systems call into, is key for internal efficiency and responsive customer feedback.
Previously, accurate data capture required various levels of time-consuming human oversight or significant technical investment. Now, APIs and Agentic architecture can enable the flow of documents and data in high volume environments as hyper-specific lightweight AI models for mortgage documents and data return remarkably accurate results. Best of all, the amount of data that can be captured quickly has also increased.
2. Preserve the evidence behind every data point.
Context establishes reliability for captured data. That makes the audit trail of how it was captured is just as important as the finalized data value, especially when using an AI-enabled data capture approach (regardless of whether it is based on deterministic AI or generative AI or both).
Documenting the context of how each data point was captured, are key for defending the result, and being able to replicate it, including:
- all model details and inputs used to finalize that decision
- unique identifiers for how it was sourced, right down to the coordinates on the page of a specific document instance
- any data formatting, normalization, or transformation applied
- other data provenance details
This evidence allows a user to move from a field in a system back to the supporting document and examine how that value was established.
Context for data capture really means you can show your work “on demand” (or more commonly, your system’s work) — for traceability of the inputs and transparency of the output. Retaining this metadata enables you to demonstrate repeatable results, and measure consistency of results over time.
Creating confidence in data capture results means being able to verify them on demand, and over time for ongoing measurement of data decisions.
3. Cross-validate data before releasing it into downstream workflows.
Being able to reasonably rely on data for lending decisions means cross-checking incoming data, and where possible, cross-validating to data you already have.
For example, triangulating data within a document, such as validating “the transaction math” on a document like a bank statement or paystub, across documents from the same document issuer such as across all bank statements, across documents with similar data, and other checks can help identify a potential fraudulent document.
Cross-checking data consistency across documents, including options for semantic matches vs. strict matches, and validating document completeness streamlines audit reviews.
And of course, evaluating data values from source documents to datasets received or already ingested into systems of record, can highlight potential discrepancies. These comparisons can reveal issues that would otherwise carry forward into underwriting, investor delivery, servicing, or portfolio analysis.
Being able to manage these data validation checks, including the ability to modify them and change under which circumstances you want to use them, is also critical. And results should distinguish data that passes data evaluation from evaluation issue categories such as missing or incomplete documents, failed data comparison checks, and other exceptions requiring review.
Put trusted data to work sooner.
Mortgage lending and servicing operate against many clocks: rate lock expirations, closing dates, investor delivery deadlines, servicing transfer effective dates, and borrower response needs. Errors discovered late consume time when there is little time left to spare.
With today’s technology, data certainty can deliver results faster than ever before; and now, before incoming data is ingested for workflow process use. However, humans need to know how to monitor the results of any automation – whether it’s automated data capture, robotic process automation, or their agentic AI assistant. Trust, but verify – on demand.
Establishing data certainty at entry gives downstream systems a stronger foundation and enables them to run more efficiently. Being able to capture and clean data before it is ingested makes it ready for use at the speed your business requires.
(Sponsored content includes material submitted independently of the Mortgage Bankers Association and MBA NewsLink and does not connote an MBA endorsement of a specific company, product or service. For more information about sponsored content opportunities, contact Bill Farmakis at bill@jlfarmakis.com or 203/834-8832.)
