How Truework uses machine learning to accelerate verifications
“AI solutions need to be more than gimmicks. We are in a heavily regulated and compliance-driven industry. We need to dot all of the i’s. Not most of them.”
Building for Trust
Since our founding, we have incorporated machine learning into the Truework platform while maintaining a constant focus on compliance and accuracy. This balanced approach is how we have been able to consistently deliver the most value to both customers and consumers.
Applying machine learning to use cases involving financial data requires patience, craft, and attention to detail. We are lucky to have some of the smartest and most thoughtful engineering minds working to accelerate financial workflows using these technologies.
After years of putting in the work, here’s where we stand when it comes to applying machine learning to income verification:
- Never compromise trust and accuracy by modifying source-of-truth data. I'll highlight some of the creative ways we are streamlining access to source-of-truth data, but machine learning can never modify the data itself.
- We need “humans in the loop” to supervise and correct recommendations. With the current technology, fully autonomous agents are not useful beyond gimmicks and are not appropriate for financial services use cases.
To maintain accuracy and uphold our commitment to compliance (which requires strict tolerances on false positives), we deploy human supervision in all use cases. When needed, a Truework team member will jump in and complete a task manually instead of introducing unnecessary risk to data integrity.
This focus, drive, and discipline over the years, coupled with these tenets, has enabled us to build a truly technology-driven verification platform leveraging data at scale.
Here are a couple of examples of how we’re already leveraging machine learning.
Intelligent Orchestration Flywheel
In our industry, the best way to capitalize on machine learning is using this new technology to make a faster, more efficient verification platform that does not rely on a single data source.
A better platform requires intelligent orchestration. When a customer submits a request to Truework, they don’t need to resubmit it if any individual method is unsuccessful. Instead, Truework orchestrates each request across the platform, seamlessly engaging with one or multiple verification methods simultaneously until the request is complete.
We accelerate completions by identifying the optimal verification method using over a dozen employer and employee attributes. We then deploy machine learning to constantly optimize the time-to-verify by ingesting new data from the thousands of requests we process every day.
Employer Standardization
Let’s start with an employer name as an example. Large corporations with complex subsidiary structures (eg. Alphabet has numerous subsidiaries Google, Nest, Waymo, Wing, etc.) are difficult to standardize and navigate. This problem type is one of many challenges faced when trying to standardize employer names for hundreds of related entities.
By using OpenAI’s embeddings, our platform can now standardize corporate hierarchies and solve other categories of issues to orchestrate requests faster.
The Google example highlights how we have been able to accelerate the time-to-verify using an orchestrated approach.
Making Smart Outreach “smarter”
One of Truework’s most effective verification methods is Smart Outreach, our automated solution for otherwise manual verifications. Over the past 6 years, we’ve been using technology to take what used to be a 5+ day manual verification process, and make it faster and more efficient. What began as a call center (and still is for most competitors in our space) has transformed into a cutting edge machine learning platform. Machine learning helps us accelerate verifications by constantly mapping and optimizing contact information for hundreds of thousands of employers.
Once we’ve verified employment for a net-new employer, we can automate the standardization and routing for every subsequent request, eliminating manual processing. When we come across a new employer, we are able to recommend the best tasks using look-alike data models, helping our team accelerate hard-to-verify employers.
Document Extraction
For some consumers, the only way to verify income requires a document to be shared. This may be in the form of a W2, 1099, tax return, child support payments, or many other types of documentation. To complete document-based verifications, we utilize different forms of machine learning to extract the data from an array of documents and convert them into standardized income reports (while also running multiple fraud checks simultaneously). This process can be complex but also highlights where a human-supervised process is vital.
Estimated Income
Estimated Income helps identify and flag errors in reported income, reducing the need for teams to manually check income numbers.
Our estimation model takes into account a large set of reference and aggregated verification data to understand income across various inputs (e.g. job title, employer). From there, it organizes, cleans, and labels the data to be matched against what is found during a verification process, removing bias and errors.
It takes the reported annualized salary of the applicant along with other inputs and tells us if that number fits within the identified income range. In other words: Based on what the person told us about their income and employment, does it fit within the range of what we expect that person to make?
Now let’s look at how this all comes together. When we receive a verification request, we apply multiple instances of machine learning to initiate the most efficient orchestration for an employee and employer combination while maintaining data integrity and accuracy. Our platform can now route hundreds of requests simultaneously across every major verification method. This efficiency is transferred to our customers in the form of speed, accuracy and cost-savings.
The Future
Our experience over the past few years has confirmed one of the core hypotheses we held when founding Truework: the income verification space is ripe and ready for meaningful innovation.
We will continue building machine learning into our infrastructure to make our platform smarter and faster for every customer and consumer. Ultimately, this measured and patient approach is how we fulfill our vision faster of creating trust in every financial transaction.
Published on September 26, 2023 | Updated on March 12, 2025
Ready to unify your verification strategy?
Join the thousands of lenders who use Truework Income as the one-stop verification platform.
