Infrrd and the paperwork that refuses to behave

Consider a mortgage file in which every number has been read correctly. The salary is legible. The bank balance is legible. The borrower’s name is legible. Yet the file may still demand a small army of people to decide whether those numbers belong together. A scanner can turn a page into text. A loan reviewer has to turn a pile of pages into a defensible conclusion. Much of Infrrd’s business lives between those two accomplishments.

Infrrd sells intelligent document processing, or IDP: software that sorts documents, extracts information and validates it against business rules. It also sells specialist mortgage and insurance tools, plus processing services with human review. The name sounds as though a typesetter lost interest halfway through “inferred.” That omission is deliberate. You supply the missing vowels from context. The company would like its software to perform a considerably more demanding version of the same trick.

01 The trouble starts after the scan

Take the distinction seriously and the product becomes easier to understand. Optical character recognition supplies text from an image. A business process also needs to know which document it is looking at, which fields matter and what should happen when information conflicts. Infrrd’s platform includes preprocessing, classification, extraction, correction queues, reporting and API integration. Those last, rather unglamorous functions help explain why document automation is an enterprise software business.

For a buyer, the output has to travel somewhere. An invoice’s values may need to reach an accounting system; a loan package may need to reach an auditor with discrepancies already identified. Returning a tidy table is useful. Returning a tidy table that someone must compare with the original file all afternoon is a different proposition. The relevant unit of progress is the completed piece of work.

02 The mortgage expert was part of the product

In August 2024, mortgage compliance company Asurity and Infrrd announced MortgageCheckAI. Asurity supplied mortgage processing and auditing knowledge. Infrrd supplied the document technology. Together they embedded loan auditing procedures and practical expertise into a product for pre-fund and post-close reviews. Asurity also used it in its own operations. The partnership’s announcement claimed a reduction of at least 50% in manual loan-document review.

The backstory is more revealing than the percentage. Infrrd’s account of Asurity’s earlier vendor evaluations describes rigid templates, continuing manual effort and customization costs. Older loans and diverse document types made the problem harder. A market slowdown added budget pressure. In that account, the deciding attraction was Infrrd’s willingness to listen and co-create. The partner’s own announcement confirms the substance of that collaboration.

There is a useful product lesson here. Knowing where an amount appears on a page and knowing how a mortgage reviewer should evaluate it are separate kinds of expertise. Bringing the reviewer’s knowledge into development gives the extraction engine a purpose. A subsequent partnership with mortgage auditing specialist UHS America, announced in November 2024, reinforced that industry focus.

03 A policy is not merely a page

Insurance presents the same difficulty in another costume. On Infrrd’s website, State National executive David Crawford describes dealing with 2,100 insurance companies, each with versions of its own forms. At that point, maintaining templates becomes a job of its own. The awkward document is no longer an exception; variation is the normal operating condition.

Infrrd’s iTrackPro insurance product splits scanned packages, classifies policy documents, extracts information and routes work according to staff expertise. It offers coverage checks and reporting against service-level agreements. The distinction matters: a declaration, a cancellation and a reinstatement can all contain readable words while requiring different actions. The product is designed around the policy’s meaning within a workflow.

This is also where broad claims about “accuracy” become less illuminating. Field extraction, policy verification and the share of documents needing review are different measurements. A buyer who treats them as interchangeable can purchase an impressive number and inherit an unimpressive queue.

04 Forty percent is a useful number

One Infrrd mortgage case study reports more than 150 million documents processed, a 60% reduction in processing cost per loan and a 40% no-touch processing rate. The customer is unnamed. These are the vendor’s reported results for that engagement, rather than a promise for the next customer.

The 40% deserves attention precisely because it is modest enough to be informative. It leaves room for documents that require some intervention, without telling us exactly how much. It also appears beside a substantial cost reduction. The reasonable inference is that useful automation need not remove every person from every file. Eliminating a portion of repetitive work can matter economically. Establishing how much it matters requires measuring the remainder.

Infrrd’s wider work includes logistics documents and engineering diagrams. Its engineering case study describes a water-management business handling dense construction plans and civil diagrams. Here the data includes visual components and configurations. The challenge extends beyond paragraphs into the relationships between marks on a drawing.

05 The price of getting the work finished

Infrrd’s commercial menu reflects different appetites for review. Enterprise customers can request the platform, human-in-the-loop processing or no-touch processing. The pricing page advertises performance-based pricing for the latter managed offerings, alongside accuracy and turnaround commitments. Human review is explicitly available, rather than wished away by a marketing adjective.

For an operations team, the sensible comparison is cost per accepted output. Include integration, rule configuration, correction time and ongoing review in that calculation. A cheap extraction that leaves expensive checking behind can be a poor bargain. Conversely, paying for reviewed output may suit a team that needs predictable delivery more than it needs to operate the review queue itself.

There is another entrance now. IDPForge offers developers self-service, pay-as-you-go document automation, with 15,000 free ForgeCoins advertised on the pricing page. In its September 2026 update, Infrrd highlighted that route beyond enterprise-only engagements. Enterprise purchasing remains a conversation about the actual workload and service requirements.

06 A decade of teaching software the exceptions

Founder and CEO Amit Jnagal established Infrrd in 2016, according to the company’s timeline and investor Riverside Acceleration Capital. Infrrd opened a dedicated AI research lab in 2018. In August 2024, Riverside announced its investment and named customers including PwC, Sedgwick, Unum, Rocket Mortgage and State National. Its investment account singled out mortgage origination and auditing as an area of strong product-market fit.

The newest mortgage ambition is Ally, an agentic auditor that coordinates sequences of audit tasks. The product page describes extracting mortgage data and carrying out checks, rather than stopping at extraction. That is a larger operational claim: success depends on whether the sequence finishes usefully and whether exceptions reach the right person.

“Our research exists for one purpose: to solve the last-mile challenges that traditional automation leaves behind.”Amit Jnagal / November 2025

Infrrd appeared in HousingWire’s 2026 Tech100 Mortgage list and announced a second consecutive Gartner Magic Quadrant Leader placement that September. Its stated workplace values include innovation, Team First and Value Over Gloss. Meanwhile, the tenth-anniversary celebration introduced Mattie, a squirrel mascot. Enterprise document software has found room for a woodland colleague.

07 Borrow the test, not the promise

Infrrd operates among document automation vendors such as ABBYY, Hyperscience, Tungsten Automation and Rossum. Teams can also assemble workflows around extraction components such as Amazon Textract. Infrrd’s emphasis is on combining document understanding with industry rules, review options and downstream work. Whether that combination wins depends on the buyer’s documents and operating model.

The transferable idea is a better audition. Give the system a representative mix, including the difficult scans, unusual layouts and conflicting documents. Count corrections. Track turnaround. Check whether the receiving system can use the output. Define which exceptions need a reviewer and who owns that queue. A successful demonstration on tidy files tells you little about an untidy working week.

Automation will disappoint when rules are unclear, integrations are unfinished or review responsibilities are treated as somebody else’s problem. Infrrd’s story makes those dependencies visible. A readable document is the beginning. The interesting business is what happens next.