InferIQ takes the paperwork out of lending

The trouble could begin with a few extra characters. In a published Cync case study, the software used to extract accounts payable and receivable data needed its templates adjusted even for small changes in a field’s length. A borrower sent a slightly different document. Someone had to repair the instructions. Before a lender could consider the business, the software needed help understanding the paperwork.

The template was the bottleneck

Cync makes software for commercial finance, including asset-based lending. Its customers need information about what borrowers owe and what customers owe those borrowers. Those accounts payable and receivable statements feed a Borrowing Base Certificate, a document used in evaluating borrowing eligibility.

The small businesses producing those statements did not follow one format. The old approach required locating and labeling content for extraction. Idexcel’s case study says a new file type could take around 24 hours of template creation. Routine variation had become an onboarding problem.

The pressure to change came from the workflow itself. Each new layout demanded attention before the business data could move. As the case study describes it, the constraint was maintaining extraction instructions across varied borrower statements. Teaching a model to handle new patterns addressed that recurring work.

InferIQ’s response combined Amazon Textract with machine learning tailored to recognize unfamiliar patterns. The pipeline classified input, removed noise, extracted tables and reshaped the results for the receiving system. A reviewer could validate the extraction and feed corrections back. The destination mattered as much as the document.

The number needs a destination

Financial spreading adds another problem: vocabulary. A balance sheet can be legible and still require an expert to decide which standard accounting category a borrower’s unusual label belongs in. Copying the amount accurately leaves that interpretation unfinished.

In its financial-statement work with Cync, InferIQ extracts text and uses a custom model with natural language processing to map terms to standard accounting labels. Users can correct the mapping, and their feedback helps improve the model. The application then calculates the relevant ratios.

That division of labor explains the product’s place in the market. General-purpose OCR can supply text. A financial workflow needs recognizable categories and output that the lending application accepts. InferIQ packages those steps around the extraction, with financial expertise built into the mapping process.

“Automation is a must for timely decision-making, and InferIQ delivered.”

Customer statement in Idexcel’s Cync financial-spreading case study

The vendor reports an 80% reduction in time spent mapping statements and extracting financial data in that engagement. It also says the solution added extraction from scans. These are results for the described implementation, useful for understanding the task being automated rather than predicting every lender’s savings.

Three documents, three different jobs

The product range follows the material arriving in a credit file. Financial statements need spreading and ratio analysis. Tax returns need fields and tables extracted so declared income can enter a verification workflow. Bank statements need transactions organized so an analyst can examine money coming in and going out.

The tax-processing case describes image preparation, extraction of form fields, tables and checkboxes, then export into CSV for downstream analysis. Idexcel reports about ten minutes saved per file. The useful artifact is a dataset that another system can consume, with less manual transcription along the way.

The Bank Statement Extractor, or BSE, gives users a portal for uploading PDFs or images and receiving tabulated dates, descriptions, amounts, balances and account numbers. Its May 2025 launch announcement describes support across US banks and credit unions. Users can visually select data the extraction missed.

InferIQ’s current presentation extends that transaction work into cash flow analysis and fraud signals. It also promotes banking insights and product recommendations. The common thread is organizing financial information for a specific next action: checking income, assessing a borrower or understanding an account.

A product with a services company behind it

InferIQ acquired its current name in July 2023, when Idexcel renamed AIDE, its Automated Intelligent Data Extractor. That history puts the offering in context: a document-processing product developed inside a technology services business, with integration work close at hand.

Idexcel announced the platform’s AWS Marketplace availability in November 2023. Its marketplace description includes classification, API connections and redaction of sensitive information before material is passed to external language models. This is relevant to financial teams whose documents contain personal and confidential information. Handling that information is part of the workflow.

The redaction case supplies a second view of the surrounding work. A financial-services client needed to handle more than 1,000 files at a time. Idexcel built a custom redaction layer over Textract, with output routed onward through an API or delivered by email. The task extended beyond recognizing sensitive text to moving protected documents through an existing operation.

The purchasing model has two public faces. The broader marketplace listing invites a custom private offer and identifies professional-services delivery. The May 2025 BSE announcement describes pay-as-you-go access. A buyer should put document charges, setup, integration and reviewer time into the same budget before calculating a saving.

Count the corrections

For a team considering InferIQ, the case studies suggest a practical experiment. Take representative documents, including unfamiliar layouts and scans. Follow them through extraction, accounting mapping, review and delivery. Count the minutes of human correction and check whether the receiving system accepts the result.

That is an editorial inference from the workflow, and a useful one to copy. Template trouble begins where documents vary. Mapping trouble begins where meanings vary. A pilot needs both. The presence of a correction loop makes reviewer skill and availability part of the operating model, too.

Price the review work explicitly. An extraction charge applies to processing; a correction consumes a person’s time. A useful comparison follows the same batch through the existing process and the proposed one, recording both. Faster machine output helps only if the complete batch reaches a usable state sooner, with acceptable accuracy. For the buyer, that is the meaningful unit of comparison.

The fit is strongest where staff repeatedly translate financial documents into lending systems. Where information already arrives as clean, structured data, the extraction stage has less to contribute. Where corrections consume the time supposedly saved, the economics need another look. InferIQ’s proposition can be tested at that ordinary, consequential point: when the file leaves the queue and becomes usable.