Overview

Instabase is an automation platform for unstructured data that can help you process any type of document your business relies on, from IDs to proof of income to business letters.

The core capabilities of Instabase include classification—identifying document type—and extraction—extracting data from documents.

To illustrate the value of Instabase, consider a common use case: mortgage lending. During the underwriting process, prospective borrowers submit packets of documents to confirm their identity and income: driver’s licenses, paystubs, bank statements, tax documents, employer letters, and more. In traditional workflows, these documents are processed by employees, often at high cost. With Instabase, document types can be identified and relevant data can be extracted in just minutes, leading to faster lending decisions and greater organizational efficiency.

How does Instabase work?

Instabase uses deep learning—a form of artificial intelligence—to process documents as a human brain might.

You begin by developing a solution—the Instabase components that address a particular use case. You add sample documents to your solution and annotate them to indicate what information you expect to glean from each document. With its powerful deep learning technology, Instabase can even process poor-quality documents, like a crooked photo of a crumpled receipt, and complex data, like tables and signature fields.

In the mortgage lending use case, for example, your sample documents would include driver’s licenses, paystubs, bank statements, and any other type of record you might encounter during underwriting. Your annotation set indicates what type, or class, of document each record represents, as well as indicating the specific data you expect the solution to extract.

You use your annotation set to train a deep learning model to classify and extract data from similar documents that it hasn’t seen before. You can optionally clean, or refine, extracted data, and validate it. Finally, you create a flow, which is simply a workflow for processing the documents related to your use case. Your flow specifies each step required to transform documents into structured data, from performing optical character recognition and splitting up combined documents, to classifying records and extracting data, to refining and validating data.

When you’re satisfied that your flow performs well, you can export it for use in a production environment and integrate it with other business processes and apps, such as automatically intaking documents for processing, and automatically outputting extracted data to other systems. In production, Instabase flows that fail validation are queued for review by a human reviewer, ensuring a higher degree of accuracy and confidence in real-world applications.

Getting the most value from Instabase

As you develop solutions, you have several additional tools for improving accuracy and efficiency.

Deep learning ecosystem

When you train a deep learning model, you begin with an existing base model. Instabase offers multiple options for base models, from proprietary models to best-in-class models from industry leaders, so you can build your custom document-processing solution based on the model that performs best on your specific documents. The model architecture in the Instabase platform enables continuous upgrades of base models, so they keep pace with deep learning innovations.

Solution accelerators

As you develop solutions, you can turn to the Instabase Marketplace for various shortcuts, or solution accelerators. These resources provide faster time-to-value and foster a shared ecosystem of document understanding.

Solution accelerators include:

  • Marketplace models that are pre-trained on specific document types, enabling you to fast-track model development.

  • Full-fledged solutions that address particular use cases, such as extracting data from driver’s licenses or US W-2 tax forms.

  • Developer packages consisting of low-code and pro-code functions and libraries that support complex document processing tasks, such as converting HTML to text or extracting checkboxes from documents.