How Enterprises are Leveraging Intelligent Capture, Machine Learning and Artificial Intelligence
In spite of efforts to digitize, many firms still have high volumes of paper that causing operational headaches, potential legal risk for the enterprise, and now even physical risk for employees who process paper. Intelligent tools are needed to drive digital transformation, and to capture documents at scale in situations where paper can't yet be eliminated.
We recently hosted a virtual gathering of practitioners from our clientele and network to discuss the state of intelligent capture and best practices that they’ve developed within their capture programs. In this article we present highlights from the virtual roundtable.
Good Use Cases for Intelligent Capture
There are three clear use cases for intelligent capture:
- Where you need to improve the accuracy of recognition across all document types, both easy and difficult (the next generation of capture engines raises all boats)
- Where you have historically intractable problems - like handprint, or multilingual, or document classification on 2 similar (but different) types of letters.
- Where you add intelligent capture to RPA (robotic process automation), converting images into usable text data that RPA can process.
Foundational Capabilities for Intelligent Capture
So if you have a use-case for intelligent capture, what are the foundational capabilities that should be in place to use it?
Practitioners agreed that it’s critical to understand the different capture capabilities among the various providers. Once the set of potential vendors has been narrowed down, they strongly recommend starting with a proof-of-concept (POC) project.
Practitioners also emphasized that it’s important to have a secure sandbox in which to run samples during testing. Opinions on what makes a good sample size varied, in part because the sample size can depend on the type of project.
One good practice shared by a participant is to implement protocols and procedures that guide staff on how to collect and create the sample set. Another tip shared was to watch out for potential issues getting approval for sample sets – don’t assume approval will be automatic.
Measuring Intelligent Capture Success
Practitioners report a variety of success metrics that they use to get projects funded, including:
- Accuracy – particularly for unstructured content
- Speed – reducing processing time, and scaling that across the organizations
- Cost – achieving cost reductions by reducing errors and minimizing staff touch points
- Customer service – delivering products/service more quickly to the customer, which is made possible by faster, more accurate capture
Poor Image Quality – a Roadblock?
Are poor image quality issues bound to make intelligent capture a non-starter?
The participants discussed problems they have experienced due to poor image quality, but it has not been as huge a barrier as initially thought. One reason is that when there are fewer hands touching the documents, there are fewer issues with image quality to begin with.
Common image issues include:
- Small image size/low-resolution
- Low-quality image scans
- Faxes (inherently low quality)
Fax images continue to be problematic, so much so that practitioners avoid them whenever possible. One practitioner recently implemented document upload, reducing dependency on faxes significantly.
Structured vs Unstructured Document Capture - What's Feasible?
There's a spectrum of structure across document types.
- Structured documents have predictable layout and content.
- Semi-structured documents have partially predictable layout and predictable content.
- Unstructured documents don't have predictable layout and may not have predictable content.
Next-generation capture solutions – particularly those that incorporate Artificial Intelligence or Machine Learning – are doing better in dealing with unstructured and semi-structured documents. For example, one participant reported that his firm is using an AI engine to uncover customer complaints and then route them to whomever can address it.
Another practitioner shared that close to 75 percent of documents processed today by his firm – which typically come from third-party sources – are unstructured, and they are relying heavily on Machine Learning to address it. As part of their program, they have developed standards for how unstructured content comes into the capture system, the way it exits, and the way it goes into the automation system.
Expect to See Continuing Innovation in a Fragmented Market
The rountable participants commented that they expect to see continuing innovation in this category, including:
- A “unicorn” solution that combines intelligent capture and RPA, adds Artificial Intelligence and Machine Learning to both, and then orchestrates and manages them all with business process management.
- An efficient combination of modules ( e.g. OCR, computer vision, natural language processing, etc.) that is managed and configured by AI so that AI decides which modules are going to be used where, instead of staff having to do that.
- Advances in how the next-generation solution providers handle messy documents.
The practitioners also expect continued fragmentation in the solution market, and do not predict significant consolidation in the near term. And while they don’t expect any one vendor to become immediately dominant, they have noticed that there seem to be a lot of creative integrations among solutions from different vendors, and that integrations are a lot simpler than they used to be.
The roundtable practitioners are seeing results with intelligent capture, and expect to see significantly higher performance as innovation continues. If you’re in the early stages of implementing intelligent capture, or looking to revamp your program, reach out and let us know. We’ll invite you to the conversation.