100% Accuracy and the (Hard) Choice Between Automation and Manual Document Processing
by Sujith Parakkunnath, on August 3, 2022 9:00:00 AM PDT
During an interaction with an industry leader about his experience with IDP, said he had recently decided to automate his document extraction process because of the challenges faced with data inaccuracy and cost. He said their IDP platform currently provided 90 to 95% accuracy. He added that a manual team handled the remaining 5% to 10% of corrections.
What he said next inspired this blog post.
He said 95% accuracy does not look like 95% accuracy and his manual team had to put a lot of effort into corrections. This is a familiar case with many businesses that have chosen automation with high expectations. What exactly is the problem in this scenario? Let me try to shed some light.
One of the popular perspectives or oversights many of us have is that 95% accuracy means 95% correct values. This is not true. 95% accuracy does not translate to 95% correct values. It means that there is a 95% probability that a value is correct, or in other words, there is a 5% chance that the value is completely wrong.
95% is not the same as 100%. Unfortunately, even with 95% accuracy, customers must still manually verify all their documents to ensure the data is 100% accurate.
Your Business Outcomes Depend on 100% Accurate Data
You are not in the business of extracting data from documents… we are. Your business needs this data from the documents to make lending decisions, settle a claim, or approve expenses. When you are forced to keep a data extraction team in-house or contract with a partner, you run into new challenges that have nothing to do with your core business. Challenges such as:
• Limited Work Hours - Data processing teams log out after 8 hours of work, whereas servers do not need to log out. This means you can get a lot of data processing done while your team is sleeping. When your team starts their work the next day, they can focus on reviewing the work of the algorithms, instead of processing data from scratch. And a good IDP system will learn from each of these corrections so that your team does not have to review the same mistakes every time.
• Lack of Organization - When data processing teams work with documents, it takes massive effort to organize their work to make processes more efficient. IDP systems can help organize this work automatically. For instance, if your data processing team processes mortgage documents like W2 forms, bank statements, and paystubs, you can route them intelligently to make sure that each data processing executive sees all the W2 forms first and then switches to bank statements. There is a 10-20% efficiency gain to be had from organizing work items in this way.
• Scalability Challenges - Hiring, training, and retaining employees take a lot of work. When your business volumes pick up, it generally takes a lot of time to increase the size of your team, train the new staff and wait for them to become efficient. Scaling an IDP platform simply requires more infrastructure which is relatively inexpensive and rapidly available. IDP investments can help you scale your operations without having to proportionally increase your team size.
• Processing Inefficiencies - When work is being done manually, there are often unneeded processes - automation can help to remove some of these completely. One example is the document scanning process. When multiple documents are scanned in a single PDF file for the data processing team, a blank page needs to be inserted where one document ends and another one starts. This time-consuming process can be completely eliminated thanks to IDP platforms.
To ensure data accuracy, you suddenly find yourself training and managing people who are not core to your business. Infrrd takes all of these problems out of our customers’ hands and manages human-in-the-loop (HITL) until the system matures and offers you 100% accuracy. This is a game-changer for customers because you no longer have to worry about the pain, effort, and cost to recruit, train, and manage a corrections team.
Perceptions about 100% Accuracy
One of the perceptions about 100% accuracy is that it is an expensive proposition. This perception is True as well as False. If you get human-in-the-loop verification from a company whose primary offering is people then it turns out to be a lot more expensive than having an in-house team. But if you work with someone like Infrrd, a software technology company, then the human effort is substantially reduced due to the extensive use of AI. This is much more cost effective than managing an in-house team or working with a totally people-based organization.
But how can you tell what kind of partner you are talking to? Every vendor makes similar claims during the sales process. There are three ways to figure out if you are talking to a people organization or a software technology company.
1. Demo - A people-primary company will likely not have a solution that can extract data from documents but rather use technology for routing documents to people. Most of these companies give you an email address to send a document to and get the results back. On one of the websites of people-primary companies that I looked at, it mentioned the following on how long it takes to process a document: “It may be minutes or several hours. It really depends on a number of factors including the load factor on our servers.”
2. SLA - People-primary operations have turnaround times in hours. For software technology companies, at least 30-40% of documents should come back in a matter of seconds.
3. Cost - People cost a lot more than servers. If you are dealing with a people-primary vendor, then the price will reflect the cost of people.
As we discussed, one consideration in choosing 100% accuracy may be the additional cost involved at the onset, but the following benefits override that cost in no time:
• Right focus: No need to manage a data team. You can focus on your critical and expert areas while your IDP partner takes care to deliver results with 100% accuracy.
• Better output: You can expect to get better results or output as corrections are handled by experts.
• Compensation: You may get rewarded if the processing results are not absolutely 100%. Ensure that this clause is included in your contract. Even though this cannot be considered a benefit, this is an indemnity when things don’t work as planned or expected.
• No technology risks: Now, this is very important. When your IDP partner takes care of corrections, it does not matter whether the system delivers automation results with 60% or 90% accuracy because you finally get 100%. However, it is still important you select an efficient system, a good AI-based machine learning system that has the potential to exponentially mature to achieve real-time automated results with 100% accuracy.
• Fast deployment:If you choose a good IDP system that provides high accuracy out-of-the-box and relies on IDP experts to handle corrections, the training or maturity time will be minimal, enabling you to go live in a very short time span.
100% accuracy is a game-changer in the IDP industry because it transfers the huge pain or responsibility of manual corrections from your end to your IDP partner. Infrrd delivers human-verified documents at a fraction of the cost because our AI does most of the heavy lifting. You can learn more about Infrrd’s 100% accuracy offering here.