Machine Learning OCR

The #1 Question People Ask About OCR

by Mark Clark, on August 25, 2020 7:00:00 AM PDT

A Gartner analyst takes 20 calls a month on OCR and Intelligent Document Processing from enterprise buyers looking to make their next moves.

We asked him what they’re looking for in a data extraction solution.

And he said, “solution accuracy.” 

Here’s why: Vendors aren’t generally willing to provide this most critical metric. Instead, they say, “it depends.” They don’t want to commit, which is unsatisfactory to the enterprises who have real money on the line. 

Accuracy Matters

Smart buyers don’t want to get involved with uncertainty. They want to know how much they can automate, and what kind of manual effort they’ll have to supply to ensure their data is extracted meaningfully.

Let’s see what the business impact of accuracy might look like with a simple example.

A firm processes 100 documents a month.  The processing costs are as follows:

MANUAL:  If a document cannot be processed by automation (complex, unstructured documents can present an extraction problem), then a skilled person will have to do the extraction without automation. This costs about $15.00 per document.

AUTOMATION: Automating data extraction with 100% accuracy means no human costs. It’s called Straight Through Processing (STP), and people don’t need to touch the document.  Let’s call the automation cost $1.50 per document. 

CORRECTION: If automation does not achieve 100% accuracy, and people have to make the corrections manually, that adds another $3.00 per document. Automation makes corrections easier compared to a fully manual process. The automation system will identify low confidence extraction fields and guide a user through a correction workflow. 

Let’s look at an example of how higher accuracy can cut your costs.

The chart below shows the firm’s cost to process 100 documents for various accuracy levels. The baseline case has no automation and it costs $1,500 to process the 100 docs.

We know that costs decrease as automation accuracy increases.  In our example, the chart shows it costs $165 to process 100 documents with 90% automation accuracy.

Cost Chart - Automation Accuracy Rates

If the solution can achieve 100% automation then we will have the lowest cost per unit of work and it will cost $150 to process 100 documents. A 10x savings.  

So, accuracy tells us:

  • The cost to process a document 
  • How many staff members are required to support the process
  • What the corrections' infrastructure needs to be
  • Process SLAs such as cost and cycle time

The example illustrates how automation accuracy reduces processing costs and shows why accuracy is such an important metric to track.  

So where can we find an Intelligent Document Processing solution that delivers high accuracy? 

Infrrd Does Accuracy 

Here is a real-world example where the enterprise’s challenge was that the source documents were so complex that only the cognitive power of a human could extract the needed data.  

An insurance firm was spending $5.8m on manual extraction each year because its insurance claims documents were too complex to automate. The firm was looking for a solution that could handle the documents’ complexity and reduce its manual load but kept coming up short. 

Infrrd was able to provide an automated solution delivering 60% automation at the start of the project, with an upward accuracy trajectory as the solution learned and improved over time. The client was able to reduce staff size by 50%, which generated $2.9m in annual savings.

Here Infrrd made the “impossible” possible for this customer by automating a complex human task.

Process and Performance 

There are two main factors that drive accuracy: the characteristics of the documents being processed, and the performance of the automation system.

Some documents have more complex layouts and more overall variation. The simpler the document and the higher the image quality, the higher the accuracy that can be achieved.  

OCR-type solutions are good for simple documents and can achieve good accuracy rates. But OCR accuracy falls apart when documents are complex and when documents change. So to start, it is important to understand your source documents.

Variations Vs Complexity Documents

Solutions such as Infrrd’s IDP offering are capable of handling more complex documents and documents that change and vary. IDP uses AI to manage complexity and uncertainty, which lets it achieve much higher rates of accuracy than OCR can provide.

Throw a complex, unstructured doc at OCR, and your accuracy will suffer. Then the extraction and correction work will fall back on people. 

Throw a complex, unstructured doc at IDP, and high accuracy can be achieved and maintained.

Why Don’t Vendors Commit To Accuracy?

OCR and IDP solutions will make accurate claims that are difficult to understand. That makes it difficult to determine how those claims will impact your business case.

Vendors might claim high accuracy for a specific, well-defined document type, such as a form.  

This vendor says, “up to 95% automation, and greater than 99.5% accuracy.” Saying “up to” lets the vendor showcase the best results, and leaves a lot of wiggle room.

Infrrd guarantees 100% accuracy.

Or vendors tell you automation accuracy might just be too difficult. 

Infrrd says no document is too complex.

One OCR vendor says “garbage in, garbage out”, meaning that the source documents have to be well-defined to achieve high accuracy. But the source documents are not always in your control.  

Infrrd is ready for anything.

Another OCR vendor says “If the SOW says 80% field accuracy and you have 76.5% field accuracy, should you be expending effort for that last 3.5%? Probably not. The effort to remove the last 3.5% probably produces diminishing returns, while effort spent in productivity gives far greater gains.”

Infrrd’s platform makes it easy to reach your accuracy goals.

Devoid of straight answers from vendors, enterprises turn to Gartner and ask “what will the accuracy be?” 

Wouldn’t it be much nicer to know your accuracy from the get-go?

The Buck Stops Here

Accuracy uncertainty then leads to uncertainty about the business case.  

You assume the uncertainty for unclear automation accuracy. That means will need to size staffing to account for the low end--not the best case the vendor will pitch--of accuracy rates. 

And you’ll have to keep extra staff around to handle fluctuations in processing volume. Otherwise, you will miss your processing SLAs.

Infrrd takes a different approach. We guarantee your accuracy and assume the performance risk, so you don’t have to.

Understanding Accuracy

You need quality data to feed your business process, so 100% accuracy would be nice.

But understanding accuracy is tricky. To really understand how accuracy will impact your business, ask these additional questions:

  1. Will the solution work on my specific documents at the accuracy I require?
  2. Where and how is accuracy measured? 
  3. What will be the effort to close the accuracy gap?
  4. What will be the effort to maintain accuracy levels?
  5. How will the solution improve performance over time, and what will that cost?
  6. What if my documents change?
  7. What if I want to add a new use case or document?

Vendor claims about accuracy can get complicated -- proceed with caution! 

Accuracy is a Great Question

Accuracy is a key driver of business performance, and Gartner’s hot topic because of the impact it has on enterprises. 

Accuracy tells you how you need to design your process, what SLAs can be offered, and how automation can impact the overall business process.

But uncertainty about accuracy creates risks that make it difficult to design a lean process without the extra cost needed for manual buffers. 

What’s the Answer?

When it comes to accuracy, enterprises want to be sure. 

And nothing’s more certain than a guarantee. 

Guarantee accuracy performance and enterprises can rest easy, knowing that they can focus on business impact, without worrying about what accuracy they can achieve. 

High-quality data to fuel your business processes -- that is Infrrd’s answer.

Topics:Intelligent AutomationAI ReadinessBusiness Insights

About this blog

AI can be a game-changer, but only if you know how to play the game. This blog is a practical guide to turning AI into real business value. Learn how to:

  • Make sense of complex documents and images.
  • Extract the data you need to drive intelligent process automation.
  • Apply AI to gain insights and knowledge from your business documents.

Say Goodbye to OCR Blues & Burnout

Subscribe to Updates