I have a friend who is the accounts payable manager for a medium size logistics company. Last year she told me that for the last few years she spent about 2 hours every morning entering data into a spreadsheet for all the invoices her company received. She said it was hard work but wasn’t difficult to do. She is getting paid to do copy work as a very expensive human.
That conversation stuck with me. Most people have no idea of the huge efforts that Accounts Payable teams go to on a daily basis to process documents for most organizations. Here is a video that graphically shows just how long it takes to manually process individual documents, on a daily basis, day in and day out. Yes, it’s not that hard to manually enter a few fields of data, but so dull and time consuming that it becomes endless.
The actual cost nobody talks about
Even if manual data entry were fast, there is the more serious issue of the damage that manual data entry does to organizations over time. A small leak in a roof can cause a lot of problems over time. A single transposed digit or a single error in a vendor code can result in an invoice being filed under the wrong project. Three months later, that incorrectly filed invoice surfaces during an audit. Everyone wants to avoid situations like this.
Studies have found that the rate of data entry errors averages about 1% per manually inputted piece of information. For example, one study found that errors occurred in 1% of manually keyed-in information in accounts payable processes. The problem for companies with large volumes of documents processed in accounts payable is that this translates to hundreds of mistakes per month for companies processing tens of thousands of documents.
The biggest cost however is the huge amount of potential being wasted from the very experienced people in the company. Instead of managing a very complex accounting process, they could be focusing on the intricacies of a process and not on counting and recounting information in spreadsheets. This would be like having a very experienced carpenter but instead of building houses, they spend their time sorting through all the different types of nails and classifying them into different sizes. Yes, they would be organized but a huge amount of potential would have been wasted.
What intelligent capture actually does
This is also where the old world of OCR meets the new world of intelligent document capture. Old OCR, or optical character recognition, was once a great technology but for the last 15 years or so it has gone into a type of hibernation. The technology relies on simple pixel based matching of characters on printed or written pages. Such systems are notorious for incorrectly recognizing an “8” as a “B”. Intelligent document capture, uses advanced machine learning within a framework that understands documents within the context in which they are used.
There are also additional post capture processing steps that documents go through after they have been captured by Intelligent Automation (IA) solutions. These automation processing steps include identifying what type of document has been received, extracting key data fields, and then matching the newly captured document to existing data already stored within the organization’s business systems. In situations where all of the processing of a document occurs after it has been captured Intelligent Capture greatly reduces the potential for human error prior to the document being distributed via email to the appropriate staff member for review.
- Invoices get matched to purchase orders and flagged if something doesn’t line up
- Forms are sorted by type and routed to the correct department without manual triage
- Business records are indexed automatically, making search and retrieval dramatically faster
- Exceptions — the genuinely weird edge cases — get escalated to a human rather than silently processed wrong
Again, a system which can extract information from documents is only as good as the quality of the information it was provided with in the first place. Good AI systems will be able to indicate when it is unsure or not confident of having identified correct information. This is actually a very trustworthy feature of intelligent capture systems and is far more important than being able to say with absolute certainty that all information has been correctly extracted from all documents, when it is clear that this is unlikely to be the case.
A quick look at where the time actually goes
| Task | Manual process | AI-assisted process |
| Invoice data extraction | 2-4 minutes per document | Seconds, automated |
| Document classification | Manual sorting, error-prone | Automatic, consistent |
| Routing to correct team | Depends on who opens it first | Rule-based, immediate |
| Error correction | Discovered late, fixed manually | Flagged at point of capture |
At 400 documents per day for a team of 10 employees processing paperwork, that would amount to 20 hours per month of time saved. Time that could be more productively spent elsewhere.
This isn’t just for big companies
A lot of people believe that intelligent capture solutions are the domain of large enterprises with deep IT budgets and teams of experts, however recently intelligent capture solutions have evolved and are now available to businesses of all sizes. Many cloud based document management solutions now have intelligent capture as core functionality of the solution and are no longer an expensive add-on developed by a 3rd party vendor.
It can also be of great value to medium sized businesses with a few skilled staff using large numbers of similar manual paper based document processing methods. In these businesses, the time currently being spent, processing large numbers of similar documents by hand, can be used to recover the cost of a cloud-based paper document management system that contains intelligent capture functions as standard.
Calculating out the hours spent manually processing all of the documents as if they all were processed in the same amount of time as the first 100 documents to arrive at a number of hours spent processing all the documents is a very different number than the estimated hours spent processing all of the documents.
Evaluate Requirements
- Does it handle your specific document types out of the box, or does it require extensive training before it’s useful?
- How does it handle exceptions and low-confidence extractions?
- Can it integrate with the accounting or ERP system you already use?
- Is the learning ongoing — meaning does it genuinely improve the more documents it processes?
Each vendor will claim that their product is the best for your business. That a vendor is willing to have a web demo is not in itself evidence that the solution will work for you. These questions can generally be answered in a web demo but it is only by really digging into the details of a particular solution that you will really know if it is the right one for you. And of course there are more than a few vendors who are only looking to make a quick buck and will thus do everything in their power to downplay any features or limitations that may indicate to you that their product is not right for you. So you need to be very cautious of vendors who:
(One more thing – check that the system can integrate with other systems that you use).
The part my friend never expected
We no longer have to manually enter in all of the invoices from the night before for a 2 hour morning ritual of data entry. All of the documentation for the invoices are captured and accurately accounted for by the system. I have been checking the capture of the documentation and the accounting for the documentation for the first few weeks to ensure everything is being entered in correctly and that the system does not have some sort of catastrophic failure. I haven’t found any issues, and I have found plenty of time in my days to do some really valuable work with all of the time that I have been saved. Managing relationships with vendors, forecasting cash flow, etc. This work really compounds in value as opposed to just maintaining the status quo.
In summary, technology should be used to return wasted hours of work removing tasks from humans that use their skills, freeing up time to add more value each hour as they tackle work that compounds in value.
