Testing accounting compliance

Wiktor Sarota
Is this the future of accounting?
4 min readDec 9, 2016

About inFakt

inFakt has been delivering accounting solutions since 2008. More than 400,000 registered users and 10 million processed documents testify to thesuccess we have achieved on the Polish market.

Our goal has always been to create easy-to-use tools and provide excellent support for accountants. We believe in the importance of both UX/UI and people.

inFakt’s first product was an application for processing invoices. Then, in 2010, we introduced another product, a self-service accounting app. This was followed in 2012 by an advanced service now known as Full Service Accounting, which connects businesses and accountants on one platform where they can work together. All of our products are cross-platform, accessible through web or mobile applications.

Accounting Automation

We’re looking at accounting automation for two reasons.

First of all: While creating the Nationwide Accounting Office, we selected the 100 best accountants out of more than 2,500 candidates. We trained them in our methods of accounting, tax calculation and social insurance liabilities in order to be sure they could provide expert customer support and have thorough technical knowledge of our solutions. This is a costly and time-consuming process but it makes it possible to minimize the risk of making even the smallest mistakes among businesses that use our services.

Also: A conversation with my wife, who uses the Nationwide Accounting Office service, alerted me to a pretty obvious and easy-to-see mistake in the report sent to the national insurance office sent by her accountant.

So we asked ourselves a question: What if, instead of looking for a new accountant, more accountants used automation made possible by machine-learning technology?

Automatic accounting of national insurance payments

We started work with an attempt to automate the processing of payments to the national insurance office. To be sure of its accuracy, accountants still handle this matter in the traditional way, working with the system and supplying the data from the previous twelve months. Their task was to make a decision regarding the accounting of national insurance payments for the coming month.

Comparing results after a month

Document recognition

One hundred percent of documents were recognized.

Difference in decisions made

It turned out that 87% of accounting decisions made by people and machines were the same.

The source of errors

We verified the 13% figure for the differences in decisions manually.

We found that:

  • ⅔ of the mistakes were human error — The accountants that we selected were chosen and trained from 2,500 candidates.
  • ⅓ of the mistakes were due to errors in algorithms that we modified and improved so that they did not make the same mistake again.

Automated accounting costs

The next challenge was accounting beyond that which is necessary for simple confirmation of payments to national insurance or taxes, but all remaining documents and invoices to be processed. We used a similar mechanism for this task.

Accounting effectiveness

The Accounting Automation system was able to process 55% of the documents.

Accounting accuracy

Among the 55% of documents processed, 97.5% of those decisions were accurate.

The source of errors

An analysis of the source of the errors allowed us to identify three types of mistakes and provided us with some surprising results:

  • 75% of the wrong decisions were the result of human accounting errors that caused the system to learn incorrect methods
  • 15% of the mistakes were due to errors in the algorithm
  • 10% were split between the human accountant and the system but did not affect the tax situation of the business

Results

The results of our research gave us some more interesting insights:

  • We took great care in recruiting the right accountants but the automated system still made fewer mistakes
  • Making decisions on 55% of the documents is a very promising start towards full automation and reducing the time needed for repetitive, monotonous work
  • Human accountants are still very necessary for us and our clients. The system works to change their role into one of auditor and teacher for the technology. Accountants also explain the process to clients and are the first point of contact with tax offices and other governmental institutions.

--

--