Home Bots & Business Blog: How to automate backoffice tasks with OCR and machine learning

Blog: How to automate backoffice tasks with OCR and machine learning

by Guest

Over the past 20 years, back office outsourcing has been the go-to solution for large enterprises looking to save costs on repetitive, labour-intensive work. Next to cost reductions, a big benefit of moving the back office to low wage countries used to be the availability of a large labour force that allowed easy scaling. Countries like the Philippines, Indonesia and India used to be very popular countries to go to.

As with everything, technology is now catching up with traditional ways of back office outsourcing. The new norm is not to simply outsource, but to actually automate back office tasks using modern technologies like RPA, OCR and machine learning.

Automating allows for even higher cost reductions, faster turnaround times, lower organisational complexity and fewer compliance risks. In this blog, we will explain how you can automate your back office and what the benefits of back office automation are.

What is back office automation?

To answer that question, let’s first determine what back office tasks are. Back office tasks are all administrative tasks within an organisation. These tasks usually arise from legal (contract processing), customer onboarding (KYC), compliance, HR (employee onboarding) or financial procedures (accounts payable). Think about tasks like data entry, document sorting, approvals and fraud or compliance checks.

A back office task every company will know is invoice processing, usually done by the finance department. Of course, at low volumes it’s totally fine to have a finance employee manually processing 50 invoices per month. But what happens if you need to process and approve a million invoices per month? Will you scale your finance team from 1 to 100 people, or will you try to automate this task?

This is where back office automation comes in. Back office automation is applying software to automate repetitive tasks that administrative employees are now performing manually.

Why is back office automation a good alternative to back office outsourcing?

Even though back office outsourcing can deliver big cost- and scalability benefits, it also has a few downsides. One of the major drawbacks is compliance risk. Offshoring back office data to countries like India or the Philippines comes with (privacy) risks. Are the employees trained well? Do you have permission from your customers and local authorities to share their data (GDPR)? How do you prevent a data breach? These are some of the challenges you might face while outsourcing.

The cost of processing back office work in a low wage country is of course lower than in Europe or the US, but what if it could be even cheaper? And from a scalability perspective you could theoretically try to infinitely scale an outsourced team of people, but is this manageable in reality? And can you speed up the turnaround time towards real time processing by training people? Probably not.

So to solve these issues you have to look for an alternative to back office outsourcing: back office automation. In the next paragraph we will highlight why you don’t have these issues with a back office automation solution and show that it’s a good alternative to back office outsourcing.

What are the benefits of back office automation?

With automation you can take the human factor and foreign regulations out of the equation. This has benefits from the perspective of cost, turnaround time, organization complexity, scalability and control:

  • 100% control: If you choose an automated solution over an outsourcing solution, you ensure total control. Automated solutions can be deployed in your own country, under familiar law and within your own IT infrastructure. Therefore, you do not depend on third parties and their local regulations for your business needs and keep 100% control.
  • Cost savings: While outsourcing might already reduce your cost over local processing, automation can get you an even bigger cost reduction. We see cost reductions in the range of 50 to 70% on average depending on the task complexity, when compared to outsourcing.
  • Realtime turnaround times: While outsourcing might speed up your processing time, this is mostly due to a bigger workforce. This bigger workforce can simply get more work done within the same timeframe, but does not really do a single task faster. But what if you are looking to reduce your turnaround time to nearly real time? Then, outsourcing is not the way to go. An automated solution can process most administrative tasks within a few seconds, so nearly in real time. Something a large outsourcing crew could never achieve.
  • Infinitely scaleable: While you could theoretically scale an outsourced back office from 100 to 1000 or even 100.000 back office workers, the operational complexity that arises with scale makes it very hard in reality. An automated solution, on the other hand, can automatically scale its servers on demand without any additional recruitment, office space, contracts and so on.
  • No compliance risks: Outsourcing your privacy-sensitive data to a low-wage country may remove your business from compliance to, for example, GDPR regulations. When extending these tasks outside of your control, you can never be sure if your data is handled securely. With back office automation, you are always in the driver’s seat. Servers on which data is processed can be placed locally and data that you send to those servers are never saved, but simply returned to you. This is a form of safety and security that an outsourcing company could never guarantee.

How does it work?

OCR and machine learning are at the base of many automation solutions. With the help of OCR, it is possible to identify the text in documents and images. As soon as you have the text, you can start gaining an understanding. By using machine learning, it’s possible to identify data points that are interesting within text and you can mimic human behaviour by learning from previous examples.

Through OCR, the pixels that contain text are identified and extracted into digital text. The act of manual data copying is replaced directly with OCR. With an accuracy of more than 95%, all text is extracted, whereas manual data copying would have a significantly lower accuracy and costs much more time.

Machine learning grows in effectiveness when fed with more and more examples. An AI is trained with numerous examples of documents and specified data sets so that it can automatically localize and identify specific text on a specific position on the document. Over time, a machine learning model can therefore only get better. All data is automatically contextualized and converted to a JSON-format.

Now it’s clear how automation can work, let’s discuss what the process of implementing a solution looks like in four steps:

  • Looking at current procedures: Identifying the steps taken in the current back office workflow is the start of any innovation process. What documents are you processing, who is doing the processing, what are the steps they take at what stage? These are all relevant questions at the start of your journey towards automation with OCR and machine learning. Once you’ve identified the taken steps, you can start determining which steps can be replaced with automation.
  • Gathering data and data annotations: It is important to gather as many and various examples of processed documents as possible. In these example documents, specific segments are all annotated with contextual information to identify what each segment contains. The more examples are annotated, the merrier you will be with the outcome. You’ll create an enormous dataset full of valuable information, which is imperative for the training of a model.
  • Training an automated solution: With the aid of a framework such as Yolo, the dataset can be quickly fed to a machine learning model, so that it is trained with all possible examples. The model learns how to interpret data thanks to the annotation performed in the previous step. With enough training, such a model can perform data contextualization by itself without the need of human involvement. When the model proves accurate and quick enough, it can go live and can be implemented in a solution.
  • Implementing the solution: The trained model is now ready for deployment in a solution. You send a request to the API, the API deploys the trained model to automatically identify the information on a document and quickly and accurately returns the desired data for further processing or application. With this API, you have your automation solution ready to be implemented.

Instead of setting up and implementing an entirely new tool yourself, you may prefer to integrate these technologies into your existing workflows. Many enterprises are already using RPA vendors like AutomationAnywhere, UiPath, BluePrism, Mendix or others to automate certain workflows. From a workflow perspective these solutions are all good. But what you will often see if you look at their OCR and ML capabilities is that their built-in solutions are not adaptable enough to reach a high degree of automation for your niche use case.

Human-in-the-loop automation to handle dropouts

Of course, technology is not always perfect. Even though it evolves very rapidly, you might have certain back office tasks that can not be automated 100% (yet). Luckily, there are solutions for this as well. By combining software with the power of humans you can create a best of both worlds scenario, also called human-in-the-loop automation.

With a human-in-the-loop solution you can automate a large portion of the procedure and then have a human review and complete the work. Meanwhile, the system will learn from any human-made changes and with that improve over time. So while you might start at 70% automation, which is already great, the human-in-the-loop processing will get you as close to 100% automation as possible.

Yeelen Knegtering is CEO & Co-founder at Klippa

Misschien vind je deze berichten ook interessant