April 3, 2020

Enterprise process automation made easy

Mike Khanna

Enterprise process automation improves efficiency and slashes costs. Here, we show how unsupervised machine learning can help you automate any business process.

Over the past decade, enterprise process automation has gone from niche concept to mainstream business solution. The early versions used robotic process automation, dumb systems that made it easy to do tasks like bulk mailing, completion of forms, or processing an order. If you want to create a new RPA bot, you need a specialist in RPA and a domain expert to explain exactly what needs to be done. Typically, this takes a couple of months and results in a simple bot that can only operate in isolation. Worse, if you present your bot with data that differs in any way from what it expects, the bot breaks. For instance, an order processing bot could be thrown simply by the address being given in a different format.

RPA 2.0 takes enterprise process automation to a new level by adding artificial intelligence. Now your RPA bots can make intelligent decisions when confronted with unexpected data. For instance, they can learn all the different ways a customer writes their address. The problem is, to create intelligent enterprise automation bots you need to turn to artificial intelligence. This means you now need to combine three different skillsets: a domain expert, a data scientist, and an RPA specialist. So, the process of creating robust intelligent bots now takes months. However, you can speed things up by choosing an approach based on unsupervised machine learning. And, as we will see later, systems like Sonasoft Saibre can cut this time to just a few weeks.

Machine learning refresher

Machine learning is the science of teaching computers to identify patterns in data. It underpins most artificial intelligence applications. We explained it in some detail in another blog post but here is a quick refresher. There are 3 main forms of machine learning.

Supervised learning requires a large amount of data that has been labeled. That is, you have specified what the data is showing. The computer is trying to learn how to identify the data for itself. The classic example here is teaching a computer to recognize if a picture contains a cat by showing it millions of images of cats.

Unsupervised machine learning takes a different approach. Here, you don’t tell the computer what the data shows. Instead, you allow it to learn for itself and try to identify interesting patterns in the data. For instance, you might give it a huge number of addresses from all over the world. From these, it can learn for itself the different address formats used in each country. It might even be able to learn details like which ZIP codes relate to which city.

Unsupervised machine learning is great for identifying clusters
Unsupervised machine learning is great for identifying clusters in data

Reinforcement learning is more about learning by trial and error. The computer is rewarded in some way each time they get things right. This is very similar to the way babies learn skills like walking. It is also how voice assistants like Alexa have been taught to become so good at recognizing different accents. They do this by getting humans to listen to snippets of voice and check whether Alexa correctly understood them.

Enterprise process automation

As we mentioned above, enterprise process automation grew out of robotic process automation (RPA). The aim is to help automate as many business processes as possible, improving your efficiency and freeing up staff. RPA relies on dumb bots to do this. These typically interact in a simple fashion with your existing systems and try to replicate repetitive tasks. The best way to understand how it works is with an example.

Automating employee onboarding

Imagine you are an HR team member responsible for onboarding a new employee. First, you need to add them to the HR system. Then you need to ask the system team to create an email address and any required accounts. The IT team also needs to assign new hardware including a computer/laptop. You have to contact payroll and ensure the employee is added to that system. The new employee’s line manager needs to be told the details of their start date, etc. Finally, the office manager needs to be contacted and asked to assign a desk space.

With RPA, you can automate many of these steps. For instance, when you add the new employee to the HR system, an RPA bot can automatically contact the IT team and ask them to create an email address and assign hardware. It can also email the office manager and line manager. You can even use it to integrate the HR system with payroll. All this works really well until you have a new employee who only works part-time and is home working. Now, you have to go back to a manual process to ensure everything is done correctly.

How RPA 2.0 helps

If you add some intelligence to your RPA bot, it can be taught how to handle different types of employee, without manual intervention. You need to include a machine learning model within your bot. This model needs to be able to understand the different onboarding process for a home worker and an employee based in the office. It could do this using supervised learning or reinforcement learning. But for a case like this, unsupervised machine learning would probably work well.

To teach your bot, you need to show it how several different types of an employee are onboarded. This will allow it to learn that if someone is a homeworker, they don’t need to be assigned office space. However, the IT team needs to be asked to dispatch their equipment to their home address by courier.  

But there are problems even with this. Creating a reliable unsupervised machine learning model isn’t easy. There are many hundreds of different models you can choose from. You need to be able to get the correct data in a usable format, always tricky when you have multiple systems. Finally, you will have to verify that your model is working properly. Typically, even a good data scientist will take several iterations to get this right.

Using a bot factory instead

Fortunately, there is another way to do things. Sonasoft Saibre is an AI bot factory. Saibre makes enterprise process automation as easy as pie. This is the technology that powers Sonasoft Saibre, which adds intelligence to any customer or HR support system. At its heart, Saibre is a universal AI platform that is able to take raw data and create fully functional bots with no human intervention. It is particularly well-suited for problems like automating employee onboarding. Saibre can understand and process almost any data. This means you can feed it things like the employee handbook, details of past onboarding processes, and even employee contracts. From these, it will be able to establish exactly what the onboarding process should look like.

Critically, Saibre does more than just unsupervised machine learning. It also tests its assumptions for causality, understands data is time-dependent and tries thousands of models to pick the best. The result is a system that can solve enterprise process automation issues in a fraction of the time to do it manually.

White Paper

SAIBRE AI Ecosystem

End-to-end AI applications that solve any business problem