RPA + AI: Smarter business process automation

Companies that want to take full advantage of the benefits of Robotic Process Automation (RPA) are beginning to see artificial intelligence (AI) as a key technology to optimize their processes.

Building artificial intelligence into business process management is not easy. Many companies add artificial intelligence to processes by building or purchasing single-tasking robots (for example, natural language processing systems or visual recognition tools) and adding them to the process using traditional non-artificial intelligence methods. For example, engineers write scripts and business analysts use process visualization tools to create automated workflows.

However, combing the process, connecting disparate systems to a unified process, modifying processes as the business evolves, and discovering and resolving problems still require humans to develop their ingenuity.

Artificial intelligence, machine learning and related technologies are now entering this field through Robotic Process Automation (RPA). According to McKinsey, the combination of artificial intelligence and RPA constitutes Intelligent Process Automation (IPA). In addition to RPA and machine learning algorithms, IPA also includes process management software, natural language processing and generation, and cognitive agents, or “robots.”

According to McKinsey, IPA means that efficiency can be increased by 20% to 35%, processing time is reduced by 50% to 60%, and return on investment will reach three digits. However, it may be premature to say this because most companies are in the early stages of development and are still using the various components of artificial intelligence, and rarely connect them to a complete end-to-end automated process, let alone Connected to the artificial intelligence process.

Gartner analyst Moutusi Sau mentioned RPA’s application in the financial services industry: “There is no application scenario that can reflect the whole process. There are already some chat robot engines and artificial intelligence decision tools, but it is difficult to use. On a specific solution, the bank wants to do more than one thing.”

Unremarkable robot

For many companies, the journey to automate intelligent processes begins with a single intelligent robot, usually a chat robot that answers questions from customers or employees.

This is the case with the German ZF Group, which is the world’s third-largest automotive supplier and has only begun to apply intelligence to its business processes more than a year ago.

Andreas Bauer, the company’s IT manager, said: “There is a lot of repetitive work in the communications field of our company. There are a lot of emails in our inbox, many of which are duplicates.”

Using containerized workloads can greatly increase the efficiency of the software development department – ​​which is why we are expected to fully adopt containers in the next few years.

The first step is to create a robot, which is the basic tool for answering the most repetitive questions.

He said: “As a first step, it is very lightweight. For example, if someone asks if they can do a certain job, or can they find an open position? We can follow one use case and solve it. One, and then another. We are now developing a financial robot that can help customers understand the status of their invoices or bills.”

However, once most of the steps in the business process are automated, then a higher level of intelligence can be applied—the process itself is intelligent. In this way, companies can focus on the future when choosing a supplier for their robots.

“We are moving in the direction of automation of the entire process chain,” Bauer said. “We are not just looking for a robot. What we are looking for is a process orchestration and integration platform where we can easily adopt these technologies and They are combined with intelligence.”

He said that the ZF Group is looking for a platform that can learn from experience while avoiding unintended consequences. “I think, many people have heard of Microsoft’s robots going crazy.”

Since automation integration and process orchestration are the ultimate goals, the company also needs a platform with built-in checks and checks and balances. He said: “Some people are worried that certain things will become crazy, but we have no control. We must be careful and we must pay close attention to technology. Technology does not protect itself. You must work hard.”

ZF Group chose Vizru, a robotic platform that provides management, governance and language support for robots. It is called the Artificial Intelligence Process State Network (SNAP), which blocks robots if their behavior is abnormal. According to Vizru, SNAP can also mark or stop a session process if there is a violation or if sensitive data is shared between processes that should not be shared.

Decision point

Another approach is to add intelligent decision points to traditional automated business processes.

American Fidelity Assurance is working on this. The Oklahoma-based company provided 2.5 million policies to 1.5 million policyholders. The challenge for Fidelity in the United States is how to automatically send a large amount of emails received every day to the right destination. In the past, it was up to people to decide where each message should be sent.

Shane Jason Mock, vice president of corporate R&D, was inspired by Amazon’s work in this area: “Is there a way for advanced machine learning technology to learn from past data, past decisions, and humans? The same decision.”

He said: “It is a challenge to observe Amazon’s warehouse. I know that many people are doing various packaging or other incredible things there. Maybe this is not what the insurance industry can do. But the point is not what others have done. And how can it help our customers.”

Fidelity chose UiPath, the enterprise RPA provider, and DataRobot, the artificial intelligence platform, to make its processes smarter.

He said: “In the new email process, we combine the RPA components with the machine learning components. The combination of these two components determines where the email should be sent.”

In many cases, the traditional RPA approach encounters a decision point that is too complicated for a simple automated process.

The company also wants to use artificial intelligence for process mining to automate process discovery, rather than letting business analysts figure out how the company works internally.

He said: “We are working on some proof of concept. But it is too early to comment on this.”

Process mining

Traditional business process management methods include business analysts talking to managers and employees, auditing, and then creating charts to illustrate the various business processes of the business.

Sumeet Vij, director of strategic innovation at Booz Allen Hamilton, commented: “Many customers are involved in our work and there is a process workflow on the office wall. But is this the case? You will find The actual situation is different and there are various difficulties. Using machine learning for process mining helps people understand how the actual situation occurs.”

In addition, these tools can update processes as your business grows, and even discover anomalous behavior in real time.

Chart Industries, Inc., a manufacturing company that serves the energy industry, is headquartered in Ball Ground, Georgia, and has adopted an intelligent process mining system.

A few years ago, the company faced many difficulties. The energy industry was hit hard by falling oil prices, the company’s share price fell, and top management was frequently replaced. The new leadership wants to reform. For example, Chart has three main divisions. Although they share an ERP system for Oracle and JD Edwards, there are multiple back office departments that handle accounts payable, accounts receivable, and other background tasks. Each back office has its own processes and Procedures.

Bryan Turner, executive vice president of IT at Chart, explained: “We found that our customers actually accounted for the late payment to us.”

There are other opportunities that affect cash flow. For example, in some cases, companies can use a discount over a certain period of time to pay a supplier; in other cases, the longer a company holds cash, the better it is for the company. Turner said the benefits of improving efficiency in this area can reach millions of dollars.

Chart turned to process mining provider Celonis to help identify such opportunities.

Turner said: “We are currently running it on some custom systems. As long as it has a database, session and timestamp, you can push it to Celonis. A lot of the heavy work is how to work with our SaaS application or Celonis in our department. Moving data between Amazon backends.”

Celonis examines and identifies business processes in detail—not a theoretical process, but a process that can be actually implemented. Then it uses machine learning to identify patterns and find anomalies.

Business processes can be viewed in the form of charts such as Visio diagrams, and managers can drill down into the process up to the level of a single session.

Turner said: “Only overdue payment, we saved $240,000 a year. The software has already recovered its own costs many times. We also noticed that the cost opportunities are related to our suppliers and ours. Customer related.”

How much data do you need?

Artificial intelligence systems typically require millions of data points to make available predictions. Few companies have so much internal data about business processes.

Ray Wang, principal analyst and founder of Constellation Research, said that due to the existence of its business process mining platform, Celonis is leading the way in using artificial intelligence to help companies automate intelligent processes. He said other business process providers, such as Workday and Salesforce, are also good at helping customers automate the discovery and management of business processes using their historical data.

He said: “They may be able to reach a certain height: they can orchestrate a process and then plan for the next best practice. But it still takes time.”

Some vendors that offer enterprise ERP, CRM, and similar platforms may begin offering intelligent process automation tools in the future (if they don’t have one yet). For example, Salesforce is providing smart tools through its Einstein platform.

For such situations, companies benefit from artificial intelligence that has experienced all supplier customer data training. In other cases, companies can purchase pre-trained models and adapt them to their needs, as well as open source or commercially available training data sets.

In addition, enterprise data can be augmented with external data that helps inform business processes, such as weather data and financial market data.

Vij Allen Consulting’s Vij commented: “The more data, the better the robustness of the algorithm. But we also realize that many times, when we are involved, the customer does not have all the data.”

Vij says smart applications can be implemented without the need to fully automate the entire business process. Many business tools do not have digital interfaces or APIs, and some business processes require a lot of manpower. The intelligent process knows when to send the task to someone for processing. Sometimes, a step that seems to require human involvement may not actually be needed.

Vij said: “People in SharePoint and Drupal are unstructured and need people to view and find them. But you can apply advanced natural language processing techniques to extract structured information instead of letting people find it. ”

He said mature management processes include HR processes such as on-the-job and financial processes such as claims processing.

Business process analysis

Seann Gardiner, senior vice president of business development at DataRobot, an artificial intelligence platform provider, said that some of the most advanced companies have enough business process data, and now they can look at the overall situation of what is happening, and analyze and predict.

He said: “They found problems in the RPA process, tried to grasp these problems, and learned from them to make these processes smarter. I dare not say that this is widely used in enterprises, but it does have “”

He added, “If a company is very focused on process-level automation and can break down data silos, then it is indeed ready. But there must be business leaders who believe in automation and artificial intelligence prioritization, and can carry out the necessary organization. Structural reform.”

He said that Fortune 5000 companies are ready and have developed processes that combine artificial intelligence with RPA. “The question is, will they really implement it and carry out large-scale reforms in organizational structure?”