Robotic Process Automation (RPA) will be at the forefront of AI. Combining the two technologies will allow AI to complete tasks without intervention from developers. Ultimately increasing corporate productivity and reducing the need for expensive developers. The pieces are already in place and the technology will rapidly improve.
AI will replace RPA in the near future. Scripts that RPA developers create in low-code platforms will be generated by AI prompts, tested and adjusted through fully automated processes. ChatGPT Code Interpreter, Zapier and Microsoft all have first generation solutions already available.
Artificial Intelligence is improving at incredible speed.
ChatGPT entered the mainstream in late 2022 and in less than 6 months has added features allowing it to search the internet, integrate directly with third party services through plugins, and will soon release Code Interpreter which will enable people to upload files and interact with them using Python scripting.
Zapier, Microsoft, Adept AI and many other companies are using ChatGPT style text prompts to automatically convert text to process automations.
To understand how the future will unfold, it’s important to have an understanding of how RPA currently works.
Let’s jump in!
How does RPA Work?
The RPA industry loves talking about bots. It captures people whose imagination thinks of a Jetson’s style robot of the future clicking buttons and completing tasks 24/7.
Sadly, RPA bots are not robots at all, but a marketing name for programming scripts that typically run on a schedule, a loop or when a trigger action tells them to start.
It’s much less exciting to imagine a computer being a computer running programs but it’s the truth.
RPA developers examine existing processes and replace the process with scripts that are built out in low-code software development tools.
A typical process automation flow looks like the following:
What does an RPA Developer Do?
RPA development became a very popular job role in the early to mid 2010s. It’s increased in popularity but is at high risk of automation.
RPA Developers have two primary job duties.
- Partner with business process owners to examine and understand existing processes.
- Convert observed processes into scripts that can be scheduled, triggered, or set on repeat.
Understanding existing processes is arguably the most difficult part of an RPA developer’s job duties. They then take the process and convert it into a computer script often using a low-code scripting tool.
As an experiment, write down how you toast a slice of bread.
Did you remember the step where you first walk to the refrigerator, or the step where you remove the twist tie from the bag of bread before picking a slice that’s not the end piece?
These are the challenges that RPA developers have to deal with, but on a much more complicated scale.
To make things more difficult, many processes being automated are performed by lower skilled knowledge workers that may not fully understand the why they are performing each step that they were trained to complete.
Many of popular RPA tools require a minimal amount of computer programming knowledge and script creation itself is relatively easy.
The hard part of RPA development is exception handling.
Let’s face it, computers break. Systems that utilize RPA to complete processes are typically outdated and break very often. RPA developers have to take these exceptions into account, and create processes that can be run repeatedly but also have error handling built into them so that the process won’t break if an error message pops up.
For AI to replace RPA it will have to address both of these primary job duties of RPA Developers
Let’s look at how it’s going to work.
ChatGPT and AI Process Mining
To tackle the first problem of understanding processes, there are two approaches being employed by companies today. The technology is still in early stages but will improve rapidly.
We will use examples using Microsoft’s RPA platform, Power Automate but every major RPA platform on the market today has similar technology and solutions.
How businesses are using AI Process Mining
Systems are being deployed to automatically track what employees are doing on computers to identify areas that can be automated.
AI Process Mining utilizes data extraction methods to scan event logs from business systems. Machine learning uses this data to model existing workflows, predict future events, identify inefficiencies, and suggest improvements, enhancing traditional process mining techniques.
The video below highlights how process mining currently works using Microsoft Power Automate.
How people can automate processes on demand with ChatGPT
Microsoft Power Automate Flows and Zapier are two of the largest players in cloud process automation.
Cloud Process Automation is a technique that leverages cloud-based software and services, such as Microsoft Power Automate and Zapier, to automate repetitive tasks. For example, someone could create a Power Automat Flow to automatically save email attachments to OneDrive, while Zapier could be used to trigger an action in Slack whenever a new row is added to a Google Sheets document.
Processes would historically be built out using a step by step low-code process and would look something like:
Step 1.) When an E-mail Arrives, check if subject line contains “Invoice”
Step 2.) IF Yes, go to Step 3. If No, End.
Step 3.) Download attachments from E-mails containing the word “Invoice”
Step 4.) Save e-mail attachment to a OneDrive Folder
The revolutionary change recently has been the integration of ChatGPT to automatically generate process flows on demand.
While still in early stages, people no longer have to add each step individually they can simply describe a process and the RPA tools will automatically generate each step in the process for you.
For straight forward processes, an automation that could have taken an hour to put together can take less than a few minutes.
On a recently published blog article, Microsoft revealed that Power Automate would predict and suggest the correct workflow 52.6% of the time on the first try. This was recently improved to 90.3%.
They also highlight that people use the workflows created by AI 76% of the time compared to 43% when they have to build and create the workflow from scratch.
The Power Automate team attributes the quick increase in system improvement to having more user data to reenforce and train the underlying AI models that power the technology.
RPA training data process automation gold
Companies that have the most training data will be the ones who create the most advanced AI Process Automation solutions.
Advanced AI Language Models like ChatGPT are trained on absolutely massive training datasets. They use text data from public and private data sources like Wikipedia, Reddit, academic journals, and information available on the internet.
For AI to fully RPA developers, the systems will need a lot of training data.
This is the biggest challenge right now and the reason why Zapier and Microsoft AI powered RPA solutions are only being deployed for cloud services.
Cloud automations are created in a controlled environment where a computer model can suggest a workflow and test it using a set number of Application Programming Interfaces (API)
When AI is deployed to larger more open ended problems, like asking it to open SAP, run a specific T-Code and export that report to a shared network drive the AI will need to have a familiarity with the many versions of SAP that exist on the market today, know where to click, and be aware of any custom T-Codes that a company might be using.
Desktop software is also prone to breaking and crashing.
An RPA powered by AI will need to know how to handle these exceptions. That means that the exceptions will need to occur often enough for there to be training data available.
Companies that are able to collect enough training data will be the ones who usher in a new era of AI performing tasks on its own rather than suggesting ways for people to complete them.
One possible solution to collecting enough training data is Microsoft’s recent announcement of a cloud based Windows product. Windows 365 for business is already available. A cloud based Windows would make it easier for businesses to track processes and for Microsoft to potentially mine the processes across a very wide range of use cases for those users who opt in to such a program.
RPA will train itself in the future
TechCrunch recently reported on physical robots using AI to train themselves to complete tasks by watching YouTube videos.
This is an incredible feat and a fascinating video to watch the process in action.
AI robots interacting with the physical world are still in early stages of development. In some examples it is reminiscent of a child learning how to perform simple tasks around the house.
It’s more interesting when you extrapolate the ability of a physical robot to complete tasks from watching videos to an RPA bot interacting with a computer system.
Similar techniques could go a step beyond AI process mining and allow companies to train AI to automate processes by recording videos of users interacting with computers to complete their daily jobs.
ChatGPT will replace Computer Programmers
Computer programmers face the highest risk of their jobs being automated. Tasks like test automation, mobile app development, and creating webpages can already be completed by non-technical users using low-code tools. AI will take things a step further and complete tasks on people’s behalf.
Business Insider recently called out coding, computer programming and software engineers at being some of jobs at highest risk of automation. These historically high paying jobs are prime automation targets due to the nature of the work and the advancements of large language models like ChatGPT that specialize in synethsising words, syntax and setnences.
There are also many online code repositories in addition to internal code bases that can be used as AI training material which eliminate the barrier to not having enough data to train a computer on.
We expect that some high level programmers would still be needed to manage and oversee the AI systems. This could create some opportunities for career paths that do not currently exist.
Going beyond computer programmers and RPA developers, we expect that many other jobs will be displaced by AI. Accounting and Financial Analysts are ripe for disruption along with other white collar workers.
People will need to examine how to handle potential widespread disruptions over the coming years. INSEAD recently published a great article outlining how they see emerging technologies like ChatGPT shaping business, society and employment.
Does RPA Currently use AI?
Up until recently, RPA did not use AI. The closest that RPA came is building AI into part of their process. For example, an RPA developer could setup a bot to capture e-mail invoice attachments as they arrive. Then the bot sends those files to an AI powered service like AWS or Azure Form Recognizer to convert an image to text, and parse out the important information.
Conclusion
The coming change of AI replacing RPA will not be immediate. Simple processes will become fully automated first, and increasingly more complex tasks will be fully automated using AI.
The ability to collect user data will become increasingly important. User provided data is going to be the fuel that AI engines run on.
We anticipate more tech companies moving systems to the cloud in an effort to capture an increasing amount of data to feed into AI models.
As the amount of data captured improves, models will improve and the need for RPA developers will continue to decline and AI will replace RPA.