Robotic Process Automation (RPA) and Artificial Intelligence are two of the most consequential technologies in the business world today. Both can help companies save time and money. They’re often complementary technologies, making the difference between the two even more confusing.
Some of these technologies are so transformative that the Harvard Business Review describes Generative AI as need to know for young professionals.
Let’s dive in and explain what both of these technologies are.
What is Robotic Process Automation?
Robotic Process Automation (RPA) is a technique that enables companies to automate repetitive tasks. RPA solutions use bots that are designed to interact with computers the same way that that humans do, such as typing on a keyboard, clicking the mouse, closing windows, and other common tasks that humans do naturally.
When people hear about RPA robots (Bots), they typically imagine physical robots, like Rosie from the Jetsons, typing away at a keyboard to create reports or perform tasks. In reality, the process is much more boring, and bots aren’t really robots at all. A bot in terms of RPA is a piece of software that runs a repeatable script that was pre-programmed by a human.
These scripts can be scheduled to start at a specific time, start when a certain business condition is met, such as receiving a new invoice in your e-mail, run when prompted by a human, or can be setup to run on a continuous loop until a set of tasks is complete.
RPA is suitable for automating simple tasks as well as complex processes but may take a significant amount of time to program. Even in a low-code programming environment the majority of the developers time is spent figuring out how to handle errors. For example, what should the bot do if an error message appears?
RPA technology has many benefits such as automating repetitive processes, reducing human error, eliminating time consuming tasks, eliminating data entry, data extraction, and automating other manual tasks.
It’s important to know that two types of RPA exist, attended and unattended.
What is the difference between attended and unattended RPA?
Attended Automation is a process automation designed to work in in conjunction with a human or requires a human to start a bot. These tasks are often triggered by a hotkey or button. In this way, a bot will only run when started by a human.
Unattended Automation is a process that runs by itself without human interaction. Imagine a scenario where an e-mail comes into your inbox, the bot recognizes it, and instead of it going to Accounts Payable for data entry, a bot recognizes it as a candidate for invoice extraction and sends it through an Optical Character Recognition (OCR) model then processes the results by entering it directly into the accounting system.
Which Processes are Good Candidates for RPA?
Manual tasks based on structured data, that are high volume, and easily repeatable are great candidates for RPA. Because RPA technology requires a developer to create a script, it can be fairly expensive to implement. Keep in mind a project must have a return on time and monetary investment. Here are some great candidates for the use of RPA tools.
- A Process is based on Structured Inputs such as Databases, Excel Worksheets, Websites
- Simple processes that are high volume and easily repeatable.
- Tasks with similar processes where existing RPA scripts can be leveraged.
- Processes that replace human labor.
- Data centric tasks where software bots can interact with computer systems.
Which Processes are Bad Candidates for RPA?
RPA can be expensive to develop for complex tasks. These are some common tasks that may not be good candidates for automation and should be left for human workers.
- Processes with a lot of decision points that require frequent human intervention.
- Tasks where rule-based processes do not exist. (Tasks requiring judgment calls)
- Processes that are low volume and are not worth the expense of automation.
- More complex tasks such as handling large amounts of unstructured data.
- Non-Digital Systems where RPA tools have no way to interact with data
- Data Analysis, Forecasting, and Reporting that Requires Context
Good vs Bad Use Caes for RPA
In summary, to implement an RPA solution, there has to be a return on investment. It can take a considerable amount of time to automate a process and setup error handling within an automation script. Simple high-volume processes are great candidates because they take a lot of human labor to execute, and reducing labor can offset the cost of developing a software bot.
What is Artificial intelligence (AI)?
Artificial Intelligence (AI) is a branch of computer science that attempts to re-create human intelligence with computer algorithms. There are a few primary categories of AI that are especially relevant to process automation.
Natural Language Processing
Natural Language Process (NLP) focuses on the interaction between humans and computers where humans talk or type to a computer using questions and responses as if they were interacting with another human. The latest example of advanced NLP is ChatGPT which was developed by OpenAI.
Machine Learning Machine Learning (ML) focuses on training algorithms to ingest sample input and output data, then take new input data that it’s never seen before and come up with the correct output. This technique is often used in image recognition and invoice processing.
Predictive Analysis Predictive Analysis focuses on using Machine Learning algorithms to look at historical data and forecast the future. This technique could be used in something like tax forecasting based on prior year tax payments or forecasting inventory to assist in purchase decisions. It’s often used with much larger datasets and larger number of variables than what you would use with a more traditional regression model.
Computer Vision Computer Vision focuses on training machines to interpret and understand visual data from the world around them. It can be applied in many areas, such as identifying which vendor an invoice is from to more advanced areas like autonomous cars.
What does Artificial Intelligence Look Like in Practice?
Definitions are nice but let’s look at a few ways that you could deploy various aspects of Artificial Intelligence in your business to bring the definitions to life.
How to Use Natural Language Processing?
Chat GPT is one of the best examples of Natural Language Processing available today. The following is a screenshot of asking the chat bot how do you write a year to date formula in Power BI? The answer is generated based on an AI model behind the scenes synthesizing an answer from massive amounts of unstructured training data like scraped websites, encyclopedias, computer coding libraries and other large sources of data.
How to Use Machine Learning and Computer Vision?
Computer algorithms have existed for decades that can convert written text to typed text. This process is called Optical Character Recognition (OCR). Even the best models today are only 70% to 90% accurate. When applying machine learning models to an OCR task the success rate improves dramatically.
In this scenario a process might look something like the following. Software robots are setup to watch an e-mail inbox for incoming attachments. The attachments are brought into a Machine Learning model and the images are categorized. If the RPA bot recognizes the invoice it’s routed to an RPA process that handles similar tasks of converting unstructured data into structured data and extracts it for upload into an ERP system.
If a software robot does not recognize the image, it routes the invoice to human workers who code the invoice by selecting sections of it. Drawing boxes around the invoice date, around the company logo, around the total invoice amount and other important line items. In this way, the machine learning model is “trained” on what the correct answer looks like. The feedback loop is similar to human behavior in that the feedback trains a model that becomes more precise over time. Over time and with larger data sets there are less exceptions, that require less human intervention.
What is AI Automation?
AI Automation is the application of AI techniques for business process automation. It could be used to automate complex tasks, or routine business processes. Rather than a human writing scripts that are set to repeat such as with RPA, it uses Artificial Intelligence techniques such as Machine Learning, Natural Language Processing, Computer Vision, and other AI systems to re-create a process on its own.
What is Intelligent Process Automation?
Intelligence Process Automation (IPA) combines artificial intelligence, machine learning and natural language processing to automate the analysis of unstructured data such as images, and videos. IPA systems ingest massive amounts of data to feed their artificial intelligence and machine learning models. Text takes up a relatively small amount of space compared to images and videos. One can only imagine the scale of data processing that a company like Elon Musk’s Tesla uses to train its self driving cars.
How do you use RPA and AI together?
The combination of RPA and AI is often referred to as Smart Process Automation (SPA). SPA utilizes Machine Learning to automate complex business processes. It takes traditional RPA and goes a step further by adding cognitive and analytical capabilities. Advanced Smart Process Automation can even use AI for decision making and learn from its mistakes.
Smart Process Automation uses AI and ML algorithms to automate tasks such as fraud detection, medical image analysis, investment management, drug discovery, supply chain optimization and other complex data tasks.
Example Use Case of RPA and AI Working Together
Here is a scenario where these two technologies, AI and RPA, work together to complete tasks.
Current Scenario: A Car Dealership uses outdated accounting software based on the PICK programming language, which was widely used in the 1980s and 1990s. The front-end of the software has been updated to look modern, but many reports require you to use a green screen command line like you would in Windows DOS to enter and extract meaningful reports.
Because there is no way to easily extract data using modern technologies like SQL or an API, human workers currently run reports, save them to Excel and create monthly financial reports and forecast future sales data based on a 6 month rolling average.
RPA: A software robot is setup to click through the user interface of the car dealership software to select the report that is needed and enters the required parameters into the command line screen to generate a sales report. The sales report is saved into a cloud service such as One Drive.
AI: Data analysis software such as Alteryx or Power BI connect to One Drive and import the excel spreadsheets that the RPA software robots saved. It’s consolidated and re-formatted for advanced data analysis. The software applies a predictive model to the sales data and emails the results to management.
In this simple example, human labor was eliminated by no longer needing someone to run a report, internal customer satisfaction is improved because reports come out regularly and on time all of the time, repetitive tasks are eliminated to reduce potential errors, and people are freed to focus on more value-added tasks.
What are some RPA and AI Use Cases?
Our recommendation is to first identify several tasks within your organization that could be good candidates for process automation. Here are some ideas of common use cases. Keep in mind these processes are fairly simple and highly reputable.
- Data Entry
- Invoice and Accounts Payable Processing
- Accounts Receivable Processing
- Fraud Detection
- Account Reconciliation
- Basic IT Support (Resetting Passwords, Common Technical Issues)
What are some RPA and AI Solutions Available Today?
RPA and AI solutions come in many different forms. Software can range from free to hundreds of thousands of dollars to implement and everything in-between.
These are some RPA solutions split out by relative cost.
Tier 1 RPA – Can cost into the hundreds of thousands to implement and maintain.
- Automation Anywhere
- Blue Prism
Tier 2 RPA – Costs in the thousands to implement and maintain.
- Microsoft Power Automate for Robotic Desktop Automation.
Tier 3 RPA – Costs are Free to Thousands of dollars to implement.
- Microsoft Power Automate Flows for Cloud Process Automation
We outline some of the Best Open-Source RPA Tools if you’re just getting started and are on a budget. Be aware that these solutions are not representative of the features available in the more expensive fully featured platforms.
How Expensive is RPA?
The cost to implement RPA solutions can vary greatly. There are 3 primary components to be aware of.
- RPA Software Licenses
- Implementation Costs
- Maintenance and Support
- Training and Development
Of the 4 primary costs associated with RPA, development costs are often the most overlooked. Even low-code tools require some basic programming knowledge. Developers are in high demand and can be demand high salaries.
How Much Does an RPA Developer Make?
Salaries vary greatly by software package used, experience, and industry expertise. People that can demand the highest salaries typically have a number of years of experience across multiple industries and are skilled across disciplines such as sales, marketing or finance.
- Enry Level Developers: $60,000 – $80,000
- Sr. Developers: $80,000 – $120,000
- Solutions Architects: $100,000 – $150,000
Ask for an RPA Proof of Concept Before Spending Money
It’s important to keep in mind that RPA is a software robot, and being able to interact with a computer is a pre-requisite for a process to be automated with this technique. If you are working with a consulting group or thinking about bringing in RPA software, consider asking the firm you are engaging to build out a Proof of Concept. They should be able to prove that the RPA tool can work with the system that you are wanting to automate. Even if it’s at a basic level. Not all RPA solutions are built the same, and some may not be able to work with your existing infrastructure.
Do You Need to Hire an AI Specialist?
Luckily most companies do not need to hire AI specialists. They can be incredibly expensive and come with letters after their name like PhD that significantly add to the cost.
Most companies can implement AI technologies in their company by using off the shelf products. One popular starting point is Microsoft AI builder that makes AI approachable for RP Developers or Business Analysts. Many RPA platforms integrate their own AI capabilities so that you don’t have to handle complexity related to deploying AI Models.
Where to Start Deploying RPA and AI?
A very simple way to start exploring RPA and AI hands on is to sign up for Microsoft Power Automate Describe it to Design it. Without using any code at all, you can begin automating manual processes. It’s as simple as using natural language processing to describe the process that you want to automate. The software will suggest an automation workflow for you to implement. It’s a quick and inexpensive way to eliminate tasks that were previously completed manually.
For example, how many times do you have to worry about email routing? With Power Automate, you can set up a process that will do it for you. You are only limited by the computer’s ability to understand your request.
What is the Future of RPA and AI?
One of the most exciting prospects of the combination of RPA and AI is the startup company Adept.AI They are actively working on combining all of the technologies mentioned in this article to allow people to use natural language processing to describe a process and have the computer perform it. Check out the demos that they have of their latest model ACT-1, it’s truly impressive and if they’re able to execute it could be a game changer for the entire white collar workforce.
Over time we see these technologies becoming more and more approachable and easy to execute. Chat GPT is already making the otherwise impossible, possible for people that don’t even know how to program.