At the rate that Artificial Intelligence technology evolves, it sometimes feels that you need a whole new dictionary to understand AI. Terms change, definitions overlap and technologies merge.
Intelligence Process Automation (IPA) is the perfect example of that rapid change. As the concept grows in popularity, the term gets tossed around more frequently. But before you can decide whether IPA is the right solution for your business, you need to know exactly what you're discussing. So what is Intelligent Process Automation, really?
In simple terms, IPA is a comprehensive approach to business automation that draws on a variety of intelligent solutions, from machine learning to natural language processing to deep learning. It's a type of artificial intelligence that leverages the best features of common business analysis and cognitive automation technologies to create automated, intelligent processes.
This means that IPA can mimic intelligent human abilities in standard business operations. Once the program is set up, IPA can understand and automate a typical operation performed by human beings that requires both intelligence and the ability to make informed decisions. Moreover, IPA becomes more independent over time and can achieve a truly autonomous state where humans are only required to confirm high-stakes decisions.
As McKinsey put it,
"At its core, IPA is an emerging set of new technologies that combines fundamental process redesign with robotic process automation and machine learning. It is a suite of business-process improvements and next-generation tools that assists the knowledge worker by removing repetitive, replicable, and routine tasks. And it can radically improve customer journeys by simplifying interactions and speeding up processes."
This holistic approach is what makes IPA so revolutionary. It's the next level of AI, and it's raising the bar for other solutions. By combining automation and advanced AI, IPA solves business challenges at an enterprise level and delivers on desired outcomes.
Technology that's designed to optimize businesses processes works in a variety of ways, such as data collection, automation and intelligent analysis. These categories cover a range of activities, from simply executing processes to mimicking thinking and applying artificial intelligence.
IPA essentially layers AI (namely, machine learning and deep learning) on top of process automation (like Robotic Process Automation or RPA). The tools that IPA draws on break down into a few key categories:
Data acquisition and ingestion refers to how your data is collected and passed through systems that can process large volumes of data. It also includes how your data is securely stored, and what methods you have in place to review the veracity of your data.
Ingestion is done by tools like process mining, document processing, natural language processing and optical character recognition. All these methods can improve your access to, acquisition of and ability to process data.
Automation tools, including RPA and automated workflows, take repetitive processes and use software to do them for you. Without some level of intelligence, this usually requires intensive human training and oversight of a program that does concrete, consistent steps to accomplish a task.
For example, an RPA program might be software that records your mouse and keyboard strokes to discreetly learn every single action you take to complete a task. However, it's difficult to evaluate these programs: you'll often have numerous isolated automation activities running at the same time, and the success or failure of one task might impact another.
Intelligent tools fall under the umbrella of AI, including machine learning, deep learning, process models, decision engines and deductive analytics. These tools analyze how you solve problems and find ways to come up with more efficient solutions, using cognitive capabilities or neural networks to understand and optimize your business.
Image from medium.com.
IPA does it all. It sits across all the categories described in the figure above and draws on the best of each. Combining these traditionally discrete approaches — ingestion, automation and learning — enables IPA to deliver true business intelligence and execute on those insights.
In other words, IPA enables end-to-end visibility into your business's process flows. You can view all key metrics and then optimize an entire system of tasks, rather than working to eke out change in silos.
Because IPA leverages numerous tools, it's a solution that stands on its own. This represents a major step forward from its predecessors, which tackled individual tasks, rather than the bigger picture of intelligently automating business operations.
RPA automates low-level robotic processes so that a software or robot can follow a set of predetermined steps. A tool that uses RPA isn't thinking in any sense of the word or using true cognitive automation — it's just repeating the same approach to the same manual task and automating the actions.
IPA improves exponentially upon this model by adding in artificial cognitive processes. The robot can leverage intelligent decision making to choose the right automated process to apply to a particular problem (where previously, a human would have to intervene).
Machine learning is a type of intelligent program that uses rules-based automation, or "if-then" processes, to learn. You can automate tasks, like identifying if a picture contains a cat or a dog, and use reinforcement learning to enable software to improve at those tasks based on experience. The software can also use iterative processes to adapt to new information.
IPA takes this a step further by enabling intelligent automation of a whole series of tasks, with intelligent decisions made within that series. This extra level of cognitive learning means that you can not only set rules or "if-thens," but also have the IPA program choose among those rules for the most appropriate response.
The key differentiator in the examples discussed above is the extent to which IPA can operate independently without relying on human supervision. This reduces the amount of training required for an IPA program to execute effectively, and it also ensures that the program actually saves time and resources. Unlike its predecessors, IPA can execute on human tasks that require cognitive ability and automate the processes associated with those tasks.
Let's look at a use case that's often the subject of business optimization: invoice processing.
You can image that for enterprise-scale companies like Walmart, invoicing is a major undertaking. Some invoices are likely emailed as a PDF, some are sent straight to their Enterprise Resource Planning (ERP) system or business management software as an electronic data transfer, some are handwritten and delivered by a local business owner... the list goes on.
Any one aspect of IPA could boost efficiency in that invoice processing. But without the complete suite of IPA's functionality, it wouldn't deliver significant business outcomes.
For instance, a document processing program could automate the scanning of a pile of invoices. However, it wouldn't be able to automate and optimize the entire operational workflow of receiving an invoice, scanning it, extracting the right data, mapping the data to a financial system, tagging accounting codes, setting a pay date, paying the customer, balancing the books and predicting the next invoice. IPA can.
Similarly, a generic AI program could focus on extracting data from those invoices into an Excel spreadsheet. That software is running an intelligent function — but it's not intelligently managing the whole process from start to finish. It wouldn't have insight into the entire system, so it wouldn't know which patterns to track or how to optimize the entire process flow.
IPA owns the entire process. It leverages a variety of tools to recognize the type of invoice and extract all data (date, vendor, price, line items, payment terms, payment due date, etc). It enters that data into the ERP, organizes it into the right accounting fields and sends out payment at the optimal time for high cash flow.
Employees can leverage the intelligence of IPA anywhere along that process; the business manager simply needs to set the right KPIs. They could have the program track pricing over time, identify vendor discrepancies, isolate error types, determine why payments go out late, predict issues during high volume periods and more.
IPA delivers on the promise of AI software by enabling impactful business insights into the whole story. Though IPA is in simplistic terms a combination of existing tools (from AI to RPA), it's so much more than the sum of its parts.