IPA-Intelligent Process Automation

The Productivity Waves

The initial wave of automation for "service" processes commenced in the 1990s. During this period, ERP and CRM applications became standard in the corporate environment. Companies like SAP, Siebel, and later Microsoft successfully integrating core organizational processes (such as Finance, Sales, Manufacturing, and Procurement) into their software solutions. These applications enabled organizations to standardize and digitize their workflows. These processes generated structured business data that could be leveraged for informed decision-making.

The success of transactional systems, combined with the advent of cloud computing, led to the emergence of a new category in enterprise software: SaaS applications. Notable examples include Salesforce, which dominates the CRM sector; HubSpot, prominent in Marketing; ServiceNow, specializing in ITSM; and Workable, focused on HR. Despite these technological advancements, the fundamental objective has remained consistent: the standardization and automation of human-led processes.

The second automation wave started in the mid-00s, and it was all about data. data. By this time, the processing power of cloud technology had reached a level of significant maturity, prompting corporations to recognize the necessity of moving beyond "in-application" reporting and static planning to truly leverage their data . Traditional "Data Warehouses" with siloed data needed to transition to "Data Lakes" within the cloud. From the moment data is available in “Data Lakes” we have to build applications that create dynamic reports along multiple set of data. And since date data is so important there must be Data Governance tools (e.g. Customer Data Platforms) and Data Operation tools. Going further, organizations realized that they had to get insights from the outside world and not just from their own applications. The question then arose: how to effectively utilize all this data? Data Scientists provided the solution through Machine Learning applications. While the first wave focused on creating organizational (structured) data, the second wave aimed at automating its collection, management, and utilization.

The third wave of automation began with the advent of web APIs, OCR,  and RPA technologies, and is expected to reach its peak with the emergence of Agentic AI. APIs and RPA technologies have been instrumental in automating repetitive tasks, creating integrations between applications, eliminating redundant data entries, and enhancing productivity. These advancements are primarily observed in digitally mature organizations that have successfully navigated the first and second waves of automation and are seeking to expedite processes and fortify their data.

At the same period, the shortage of software engineers prompted software companies to introduce low-code platforms, which enable the creation of automation applications and workflows (e.g., Power Platform). This technological wave has resulted in organizations adopting architectures where transactional systems serve as the core, while low-code platforms facilitating organizational integration through apps, APIs, and RPA technologies.

The introduction of Generative AI did not initially impact the automation wave, as it was first positioned as an application assistant, such as Co-Pilots. Until mid-2024, the focus on AI was primarily on retrieving and processing information within applications or organizations (informational bots), rather than automating tasks. However, Generative AI possesses the capability to understand unstructured data. By revitalizing the concept of Optical Character Recognition (OCR), Generative AI can be trained on both structured and unstructured documents, facilitating faster and more cost-efficient data extraction. Once data is extracted from structured and unstructured sources, it becomes logical to leverage AI reasoning to draw conclusions. Furthermore, if AI can determine possible actions, it stands to reason to train it to input this data into deterministic applications. This leads to what can be named IPA- Intelligent Process Automation and will be the 3rd wave of automations. IPA integrates various technologies—low-code platforms, Document AI, Generative AI, Robotic Process Automation (RPA), APIs, and Orchestration Platforms—to create intelligent processes across departments and applications in hybrid environments. The most advanced IPAs will feature Agentic AI at their core.

A note on Process Mining…

Process Mining  is a decade-old technology pioneered by Celonis. It can analyze the logs and tables of software applications to create a comprehensive map of the processes within an organization, akin to how an MRI scans the human body. Once an organization understands how its processes are truly operating, their connections, and performance through a set of key performance indicators (KPIs), they can delve deeply into improving these processes. Organizations considering where and how to implement Intelligent Process Automation (IPA) projects should first invest in a Process Mining initiative. This will provide them with complete visibility of the potential value to be unlocked.

Process Mining subsequently evolves into actionable Process Intelligence. With holistic data on processes, Machine Learning can be employed to predict process outcomes through various scenarios. By incorporating near real-time data, organizations can monitor automation and deploy agents for process control. Ultimately, Process Mining and Intelligence can serve as essential tools for monitoring and optimizing complex Agentic Processes.

USE CASE

The standard procedure for recording Account Payables typically involves the following steps: a Purchase Invoice is received from the vendor by the Accounts Payable department. The accountant must perform several checks, such as verifying that the invoice matches the Purchase Order and confirming that the Payment Terms correspond with the contract. They must also enter the correct cost center, obtain the necessary approvals, and finally post the invoice.

An Intelligent Process Automation (IPA) of this process could proceed as follows: purchase invoices arrive electronically via email. An Agent, trained in classifying emails, identifies that the email contains an invoice rather than another type of message and forwards it to the Document AI Agent, which is trained to understand vendor invoices and extract the required data. This extracted data is then processed by the SAP Posting AI Agent,  which is trained to navigate the SAP interface for posting vendor invoices. The agent determines whether all necessary data for posting has been extracted or if additional information is needed—such as identifying a missing cost object. It then suggests to the user what cost object to input (via Human-in-the-Loop interaction). Once all required information has been gathered, the agent decides which bot to activate (based on its training in SAP screens and commands) and enters the data accordingly.