The Underlying Disease of Automation

Since the invention of personal computers, technology has significantly influenced business operations. The digital transformation journey, spearheaded by giants such as IBM and Salesforce, has drastically changed the enterprise landscape, positioning software as a crucial component of the modern economy. Innovations like the internet, cloud computing, and smartphones have transformed work, leading to unprecedented levels of productivity and flexibility. Nonetheless, this progress has been accompanied by challenges, particularly for large organizations struggling with an overflow of data, poor data quality, and a shortage of skilled workers, which impedes the full realization of digitalization's potential.

The narrative of enterprise software is characterized by a contradiction: despite technological advancements, companies continue to struggle with fundamental processes. This paradox is partly rooted in the design of current enterprise solutions—complicated databases with inflexible, rules-based workflows that have barely evolved, mirroring outdated technologies more than contemporary tools. Businesses often find themselves caught in a cycle of manual tasks and antiquated systems. The introduction of Robotic Process Automation (RPA) has often compounded complexity instead of simplifying operations. RPA's promise of efficiency is limited by its reliance on predefined processes, which struggle to reflect the dynamic nature of real-world business activities.

The core issue lies in the traditional approach to automation, which relies heavily on hardcoded, predefined processes. This methodology proves inadequate in addressing the intricate and evolving nature of business workflows. Real world operations are  not a series of linear, predictable tasks but rather a complex web of dynamic interactions and processes. RPA, while offering a semblance of automation, often falls short as it cannot accommodate the variability and complexity inherent in daily business operations. The reliance on static workflows means any deviation or unexpected change can render automated systems ineffective, highlighting a critical flaw in the current automation paradigm.

This realization sets the stage for a discussion on the emergence of AI and the transformative potential it holds for enterprise software. Unlike traditional automation solutions, AI offers the promise of flexibility and adaptability, capable of understanding and navigating the complexities of real-world business processes. The shift towards AI-driven automation represents a departure from the constraints of hardcoded solutions, paving the way for more responsive and intelligent systems that can truly capture the nuanced reality of business operations.

The Dawn of AI

The pace and innovation with which AI has become a mainstream technology is unparalleled. Similar to the groundbreaking introduction of the computer, the internet, or cloud computing, AI is set to usher in a new era of technological innovation and software development. The critical difference at this juncture is the advent of technology that can match human-like contextual awareness and flexibility in decision-making and reasoning. This capability means that technology can now be communicated with in natural language, making its impact and long-term implications accessible even to the most non-technical users and executives. While the exact pace and magnitude of change in the next 1-2 years remain uncertain, there is little doubt about the significant, lasting effects on the horizon.

This explains the unprecedented eagerness for adopting AI as a central component of enterprise strategy, signifying a hopeful shift where companies are prepared to embrace the risks and costs required to establish a foundation for genuine enterprise automation and enduring digital transformation.

This background is why we founded Interloom.

A New Era of Enterprise Software Automation

As outlined above, the foundational principles of current automation software have remained largely unchanged for decades. From Robotic Process Automation (RPA) to Business Process Management (BPM) Software and Intelligent Automation solutions, all adhere to the principle of hard-coding business processes within a workflow builder.

Despite the vast array of automation solutions available in the market, a fundamental problem persists. This issue stems from the fact that these solutions rely on hard-coded workflow builders, which, despite their no-code capabilities and sleek interfaces, consistently fail to grasp the complexity of the real world. This realization has motivated us to lead the paradigm shift towards adaptive automation, where work processes are inferred rather than rigidly predefined.

Rarely are processes consistent enough to be captured by a single sequence. Process mining, a significant innovation, highlights the variability in even standardized procedures like purchase-to-pay, which can have tens of thousands of variations. Our research suggests that only 30% of tasks and processes in an organization follow a predefined workflow. The remaining 70% of work relies on a "collective memory" of operational methods embedded within the company's culture. This reliance on institutional knowledge poses significant challenges, especially during generational workforce shifts, leading to knowledge gaps and inefficiencies.

Just imagine how a logistics company handles a customs declaration in a country where the logistical customs systems don't yet offer APIs, and regulations frequently change. Or consider a bank that manages hundreds of back-office processes, from account transfers and closures to meeting regulatory requirements.

It's the workforce's experience that enables these organizations to function, as much of this work is either undocumented or relies on outdated or misplaced documentation.

Well, until now…

For the first time, we have the technology to capture and process this collective memory. However, unlike many startups that have recently gained attention and funding, we believe a more sophisticated approach is needed than merely training a finer-tuned model that "understands" the enterprise or creates "AI agents for sales," etc. This realization brings us back to our consideration of tasks and processes.

Instead of defining processes through expensive consulting and development cycles—often conducted by those not intimately familiar with the subject matter or data—we suggest inferring work patterns from the daily activities of knowledge workers. These workers keep the organization running despite the challenges and limitations posed by their software tools, often stepping outside of predefined workflows to address issues directly, whether through phone calls, Teams or Slack conversations, or emails. These escalations, symptomatic of hard-coded and oversimplified processes, actually serve as valuable data sources for identifying work patterns and capturing the organization's "collective memory."

This approach offers numerous advantages, addressing many of the reasons organizations have been hesitant to adopt technology initially.

• Getting started is straightforward. Instead of engaging in costly migrations and change processes, the initial step is to equip knowledge workers involved in identified processes with tools to more effectively capture their work—by linking their email inboxes, Slack and Teams channels, and phone recordings or simply appending notes to their list of to-dos.
• After processing a sufficient number of cases through the solution, we can deduce common work patterns, using a blend of task mining, vector embeddings, and reinforced learning.
• By perpetually identifying the specific context of a task within a workflow, we can enhance decision-making by relying on the precedent created across the organization for any particular process execution, task, and case notes.
• Subject matter experts and knowledge workers, deeply familiar with the data, patterns, and challenges from handling cases daily, directly shape the precedents through their comments and notes.
• This approach significantly reduces the need for the consulting and development resources traditionally required for adopting automation. We can now identify patterns and rules using only natural language, applicable in any language, by empowering the knowledge workers handling the actual problems to immediately sway the future executions and inference.
• The exception becomes the rule: While hard-coded processes lead to escalations, we employ a human-default fallback strategy that begins with manually executing the process before automating common patterns. Although human intervention will always be necessary for addressing new problems or cases, adaptive automation addresses hundreds or thousands of process variations that are all valid, without the necessity of detailing them in cumbersome rules engines.

Adaptive automation approaches process automation similarly to how modern car autopilots are trained to navigate the complexities of real-world traffic scenarios. Instead of rigidly programming rules for every situation, such as stopping at a stop sign, we collect and train AI models on millions of scenarios. This training enables the autopilot to learn and infer the usual rules of the road, making exceptions only when necessary to prevent accidents, for example.

Based on our 15 year experience in enterprise automation, we are convinced that adaptive automation is the solution to many of the challenges and inefficiencies that currently impede automation within the enterprise environment.

Furthermore, a significant byproduct of this approach, which is likely to become its most valuable asset, is the collection of highly contextual data at specific decision points. While traditional methods might teach an AI model to never red-line contracts below a certain value (e.g. €100,000), this approach might not apply in special cases, such as an acquisition offer. This company's “lived experience” or precedence would be evaluated and considered via the model trained within the unique context of an unseen acquisition process, meaning that the general rule of not red-lining contracts becomes irrelevant.

This signifies that adaptive automation isn't merely about optimizing AI for specific decisions; it's about tailoring AI to make precise decisions within the context of specific tasks and processes. This significantly simplifies the complexity of decision-making and improves the accuracy of quality assurance and training data.

Interloom set out to pioneer this new era and category of adaptive automation. We are deeply invested in creating the necessary infrastructure, platforms, and application layers to facilitate this innovative approach to automating virtually any work pattern. While our primary focus remains on serving mid- to large-sized companies, we believe our methodology has broad implications for enhancing teamwork, even among small teams, particularly as AI agents become more integrated into daily operations. We are eager to continue sharing our insights and the conceptual framework of our developments, encouraging others to engage in the discourse on adaptive automation.

We hold the conviction that addressing adaptive automation represents one of the most thrilling and influential challenges of the coming decade. Despite its technical demands, the clear benefits it promises make us confident in its widespread adoption.