Process Reengineering For Agents & AI Platforms
Every technology initiative and purchase touches internal processes and customer workflows. Attempting to shoehorn AI and agents into existing processes and workflows causes the high project failure rates we’re seeing today. It doesn’t matter if the technology works. It must work the way people do.
For 10 years, one of V Squared’s success secrets has been process reengineering. It unlocks more ROI and accelerates adoption. With AI and agents, process reengineering is an unavoidable component of design and implementation.
When I started process reengineering with clients, I realized that most business frameworks are infeasible for enterprises. They rest on flawed assumptions:
Technology is stable for 10+ years without significant new innovations.
The business has the most modern technology available.
Processes are mapped and well-understood.
All processes generate value.
Most frameworks have been adapted from manufacturing and methods that rely on the stability of assembly lines from the 1960s. The modern enterprise is orders of magnitude more complex, touching physical and digital markets. Its processes require collaborative decision-making across the enterprise in the face of uncertainty.
The Myth Of The Finish Line
The first flawed assumption creates the perception that transformation has a start and end point. The business will transform again for the next major technology cycle, but there will be a lengthy 5+ year break between the end of one cycle and the start of the next. Most processes have been built with the assumption that transformation is periodic, but in reality, transformation is continuous.
Here are just a few significant disruptive technologies from the last 35 years.
1998: Internet
2002: BI
2005: Mobile
2008: Cloud
2012: Big Data
2015: Data Science & Advanced Analytics
2018: Machine Learning & Deep Learning
2022: Generative AI & Hyper Scaling
2025: Autonomous Vehicles, Agents, & Knowledge Management Systems
2026: Advanced Robotics (Physical AI), Decentralized Platforms, & Simulations
2028: Intelligent Internet, World Models, & Adaptive Autonomy
2030: Ubiquitous Reasoning & Nanotech Devices
2033: Quantum Computing & World-Scale Simulation
The transformation finish line is a myth. There hasn’t been a 5+ year break for over 20 years. Technology cycles are accelerating, multiple waves are happening together, and the magnitude of each technology cycle’s disruption is growing. That means business processes are constantly transforming, so enterprises must implement frameworks to manage continuous transformation.
Continuous improvement frameworks assume the process is stable and change is incremental. Continuous transformation frameworks manage processes that are being repeatedly transformed by new technology cycles and improved incrementally.
As Siddhant Khare pointed out during one of my community office hours, “Businesses still need to use the kitchen while it’s being remodeled.” That means business process reengineering (BPR) can’t stop or even slow operations. Toyota’s approach to process improvement included empowering any worker to stop the assembly line if they saw a quality issue. Doing that in the modern business will get you fired.
Any heavy, disruptive, Big Bang BPR method won’t work. Any attempt to force these methods on the business has the same result. Business leaders stop responding to meeting requests. Initiatives get slow-walked, and key resources are suddenly unavailable. Eventually, business leaders declare success, and BPR work is ‘no longer necessary.’
Legacy BPR frameworks are designed for an outdated paradigm where change was incremental, and disruptive changes were rare. We need something new.
How Well Does The Business Understand Itself?
In ‘Office Space’, the efficiency consultants ask everyone to justify their jobs. “What is it you’d say you do here?” BPR methods require 100% process mapping, and I can tell you from experience that most businesses don’t have anything even close.
Even if you ask everyone in the business, “What is it you’d say you do here?” you won’t necessarily get accurate or complete answers. Experts assume there is common knowledge about what they do and leave out pieces of their processes that they take for granted. Some people intentionally mislead process mapping efforts to protect their ownership of their domain.
The business’s process map is incomplete. Some parts will be inaccurate. The data we have about processes is in much the same state. Assumption 3 wreaks havoc on BPR efforts. BPR frameworks impose significant overhead for process discovery and mapping, but the end result rarely delivers the expected value.
Transparency & Opacity…Mostly Opacity
Most business processes are opaque to everyone except the people who manage them. Within that silo, process knowledge is high. Take a step outside the silo, and process knowledge drops to near zero because outside observers have no transparency into the process. Gathering data about a process increases transparency beyond the silo, but there’s a challenge.
The people who manage the process have high domain expertise and context about the process. Experts bring that context to raw data. Experts know what it means, so they can interpret and act upon raw data. Outside of the silo, people lack the domain expertise to know what to do with the same data. They need both data and context about the process that generated it to gain transparency into the process.
As it turns out, models and AI need the same context as non-experts to learn from information. The more time I spend building agents, the more I realize how valuable the information that powers agents is, but where does that information come from? Most methods don’t connect the dots between process reengineering and developing knowledge graphs that power reliable agents. The two have significant overlaps, and addressing them together reduces the cost of both activities.
In my last article, I explained how to run models in shadow. Spend enough time running models in shadow, and you discover a framework for building high-value information sets. When people do work on a platform, they generate data, but it’s rarely captured in a format that’s usable for model training or agent grounding.
Most businesses can’t generate high-value datasets because their workflows and processes aren’t built to generate them. That’s not how most approach business process transformation, but it needs to be. There are efficiency gains to be had through traditional methods, but getting there is expensive and takes too long.