Reading Dan Milsteins short excerpt The Golden Cesspool I thought how how amply it applies to our thinking about AI as well. Why we shy away from applying AI where it matters most to our business, and where I can create the most value.
What Is the Golden Cesspool?
The golden cesspool refers to the part of an organization’s processes and data that sits closest to its core business value and is, paradoxically, the most chaotic, opaque, and difficult to work with.
- It’s where decades of workarounds, exceptions, and undocumented logic have accumulated.
- It’s where systems were built under pressure, evolved over time, and never properly restructured.
- It’s also where the company’s competitive edge, regulatory compliance, and operational nuance live.
In short: the golden cesspool is the mess at the heart of what makes the business function and what makes it hard to change. You can’t put it on pause, you need to rebuild the plane while flying it.
Messiness Is Inversely Proportional to Importance
The processes that matter most, the ones tied to revenue, reputation, and real-world decisions are rarely clean or modular. They evolved through necessity, not design.
- In pharma, the preclinical-to-clinical transition is riddled with tacit rules, scattered documents, and fragile handoffs.
- In CPG, formulation logic and claims substantiation are buried across labs, spreadsheets, legal reviews, and marketing briefs.
- In financial services, pricing and risk models are entangled in outdated platforms and unwritten assumptions.
These are not technical problems, they are structural artifacts of how businesses actually operate. The sooner we embrace this and start working from a standpoint of seeing the mess as what actually makes us run the day to day business the sooner we can act on bringing additional value.
Why AI Transformation Stalls Here
AI thrives on structure: clear data, repeatable decisions, clean interfaces. The golden cesspool offers none of that.
- Data is fragmented, mislabeled, or not collected at all.
- Processes depend on human judgment, side conversations, or legacy tools.
- Every exception is business-critical and politically sensitive.
- There’s no sandbox. Only live operations with high risk and low tolerance for disruption.
Ironically, this is also where AI could add the most value. If it could be applied meaningfully. And I would say it can.
The False Hope of “Fix First, Then Automate”
One common but flawed approach is to try to clean up the mess before bringing in AI:
“Let’s redesign the process, structure the data, and build a platform. Then we’ll bring in AI.”
This rarely works:
- Business can’t stop while you clean.
- The complexity is deeper than anticipated.
- By the time you finish, the goalposts have moved.
- You’ve spent millions and still haven’t shipped anything real.
The perfect foundation never arrives. Meanwhile, the opportunity cost compounds. Again, you have to rebuild the plane while flying it. And with the speed of AI development, you might not even be certain of the course.
Build Through the Mess
Instead of waiting for structure, use the mess as the entry point. This is where a startup mindset is essential.
- Find a narrow, real pain point in the golden cesspool. Something concrete, painful, and high-leverage.
- Ship a constrained, useful tool. It doesn’t matter if it only handles a sliver of the edge cases.
- Let the system learn in production. Not just from data, but from feedback, correction, and co-usage with humans. Iterate with humans in mind
- Use success to earn trust and access. Not just technically but organizationally. This is how you build leverage for your next use-case.
This isn’t “move fast and break things.” It is: move incrementally and prove value fast.
What This Requires
- A shift in mindset: from systems design to system apprenticeship. You increasingly learn what you have to build.
- A willingness to tolerate partial automation. Your humans are the most valuable organizational asset, build for them.
- A tight loop between product, operations, and engineering. Cross-functional teams are a must for any AI iniative.
- An understanding that AI is not a layer on top. It’s a reframing of how decisions are made. It is between business-as-is and a new augmented reality
Conclusion
The golden cesspool is where your most important decisions live, and where AI can have the most impact. But it won’t come through top-down design, data perfection, or grand rewrites.
It comes by engaging with the mess directly. Solving something real, today. And letting that success pull the rest forward.
Not glamorous. Not clean. But real.