AI Operations Tools
AI automation is capital allocation. The same math your CFO uses for factories, fleets, and production lines applies to knowledge work automation - with one twist: these assets can appreciate.
Models depreciate (distribution shift, competitive catch-up). But verifiers, data moats, and institutional knowledge compound. The dual curve - depreciating models plus appreciating data - is what makes AI capital allocation different from buying a machine. These four tools are the decision sequence.
Is this automatable? Map any task by calculation difficulty vs. verification difficulty. The Templeton Ratio determines whether AI saves you time or doubles it.
What are the stakes? Replace accuracy with dollars. Calculate optimal thresholds from the actual cost of being wrong in each direction.
What moves improve your position? Five capital investments that shift tasks to better quadrants. Build verifiers, decompose tasks, enrich inputs.
What is the spread? Calculate the risk-adjusted return on construction cost. Assess each opportunity, then rank them to decide where the next dollar goes.
Should you automate this task at all? Score it across 9 dimensions. If any single dimension is a landmine, it is a hard no regardless of the composite score.
The Thesis
“What accuracy do we need?” is the wrong question.
“What is the expected cost of errors, how does that compare to alternatives, and what investments change the equation?” is the right question.
Every framework here encodes a specific failure mode I have seen kill AI projects inside PE portfolio companies. The Verification Trap. Symmetric threshold bias. Investing in models when the moat is in verifiers. These tools make the failure modes visible before you ship.