/Perspectives · The algorithm

When to apply AI. And when not to.

Every week an owner asks me where to start with AI. Wrong question. The right question is what to stop doing first. AI is step five of five. Most teams try to make it step one, which is how you end up automating a process that never deserved to exist.

/ 01

Question the requirement

Most process steps exist because someone added them years ago and nobody asked why. Every requirement needs an owner with a name, not a department. "Compliance needs it" is not an answer; a person at compliance who can defend it is.

/ 02

Delete the step

The cheapest automation is a step that no longer exists. If you're not adding 10% of deleted steps back later, you didn't delete enough. Deleting feels riskier than automating, which is exactly why most teams skip it and pay for software instead.

/ 03

Simplify what's left

Collapse approvals, kill the duplicate data entry, standardize the inputs. Simple processes are the only ones worth speeding up. Complexity you automate is complexity you've decided to keep forever.

/ 04

Accelerate the flow

Tighten handoffs and cycle time with the people you already have. This is where the real bottlenecks become visible, and they're almost never where the org chart says they are.

/ 05

Then apply AI

Aim it where the tool does 80% of the work and a person owns the last 20%. Measure it in dollars and hours, never in demos. If you can't name the number it moves, you're not automating, you're decorating.

This is why our diagnostics start in your operations, not in a model picker. The companies that skip steps one through four buy software. The ones that don't buy outcomes.

Try the 5-question version on one of your processes →

Next perspective

What “AI-native” actually means →

Your move

If this landed, there's money sitting in your operations. Let's find it.

15 minutes. No deck. No spam. Bring the process that annoys you most and I'll tell you straight whether AI belongs anywhere near it.