operations ai automation systems-thinking

Find Where the System Binds: How I Diagnose an Operations Problem

Slow onboarding, repeat support calls, reports nobody trusts. The delay is rarely inside a step. A three-question way to find where an operation actually breaks before you automate anything.

I learned to read systems on a massage table.

I started massage school at 18. A client would point at a sore shoulder, and the shoulder was almost never the cause. The hip was rotated. The foot was compensating. The pain showed up two links down the chain from where it started.

Operations work the same way.

The pain is rarely where it shows

When a process runs slow, the first instinct is to speed up the step that hurts. Better checklists. Faster ticket turnaround. A new tool on top.

Sometimes that works. Usually the slow step is the shoulder.

You don’t automate a broken process. You map the system first, find where it actually breaks, then decide what to build.

Three questions I ask first

1. Where does the work wait? Not where it’s slow. Where it sits. Work waits at handoffs: between two people, two teams, two systems that don’t agree.

2. Who owns the handoff? Every step usually has an owner. The space between steps usually doesn’t. If nobody can tell you who fixes a mismatch, that’s where the time goes.

3. What comes back? Repeat requests are a map. The same person asking about the same problem twice means the system didn’t learn from the first time.

What that looked like with 125 field reps

A retail field program ran 125+ reps across the US, Canada, and Mexico. Getting a new rep productive was slow, and nobody owned the slowness.

The steps looked fine. The roster lived in four systems: the master store list, the field app’s user list, the time-off tracker, and the Google Groups for each country. They never agreed. Every new hire meant reconciling all four by hand.

The delay lived between the systems.

So I built one automation that owns the rep lifecycle end to end: onboarding, leave, return, offboarding. It provisions the accounts, keeps the four rosters in sync, and manages 200+ distribution lists with a confirmation gate before every write.

Onboarding time dropped about 60%. Nobody moves faster. There’s just nothing left to reconcile.

What that looked like in a support queue

A hospitality software company had a support volume problem. The biggest ticket category was speed complaints, so that’s where everyone wanted to look.

The smaller category was overbooking. A prediction feature, sold as the product’s edge, reopened tables it expected to go unclaimed. Then the original guests arrived during dinner service.

Fewer tickets. Far more damage per ticket. And nearly every caller was a repeat caller.

Tracing it back, the problem crossed every handoff. Sales pitched the feature. Onboarding never asked about walk-ins or no-show tolerance. The algorithm ran on defaults. Support had no playbook, and nothing tagged a ticket back to the pitch that caused it.

No single team was wrong. Information dropped at every stage.

Then I build

Once I know where the system binds, the build is usually smaller than people expect.

I build automations that show their work: raw data in, visible logic out, so someone without a technical background can audit any change. If a tool only works while I’m watching it, it isn’t done.

If your team is losing time somewhere and you can’t point to the step, that’s usually the sign. Start with the three questions. If you want a second set of eyes, here’s how I work with teams.

Want the deeper story?

I write essays about identity, transition, and building a life that fits — on Substack.

Read With Jhi