Field note

How to run a Decision Cost Audit

By Ganesh R Datta. I install sales infrastructure for coaches and founders. A Decision Cost Audit is the engagement. The Framework is the lens.

A Decision Cost Audit is not a software demo. I sit with one workflow lane, inventory the recurring decisions, force Automate / Assist / Keep Human, then we implement one quick win and measure it. AI can draft. I lock the call. If nobody signs off, it does not ship.

People ask me how to run a Decision Cost Audit as if it were a plugin. It is not. It is a working pass on one messy lane, using the Decision Cost Framework I declared. If you want the definition, use the source page. This note is the sequence I actually run.

The split the audit forces. If every item is the same letter, start over.

Pick one lane, not the whole company

I run this on one workflow lane. Sales follow-up. Content publishing. Support triage. Ops reporting. If you try to audit the whole business in a week, you get a slogan. You do not get a change you can measure.

The unit is a Decision Episode: something triggers, a human has to choose or authorize, and work moves, waits, or escalates. Recurring episodes are the pile. One-off strategy calls are not.

This is the public sequence. The interview probes and scoring thresholds stay closed. I am not publishing the kitchen so someone can photocopy a fake audit.

Inventory, then force the three-way split

We list the recurring decisions in that lane. Not tasks. Decisions. Approve the follow-up. Pick who to call. Decide if the exception is real. Assemble status so you can decide if things are okay.

Each one gets a public-layer read: Load is capacity consumed, in person-minutes per week. Exposure is how bad the current mode is, 1 to 5. Disposition is Automate, Assist, or Keep Human.

If everything is Automate, you are fantasizing. If everything is Keep Human, you are protecting comfort. The method only works if some work stays human on purpose.

Human interview, AI draft, human lock

I do the interview. The notes get structured. AI can draft the analysis. I lock it. That order is the product. A chatbot that dumps Automate this on your calendar is not an audit.

Output is boring on purpose: top 5 Automate now, top 5 Assist, Keep Human (mandatory), one quick win, a before/after measure, and a 30-day recommendation list.

Then we implement one win. Not a roadmap. One. Including any new babysitting time the automation creates. If the machine needs a babysitter, that time goes back into Load.

What I will not do

I will not publish a magic multiplicative score. Frequency times time times complexity times switching times consequence times people is a vanity explosion. Three scores beat one fake number.

I will not claim a university certified this. Adjacent work exists: transaction cost economics, cognitive load, automation reliability. Those are citations. The Decision Cost Framework is the named method I stamped v0.1 on 10 August 2026.

I also will not pretend sales infrastructure and Decision Cost are different religions. Late follow-up is usually a decision sitting on a person. Booking without recovery is a decision with no owner. Same leak, named properly.

How to run the check

Pick one lane. Write the eight recurring decisions, not the eight tools.
For each, mark Automate, Assist, or Keep Human as a gut call. If they are all the same letter, start over.
Run the 10-question self-assessment on that lane only. Score out of 20.
If you are at 10 or above, book the audit. Bring the lane, not a wish list.
After the audit, implement one quick win and measure before/after, including babysitting time.

What I actually ship

Top 5 Automate now. Top 5 Assist. Keep Human, written down.
One quick win implemented, not a 40-item backlog.
A before/after measure that includes new babysitting time.
AI draft, human lock. No unsigned recommendations.
I guarantee the working pass. I do not guarantee your revenue.

When	What actually ships
Intake	One named lane. Recurring decisions listed.
Interview	Human-led. Structured notes. Not a survey dump.
Lock	AI draft. Human sign-off. Then one quick win.
Day 30	Measure before/after. Case narrative if the win held.

Sources I will actually link

The source page for the method. The definition note. Wikipedia for cousins. Cal.com for the session.

Questions I actually get

What is a Decision Cost Audit?

The engagement that inventories recurring decisions in one lane, forces Automate / Assist / Keep Human, and ships one measured change.

Does AI ship the recommendation?

No. AI can draft. A human locks it. If nobody signs off, it does not ship.

Can you audit the whole company in a week?

No. One lane. A whole-company pass in a week is a slogan.

Is this approved by anyone?

No. Original method, v0.1, declared on the source page. Adjacent research is cited. Nobody accredited it.

Who should bring what?

One workflow you actually live in. Not a tour of every tool you own.

Other field notes

All field notes · Decision Cost Framework · Sales infrastructure · Book a session