Skip to content
Yagnesh Ramalingam

The Zero Entropy framework

One trusted number.Four layers, in order.

Companies spent a decade making data easy to reach. Leadership meetings still open with an argument about whose number is right. The fix isn’t more tooling. It’s four layers, built in order, and the first two need no software.

Book a 30-minute callSee the four layers

01 The problem

Four right numbers. One stalled decision.

Reaching the data got cheap. Agreeing on it didn’t. Every new dashboard adds another version of the same number that someone can defend, so the meeting argues about definitions instead of deciding.

Asked once, Monday

What was Q3 net revenue?

  1. Finance

    $18.4M

    Recognised, net of returnsThe close

  2. Sales ops

    $21.1M

    Booked at signatureCRM

  3. Marketing

    $19.7M

    Attributed, grossAttribution model

  4. The board deck

    $20.2M

    Nobody can reproduce itSlide 14

Every one of these is defensible. No line here is an error, and no amount of query speed narrows the spread. Four definitions were allowed to exist, so the organisation now occupies four states at once. That is the entropy, and it is what the meeting is actually arguing about.

Figures illustrative. The shape is from operating roles I held between 2016 and 2025.

02 The measure

Decision latency. The cost nobody measures.

The time from a question being asked to someone acting on an answer they trust. Not query speed. Almost nobody measures it, and every leader feels it weekly. Try one question below.

Time five questions with the free template What decision latency is

People-hours spent on this one question, per quarter

195

Is it acted on before it’s asked again?

No. By the time it’s acted on, the question has already come round again.

Example values. Replace them with one of your own recurring questions.

03 The four layers

Four layers. One fixed order.

Skip a layer and you get a good demo, then a quiet rollback. The first two need a shared document and a named owner, and nothing you have to buy.

  1. 01

    Clean sources

    Governed foundation

    Every source is connected, checked and owned. Any number can be traced back to where it came from.

    Without it

    Every answer is contestable, so no answer is trusted.

  2. 02

    One definition, one owner

    Semantic contract

    Metric definitions written down as contracts, each with a named owner allowed to refuse a change.

    Without it

    Two dashboards disagree and the meeting becomes an argument about definitions.

  3. 03

    Answers where people ask

    Agentic resolution

    Questions become sourced answers, delivered into Slack, Teams or email, where work already happens.

    Without it

    The answer exists but takes three weeks and an analyst’s calendar.

  4. 04

    A person signs off

    Human interpretation

    A named person reads the result, frames the decision, and will be in the room if it is wrong.

    Without it

    A correct number with no interpretation is still not a decision.

04 Layer two, up close

One definition, one owner. Written down.

A metric contract is a short written record: what the number means, who owns it, which source may compute it, and what changed last. It’s the cheapest layer to build and the one most companies skip.

The 12-check metric contract checklist

Enterprise pipeline

Metric contract · v3

Definition
Open opportunities of $50k+ annual contract value in stages 2–5, excluding renewals, in USD at the booking-date rate.
Owner
Head of RevOps. Can refuse a change.
Source
Salesforce → fct_pipeline. No other table may compute it.
Last change
12 Aug 2026 · renewals excluded · approved by the CFO
Used in
Monday pipeline review · board pack · sales compensation
An example contract. Every number your leadership team argues about needs one of these, and one person whose name is on it.

05 What changes

Latency drops. Three things change.

  1. 1The meeting

    Starts from a number

    Not from an argument about the number. Forty minutes a week go back to the decision instead of the definition.

  2. 2The backlog

    Questions stop dying in threads

    Recurring questions become paths, not tickets. The data team stops settling folklore at 11pm.

  3. 3The AI

    Can be let into Slack

    An assistant that reads from one written definition can be trusted. One that reads from four versions of pipeline can’t.

We already have a modern stack. Isn’t that enough?

Then access isn’t your constraint. Time five recurring questions. If any of them still takes more than a day to reach a number someone will stand behind, the stack isn’t the system. The path from question to answer is.

Isn’t this a culture problem?

Partly. Culture is what people do when the definition is missing. Write it down and name an owner, and much of the argument goes away, because nobody is asked to trust a number with three parents.

Can’t AI fix this?

Not on its own. An assistant on top of contested definitions produces contested answers faster. It belongs at layer three, once the sources are clean and every metric has an owner.

Are we too small for this?

Maybe. If one analyst and one source still produce one number, you have a capacity problem, not an entropy problem. It starts to matter when two teams can each defend a different number.

Why call it entropy?

Entropy is the number of states a system can be in. Every extra dashboard and every extra definition adds one. More states, more disagreement about what is true. Zerentro is short for zero entropy.

Is this a pitch for Zerentro?

Zerentro is the company I built to run this for teams that want it run for them. You can use everything on this page without buying anything, and if it doesn’t fit your situation I will say so on a call.

Work with me

Bring one number. Find the missing layer.

A free 30-minute call with Yagnesh. You'll find out if the problem is the data, the definition or who owns it.

Book a 30-minute callGet the free template
Book a 30-minute call