Signal vs noise in decision data: how a decision engine measures it

What are signal and noise in decision data?

In engineering, the signal-to-noise ratio is the ratio of signal power to noise power, and a ratio higher than 1:1 indicates more signal than noise (en.wikipedia.org, Signal-to-noise ratio). The term is also used metaphorically for the ratio of useful information to irrelevant data (en.wikipedia.org, Signal-to-noise ratio). Decision work borrows this meaning directly.

In decision data, a signal is any fact that helps rate a dimension of the situation: a dated contract, a measured churn rate, an audited statement. Noise is everything else that arrives with it, such as duplicated reports, undefined metrics, scattered sources and complexity that hides the facts. The ratio of the two tells a team how much of its data pays for itself.

How does a decision engine measure signal vs noise?

EQOS Asia reports a dedicated measure in its case studies. The WeWork page labels it X(S), signal and noise, and reads it as the signal-to-noise ratio, with a value of 0.202 (EQOS Asia WeWork case study, as of October 2026). Other pages call the same measure an intelligibility ratio. A value above 1 means that the data is more informative than confusing, and a value well below 1 means the reverse.

The reading is practical. A low X(S) says that the information exists but sits in a form that is hard to use, because it is scattered, complex or opaque. A high X(S) says that the facts are readable. The measure sits beside the nine dimensions, so a committee can ask two questions in turn: how aligned is the system, and how readable is the data about it?

What do published X(S) values show about signal in decision data?

The values span a wide range. The 2008 crisis page, built on data public in 2006 and 2007, reports X(S) = 0.185 and states that noise exceeds signal by 5 times. The WeWork page reports 0.202, and the Theranos page reports 0.012, which it labels as readability far below 1. At the other end, the dollar page reports 1.029, and the climate page 0.776 (EQOS Asia case study pages, as of October 2026).

The pattern is useful. Retrospectives on systems that later failed show a low X(S), because the data existed but could not be read with confidence. Current-market readings sit higher. The values are EQOS Asia readings of public data, and they describe the data about a system, not a judgement of the people involved. The table lists five of them.

Which sources add noise to decision data?

Four sources of noise appear again and again. Scattered data, held in many places with different formats. Undefined metrics, where two teams use one word for two things. Complexity, such as layered structures that hide who bears a risk. Volume without relevance, where hundreds of figures arrive and a handful bear on a dimension. The 2008 page describes data that exists but is scattered, with massive complexity (EQOS Asia 2008 case study, as of October 2026).

Each source has a remedy. Centralise the data in one sheet. Define each metric with a unit and a source. Draw the structure on one page. Keep only the inputs that rate a dimension. Doing these four things in order raises the ratio quickly, and the data sheet from the data quality guide helps record each step.

How can you raise the signal in your decision data?

Start by cutting, not adding. Remove inputs that do not rate a dimension, merge duplicates, and archive old reports. A shorter list makes the remaining facts easier to see. Then strengthen the key inputs: replace a rough estimate with a measured figure, and add a second source for any number that carries a large weight.

Soft dimensions benefit from structure. Targeted interviews with the same questions for every person yield comparable answers, and the EQOS Asia method uses structured interviews that last 2 hours at the Standard level and 3 days at the First level (EQOS Asia method page, as of October 2026). A comparable answer carries more signal than a free-form comment of the same length.

Why does the signal-to-noise ratio matter when AI reads decision data?

The AI case study recommends measuring X(S) before deploying AI. It advises an audit of the internal signal and noise ratio, and it states that if X(S) is below 0.5, AI will amplify noise instead of signal. The concrete checks are training-data quality, potential biases and edge-case coverage (EQOS Asia AI case study, as of October 2026).

The EQOS Asia method page draws a related line on its own process. Scores, indices and numerical verdicts come exclusively from the deterministic Python engine, never from a generative model, while research and drafting are AI-assisted with human review (EQOS Asia method page, as of October 2026). Clean inputs and a deterministic core keep the verdict reproducible.

How do you estimate a signal-to-noise ratio for your own decision?

Use a simple proxy in four steps. Count your inputs. Classify each as signal if it rates a dimension and traces to a dated source, and as noise otherwise. Divide the signal count by the noise count. Prune and repeat. With 30 inputs, of which 20 are signal and 10 are noise, the ratio is 2.0. The proxy is an illustration and differs from the EQOS Asia X(S).

A 30-minute pre-diagnostic by video, scheduled in Singapore time (SGT), sorts the material you already have at no commitment, and a systematic NDA protects it (EQOS Asia contact page, as of October 2026). The pricing page lists the tiers for a fuller study, from a Reading at US$358 to a Mapping that integrates 100% of your data (EQOS Asia pricing page, as of October 2026).

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile