Data quality for a decision engine: what to check first

What does data quality mean for a decision engine?

Data quality refers to the condition of data, and data is of high quality when it is fit for its intended uses in operations, decision making and planning (en.wikipedia.org, Data quality). ISO 8000 is an international standard for data quality (en.wikipedia.org, Data quality). For a decision engine, the intended use is narrow and clear: each input must help rate one dimension of the situation.

This framing keeps the effort proportionate. A figure can be perfectly accurate and still add nothing, if it does not bear on any dimension being scored. Data quality in an engine therefore combines the general qualities of the data with a specific question: does this input tell us something about the dimension we are rating?

Which data quality checks come first in a decision engine?

Run five checks in order. Relevance first, because an irrelevant input wastes every later check. Accuracy second, since a wrong figure misleads. Timeliness third, because a correct but outdated figure describes a situation that has gone. Completeness fourth, to find the dimensions with no evidence. Consistency last, so that units and scales match across inputs.

The listed qualities come from the general data quality literature, which names accuracy, completeness, consistency, timeliness, relevance and uniqueness among its dimensions (en.wikipedia.org, Data quality). Putting relevance first saves time. A team that removes irrelevant inputs at the start has a shorter list to verify and a cleaner record to show a reviewer.

How do you check the accuracy of an input in a decision engine?

Trace each figure to its origin. A revenue number should lead to a ledger, a headcount to an HR export, a market share to a named report. Then look for a second source that should agree: a bank statement beside a ledger, a payroll file beside a headcount. When two sources agree, confidence rises. When they differ, record both and ask which one the decision should rest on.

Recompute where you can. Totals should equal their parts, and ratios should match their numerators and denominators. A short log of checks, with the name of the person who ran each one and the date, turns accuracy from a feeling into a record. The log also helps when a score is questioned later, because the reviewer can see what was verified.

How current must the data of a decision engine be?

Current enough for the decision at hand. A regional hub choice depends on costs and regulations that change within months, so a figure from three years ago deserves a refresh before it enters the score. Stamp every input with the day it was observed and the day it was loaded, and set a maximum age for each group of inputs.

For ongoing decisions, refresh on a calendar. The EQOS Asia model is reusable monthly through Sentinel, which re-measures the 9 dimensions and sends alerts when a tipping threshold gets close (EQOS Asia method and pricing pages, as of October 2026). A monthly rhythm shows drift early: an input that ages without an update flags itself in the data sheet.

How should a decision engine handle gaps in its data?

A gap is a dimension with no evidence behind it, and an engine that scores it anyway rests on a guess. List the gaps openly and sort them by how much they could move the score. Fill the large ones first with a document, an interview or a measurement. For the rest, record an estimate with a range and mark it for review.

Gaps tend to cluster around the soft dimensions, such as trust or transparency, where no database exists. The EQOS Asia method addresses them with structured interviews, which last 2 hours at the Standard level and 3 days at the First level, and with collected documents (EQOS Asia method page, as of October 2026). Any format is accepted, digital or physical.

How does EQOS Asia protect data quality in its workflow?

The standard workflow reserves 2 to 5 days for structured collection, with internal documents, targeted interviews and existing data, in an encrypted isolated environment. Documents are collected and vectorized, and the computation runs through the deterministic Codex core, which validates results through a systematic ethical safeguard (EQOS Asia method page, as of October 2026).

The method page also states where AI assists. Scores and numerical verdicts come exclusively from the deterministic Python engine, while research, drafting and source-checking are AI-assisted with systematic human review before delivery. The pricing page notes that from the Study tier upwards, clients can send internal and external documents, and that the more data is provided, the more accurate the model (EQOS Asia pricing page, as of October 2026).

How do you keep a one-page data sheet for a decision engine?

Give each input one line with five fields: name, unit, source, date and owner. Add a sixth field for the dimension it rates, and a seventh for its check status. The sheet fits on one page for a decision with 20 to 30 inputs. A reviewer who opens it can see which inputs were checked, which are pending, and who to ask.

Bring the sheet, or whatever you already have, to the first conversation. A 30-minute pre-diagnostic by video, scheduled in Singapore time (SGT), sorts your material into the groups of the engine at no commitment (EQOS Asia contact page, as of October 2026). The pricing page shows which tier fits the depth of data you can supply.

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