What a certified analytical answer actually means.

The word "certified" appears in a lot of analytics marketing. It is used to mean accurate, reliable, trustworthy, verified — all of which are vague reassurances that any vendor can make without commitment. What we mean by certified is specific, architectural, and falsifiable. Either an answer meets these criteria or it does not.

The three components of a certified answer

A certified analytical answer has three properties that must all be present. A result that has one or two of them is better than nothing — but it is not certified.

1. A fixed, pre-validated formula

The formula used to compute the answer was defined and validated before the question was asked. It did not emerge from the question. It was not generated by an AI interpreting the question. It existed as a permanent, documented methodology that the question was resolved against.

This is the opposite of how AI analytics tools work. When you ask an AI analytics tool "what is our customer retention rate," the AI decides — in that moment, for that query — what retention means, which columns define it, and how to compute it. In a certified system, retention has a permanent definition that was established once and applies to every query that resolves to it.

2. A traceable computation

The answer can be traced back to its source — the specific rows included, the specific columns used, the specific formula applied, and the specific result produced. Not as a log file that requires a database administrator to interpret. As a plain English explanation that an executive can read and understand in the room where the number is being challenged.

3. Reproducible results

The same question, run on the same data, produces the same answer — regardless of when it is run, who runs it, or how it is phrased. This is the criterion that most AI analytics tools fail, because each query generates a fresh computation that may differ subtly from the last.

If you cannot get the same answer twice, you do not have an answer. You have a sample.

Why this matters in practice

The value of certification becomes concrete in three situations that every organisation with significant data eventually faces.

The board meeting challenge. A board member questions a number. In a non-certified system, the best response is "our analytics tool produced it" — which invites the follow-up: "can we trust the analytics tool?" In a certified system, the response is immediate and specific: the metric definition, the formula, the rows included, the computation. The challenge is answered in the room.

The regulatory audit. A regulator asks how a reported figure was calculated. In a non-certified system, this requires reconstructing the analysis from scratch — if it can be reconstructed at all. In a certified system, the audit trail already exists. It was generated automatically when the answer was produced.

The conflicting numbers problem. Two departments report different figures for the same metric. In a non-certified system, resolving this requires determining whose methodology is correct — a political as much as an analytical problem. In a certified system, there is one certified definition and one certified answer. The conflict does not arise.

What certification is not

Certification is not accuracy. A certified answer can be wrong if the underlying data is wrong. Certification guarantees that the formula was applied correctly and consistently — not that the data itself is free of errors. That distinction matters: certification is a governance property, not a data quality property.

Certification is also not complexity. The certified formula for customer retention may be simple. What makes it certified is not its sophistication but its permanence — the fact that it was defined, validated, and locked before any result was produced against it.

The standard worth demanding

Any organisation that uses analytics to make decisions that matter — financial decisions, operational decisions, clinical decisions, regulatory submissions — should demand certified answers as a baseline. Not because non-certified answers are always wrong, but because there is no way to know when they are right and when they are not.

Certification makes the answer checkable. And an answer that cannot be checked is an answer that cannot be trusted.