AI assistant: what the numbers say and what they do not
Working through tables and metrics: how to count, what to compare with and why a pretty correlation proves nothing.
What Data Analyst returns
The shape of the answer is fixed in advance: gaps, duplicates, outliers, methodology changes, seasonality. Two conversations about the same task therefore end in the same form: the answer can be compared with the previous one and passed on without asking for a rewrite.
“Check data fitness before conclusions: gaps, duplicates, outliers, methodology changes, seasonality, survivorship.”
What Data Analyst will not do
The role has 7 rules, one of them an outright prohibition. We quote them as written rather than paraphrase: promising “it can do anything” is cheaper, but then people meet the limit only after paying for a message. Here it is visible beforehand, and a neighbouring role can be picked straight away.
“Never invent numbers, data rows, industry benchmarks or study results; if data is missing, say exactly what is missing and how to obtain it.”
How Data Analyst runs the review
The rules describe the role as “data analyst”. Below is how it holds the review, in the same words it is given before your first message — and manner is exactly what separates one companion from another.
“Compute step by step and show the formula, not just the result.”
“Separate observation, hypothesis and conclusion; never present correlation as causation and name alternative explanations explicitly.”
“Finish with a one-sentence answer to the original question and, next to it, what that answer does not prove.”
What a message to the “What the numbers say and what they do not” role costs
You are charged per message sent, not per month of access: a role needed once a quarter costs nothing in between, and it needs no line of its own in the budget.
What Data Analyst flags for checking
The danger is not a gap in knowledge but a confident tone over a guess: people act on such answers, and the cost of the mistake lands on the person, not the program. So the rules say separately what to flag, what to warn about and when to ask for a source.
“First establish the question behind the numbers, the data source, period, units and how the data was collected: without this any figure is decoration.”
“Flag small samples and short periods explicitly and do not build confident forecasts on them.”
Three questions Data Analyst offers first
A click opens the chat with the message already typed. These are not our examples but the role’s own suggestion: the same three lines stand in the empty conversation window.
Questions about this assistant
Quick answers
How Data Analyst differs from a plain chat
In a plain chat the framing is on you: the order of questions, the limits, the requirements for the answer — from scratch in every new window. Here it is already written: 7 rules, 1 prohibition of them, all quoted on this page. You never have to restate them.
✓ Rules: 7, prohibitions: 1
✓ Catalogue shelf: Code and technology
✓ A program answers, not a person: it replaces no doctor, lawyer or therapist
Data Analyst and its neighbours on the “Code and technology” shelf
This shelf holds 7 roles: neighbouring tasks, different rules — and every one of them is quoted in full on its own page.
Generonium pricingCredits work across every service
Лайт
30-day subscription
- Granted on payment and every 30 days
- Auto-renewal can be turned off in your account
- Credits work with any Generonium service
- Each service shows its exact price before launch
Про
30-day subscription
- Granted on payment and every 30 days
- Auto-renewal can be turned off in your account
- Credits work with any Generonium service
- Each service shows its exact price before launch
Макс
30-day subscription
- Granted on payment and every 30 days
- Auto-renewal can be turned off in your account
- Credits work with any Generonium service
- Each service shows its exact price before launch