Two engines for two adjacent questions. Both share the studio's shape: scope on the left, the live result on the right, a sentence describing what you are looking at.
Both are single-question tools
Neither wants a second dimension. That is the whole design.
An analysis with two breakdowns is a pivot, and a correlation with a third measure is three correlations. When you feel the pull to add one more thing, that is the signal to move to the pivot builder rather than to keep going here.
Keeping these two narrow is what makes them readable by somebody who did not build them.
Choosing between them
The question tells you which.
"How much, broken down by X" is an analysis.
"Does X move with Y" is a correlation.
If you find yourself building a correlation to answer the first, you have made it harder than it needed to be. If you find yourself eyeballing two analyses side by side to answer the second, the correlation engine exists for exactly that.
Analysis
One measure, one dimension, one period. The simplest useful question in the whole surface.
Revenue by month. Tickets by team. Average deal size by source.
Reach for this before a pivot. A great many pivots are really an analysis with an unnecessary second dimension, and the single-dimension version is easier to read and harder to misinterpret.
What you set
The measure, what to break it down by, and the window. An operator selects how the measure is treated.
That is the entire configuration, which is the point: the question is simple, so the surface is.
Operators
An operator decides how the measure is treated across the dimension: the plain value, a running total, a change against the previous period, a share of the whole.
Worth knowing because the same measure and the same dimension answer quite different questions depending on it. Revenue by month is a shape; revenue by month as change against last month is a judgement.
Reading one
Look at the shape before the numbers. Rising, falling, flat, spiky. The shape is usually the answer and the numbers are the evidence.
Check the period against the cycle. A monthly view of something with a quarterly rhythm looks chaotic and is not.
Be careful with the last bucket. A part-month always looks like a collapse, and it is the single most common false alarm in any analytics tool.
Correlation
Whether two measures move together.
Marketing spend against leads. Response time against satisfaction. Headcount against output.
What you set
Two measures, a window, and a scope narrowing which records are considered.
The engine handles the arithmetic. What it cannot handle is whether the pairing makes sense, which is entirely yours.
The warning that has to come first
Correlation is not cause, and this surface will happily show you a strong relationship between two entirely unrelated things.
That is not a flaw in the tool, it is what correlation is. Two measures that both grow with the size of your business will correlate beautifully and tell you nothing.
The useful discipline: form the hypothesis before you look. A correlation you went looking for is evidence. A correlation you found by trying pairs is a coincidence you have not identified yet.
What a strong relationship is worth
Almost nothing on its own, and a great deal alongside a reason.
The useful sequence is: you have a theory, the correlation supports or kills it, and you act on the theory rather than on the correlation. A relationship with no mechanism behind it is a prompt to go and find one, not a finding.
Reading one
Look at the scatter, not just the strength. A single extreme record can create a strong relationship out of nothing, and it is obvious visually and invisible in a number.
Check the direction makes sense. If more spend correlates with fewer leads, something is wrong with a definition rather than with your marketing.
Prefer a longer window. Short windows produce strong relationships from very little.
When it is genuinely useful
Confirming something you suspect. You believe response time affects satisfaction; this tells you whether the data agrees.
Killing an assumption. Frequently the more valuable outcome, and the one nobody looks for.
Sizing a relationship you already accept. Not whether, but how much.
Windows matter more than usual
A correlation over a short window will find relationships in noise, reliably.
Prefer as long a window as your data supports, and be sceptical of anything that only appears in a narrow one. A relationship that holds across a year and disappears across a quarter is telling you about the quarter, not the relationship.
They save differently
Worth saying plainly because it affects how you use them.
An analysis is a standing question. Revenue by month is as valid next quarter as it is today, so it belongs on a board and improves with a longer history behind it.
A correlation is an investigation. It answers something once, and the answer either changes what you do or it does not. Left on a board it becomes a claim nobody re-examines, and the older it gets the more authoritative it looks.
The rule: analyses live on boards, correlations live in whatever decision they informed.
Neither explains anything
Both engines describe the data and neither knows your business.
An analysis showing revenue falling has not told you why. A correlation between two measures has not told you which causes which, or whether a third thing causes both.
The value is in narrowing where to look, and the answer is almost always somewhere else: in the funnel, in the records, or in a conversation with whoever owns the work.
Recommendations
The workbench can suggest correlations worth looking at, which is a good starting point and a poor finishing one.
Treat a suggested correlation as a question rather than a finding. The suggestion knows the numbers move together; it knows nothing about your business.
Saving either
Both save as first-class widgets and sit on boards like anything else.
Analyses age well and belong on boards. Correlations mostly do not: they answer a question once, and a correlation left on a board for a year becomes a claim nobody re-examines.
Did this answer your question?
No, ask a person