An insight is a written observation about your business, generated from your connected data.
Where a key performance indicator gives you a number and an alert fires on a threshold, an insight is the sentence in between: what these figures, together, appear to mean.
Every insight shows its work
Each one carries the records it was drawn from, and you can open them.
This is the part that makes an insight usable rather than merely interesting. An observation you cannot check is an opinion, and an opinion about your own business from software you have owned for three weeks is worth very little. An observation with the transactions attached is a starting point for a decision.
See the evidence chain.
The records attached are the ones that actually drive the finding, not a fixed category chosen in advance. Where profit and cash have separated, that gap is receivables for one business and unpaid bills for another, and the evidence follows whichever it is. A finding whose evidence list is empty would be the "your numbers are off, go look at them" non-answer these exist to replace.
How to read one
As a hypothesis rather than a conclusion.
An insight sees your data and nothing else. It does not know that the large March expense was a one-off equipment purchase, that a customer left because they moved rather than because of price, or that the quiet fortnight was a holiday.
Read it, check the records behind it, and then decide — including deciding that it has missed something. The ones that feel wrong are worth two minutes rather than dismissal, because a meaningful share of them turn out to be surfacing something true and unnoticed.
Insights, alerts and briefs
An alert fires when a number crosses a line. Mechanical, checkable, and specific.
An insight is an interpretation across several numbers, which makes it more useful and less certain at the same time.
A brief collects both into a period summary you can read end to end.
What they are not
Not advice from an advisor, and not a substitute for one. An insight is what the data appears to say. What to do about your particular business, with its particular history and constraints, is a conversation.
When they appear
Insights are generated each morning, so a workspace opened during the working day is reading that day's set rather than something produced live as the page loads.
A fuller analysis runs at the start of each month, looking across the month that closed rather than the last few days.
Why a day can produce none
Nothing changed enough to be worth saying. An insight is generated when the underlying figures support one, not on a quota, and a quiet week producing nothing is the system working rather than failing.
New data also has to have arrived. A connection that stopped delivering leaves the figures where they were, and unchanged figures produce no new observation — see data freshness and a connection stopped bringing data in.
Where the underlying data is too thin to support a claim, it is withheld rather than hedged — see how confident a number is.