The honest empty state

Every analytics product is judged twice. Once when it has data, which is when people notice the design, and once when it does not, which is when they decide whether to trust it.

The second is decided by how carefully you distinguish four different kinds of nothing.

Not connected

The source has never been connected. Nothing is broken and nothing is missing.

The right treatment names the source, says it is not connected, and says what connecting it would add. That last part turns an empty panel into the most effective onboarding surface in the product, because it appears exactly where somebody is already looking for the thing it offers.

If the source is one a customer connects themselves, put the link there. If it requires someone else, say who and offer the request.

Connected, no data yet

The connection works and there is nothing to show. Either the first sync has not run, or the source genuinely has nothing.

Distinguish those two if you can, because they need different messages. "Connected, first sync runs overnight" is reassuring. "Connected, no shows added yet" is a prompt.

Genuinely zero

Something happened and the answer is zero. No tickets sold this week. No press coverage in the last seven days.

This must look different from unknown, and it is the most commonly conflated pair. A blank panel where a zero belongs reads as broken, and a zero where a blank belongs is a false statement.

Our rule is that a real zero is displayed as a number, in the same style as any other number, because it is a measurement. Unknown gets a dash and an explanation.

There is one exception worth making deliberately: a metric whose zero is meaningless at low volume is better hidden than shown. We hide one recognition-based figure entirely when it is zero, because for most projects a zero there says nothing about the project and invites a wrong conclusion.

Partial

The hardest one. Some data exists, and less than the question needs.

Three cases, all with the same answer: say what you have.

History shorter than the window. A 90 day figure for a project with 12 days of history. Show the gain since tracking began, with the start date.

Missing days inside a series. A gap, drawn as a gap.

Some records excluded. Two of six shows have no capacity recorded, so the sell-through covers four. Say that, next to the number.

The interpolation problem

The single most damaging shortcut in analytics interfaces is drawing a line through missing data.

It is understandable. A chart with gaps looks broken, and a smooth line looks finished. The problem is that a reader cannot tell the difference between a measurement and a guess, so the guess inherits the credibility of everything around it.

Three rules we hold to.

Days with no reading render as gaps, never as zero and never as a straight line to the next point.

Where a value genuinely was spread across a span, because a data gap was reconciled later, it is flagged as estimated in the data and annotated in the interface, so the estimate is visible as one.

A cumulative line stops permanently at the first missing past day, because resuming after a gap understates every later point by the size of the gap while looking entirely plausible.

Why this is worth the effort

The argument against all of this is that it makes a product look less capable. Gaps, dashes, explanations of what is missing.

The counter-argument is what happens the first time somebody notices a seam you did not declare. A user who spots that a chart smoothed over a broken week stops trusting every chart, including the correct ones, and that is a much larger loss than a slightly emptier screen.

A dashboard that is visibly careful about what it does not know is the only kind whose confident statements are worth anything.