Reading a Rank Chart

Rank charts mislead by default, because a better rank is a smaller number and a linear scale compresses the part that matters.

Presentation rules

Invert the axis. Better rank at the top. Without this, every improvement looks like a decline and every audience misreads it at least once.

Use a logarithmic scale. Rank is a compressed measure — the difference between 100 and 200 is enormous, between 100,000 and 100,100 is nothing. Linear axes make the long tail look dramatic and the top look flat, which is exactly backwards.

Label the axis clearly, including the direction. "Sales rank (lower is better)" on the axis itself.

Mark gaps rather than interpolating. A line drawn across missing data asserts values that were never observed.

Show the sampling cadence. Hourly and daily series look different and should be labelled.

Plot the comparison basket on the same chart. A single line invites the misreading that its movement is about the product.

The shapes and what they usually mean

A sawtooth — sharp improvements followed by decay. Individual sales, separable. A low-velocity product. Counting the teeth gives an approximate sales count.

A smooth line — sales frequent enough to overlap. Higher velocity.

A single sharp spike with decay — an event. Promotion, mention, price cut, feature.

A step change that persists — something structural. A price change, a listing improvement, a category change, or a genuine shift in demand.

A slow drift downward with no events — declining velocity, or a growing category pushing everything down. Check the basket.

A decay followed by a sharp recovery to the previous level — almost always a stock-out. Genuine decline does not rebound cleanly.

Sudden movement affecting many products at once — a market event, a seasonal transition, or a category restructure. Not about any individual product.

A vertical discontinuity — usually a data problem: a collection change, a category switch, or a different identifier being tracked.

Annotating

A rank chart without annotations invites invention.

Mark known events: your promotions, price changes, launches, stock-outs, publicity.

Mark data events: collection gaps, method changes, category changes.

Mark market events: seasonal peaks, competitor launches you know about.

An annotated chart answers "why" for the events you know and makes the unexplained ones visible as questions, which is the useful state.

Smoothing

Raw hourly data is noisy and the noise is not interesting.

A moving average over a few days reveals trend and destroys events. Show both, or show raw with a trend line over it.

Never smooth before looking at the raw series. The events are the most informative part and smoothing removes them.

State the window on any smoothed series.

Comparing several products

Same chart, same axis, same period. Separate charts side by side invite comparison of differently-scaled axes.

Consider plotting relative to a basket median rather than absolute rank, which removes market-wide movement and isolates the part about each product.

Limit to a handful of lines. Beyond about five, a rank chart becomes unreadable and a table is better.

The presentation that survives scrutiny

For anything going to stakeholders:

Inverted log axis, labelled. Gaps shown as gaps. A comparison basket. Annotations for known events. The sampling cadence stated. The observation window stated.

Six properties. A chart with all of them can be argued about on its merits. A chart missing the first two will be misread by everyone in the room, and the misreading will be in your favour or against it at random.

Charts for audiences who do not work with this data

Presenting to people unfamiliar with rank requires two additional decisions.

Consider inverting the metric rather than the axis. Plotting "rank position, better upward" confuses fewer people than an inverted numeric axis, and some audiences will misread the axis no matter how it is labelled.

Or plot the estimated units instead, with the range shown as a band. Units are intuitive and the band communicates the uncertainty honestly. The cost is that you have introduced a conversion basis and its error.

Never present both in the same document without explaining their relationship, which invites the audience to check whether they agree and conclude the analysis is inconsistent when they do not.

Annotate heavily. An audience that cannot read the shape needs the events labelled.

Lead with the sentence, not the chart. The chart supports a conclusion that has already been stated in words.