How Often Rank Updates, and What That Hides

Rank is recalculated on a schedule nobody publishes, and the schedule determines what your data can and cannot see.

What is observable

Top ranks update frequently. Products in the upper reaches of a catalogue refresh often — commonly hourly, sometimes faster during peak periods.

Deep ranks update slowly. Products far down the catalogue may hold the same value for many hours or a full day, because there is no new sales activity to reorder them against.

The cadence is not uniform across categories or marketplaces, and it changes.

None of this is documented. Everything here is inference from observation, and it should be treated as such.

Why it matters for collection

Sampling faster than the source updates adds noise, not resolution. Polling every fifteen minutes when the value refreshes hourly gives you four identical readings and a false impression of stability.

Sampling slower than events loses them. A promotional spike that resolves within a day is invisible to daily sampling at a fixed hour.

A reasonable default is hourly for actively tracked products and daily for a broad watch list, with the understanding that daily sampling can only detect changes lasting more than a day.

Sample at a consistent time if sampling daily. Rank has an intraday pattern, and readings taken at varying hours introduce a variation that looks like signal.

The intraday pattern

Rank moves through the day in ways that reflect buying behaviour rather than anything about the product.

Sales concentrate in particular hours by marketplace and by region. A product's rank is systematically better after a peak buying period and worse after a quiet one.

Daily samples taken at 09:00 and at 21:00 produce different series for the same product.

Cross-timezone comparisons inherit this. Tracking a US and a UK marketplace at the same absolute time compares different points in each market's day.

The practical rule: fix your sampling time, record it, and never compare series captured at different times of day without adjusting.

What short-lived events look like

A promotion running for six hours produces a sharp rank improvement followed by decay. Hourly sampling captures the shape. Daily sampling captures either a modest improvement or nothing, depending on when the sample fell.

A stock-out lasting a day shows as a decay with a recovery. Daily sampling may show only the recovery, which looks like a launch.

A viral mention produces a sharp spike and a long decay tail. The peak is the least informative part; the decay rate tells you whether the interest converted into sustained sales.

If you need to detect events shorter than a day, you need sub-daily sampling. There is no analytical technique that recovers a signal the sampling missed.

Gaps and their effects

Collection fails. Rate limits, network problems, page structure changes, blocks.

A gap is not a zero. Treating a missing observation as no sales, or interpolating across it as if the value moved linearly, both introduce artefacts.

Record gaps explicitly and exclude them from calculations rather than filling them.

Check gap rates per source and per period. A rising gap rate is the earliest sign that a collection method is breaking, and it degrades quietly.

Choosing a cadence

Competitive tracking of a small watch list: hourly, if you can sustain it. The extra resolution matters for detecting promotions and stock-outs.

Broad market monitoring of many products: daily at a fixed hour. The purpose is trend, not events.

Your own products during a launch or a campaign: hourly, for the duration.

Historical baseline building: daily is sufficient and much cheaper.

Do not collect more than you will analyse. Hourly data for two thousand products is a storage and processing commitment that produces a dataset nobody opens, and the daily aggregate would have answered the question.

The resolution question to settle first

Before designing collection, decide the shortest event you need to detect.

A day-long promotion? Hourly.

A week-long trend? Daily.

A quarterly market shift? Weekly is enough, and daily gives you the option of retrospective detail.

That decision determines the cadence, the storage, the cost and the complexity, and making it explicitly prevents the common outcome of collecting at maximum frequency because it seemed safer.

Establishing the cadence empirically

Rather than guessing, measure how often the source actually changes.

Poll one product every fifteen minutes for two days.

Count distinct values. If the rank changed four times in a day, the effective cadence is roughly six hours, whatever the marketplace implies.

Repeat for a fast product and a slow one. They will differ substantially, and the difference determines whether one cadence serves your whole watch list.

Repeat per marketplace.

Then set your cadence at or slightly above the observed refresh rate. Faster buys nothing but load, and slower loses events.

Recheck annually. Marketplaces change their processing and the cadence moves without announcement, which shows up as a change in your value-change-rate monitor if you have one.