Monthly Demand

Monthly Demand estimates how many units a product sells in a month. It appears on the product card as Monthly demand, and it comes from one of two places: a figure Amazon itself reports, or — far more often — our own estimate, shown as a range.

Two Sources, Clearly Labelled

| What you see | Where it came from | |---|---| | 800+ units/mo with an Amazon-reported tag | Amazon publishes a sales figure for this listing. This is their number, not ours. | | 120–340 units/mo with a High / Medium / Low tag | Our estimate. The range is the reading; there is no single hidden number behind it. | | No velocity data | Neither source produced anything usable. |

When Amazon reports a figure and our model has a read, the Amazon number leads and our range is shown beneath it as a Model · line. If Amazon's figure falls inside our range, it is marked with a check — an at-a-glance signal that the two agree.

Why It Is a Range

This is the part worth understanding, because it changes how you should use the number.

We are not forecasting what this product will sell. We are describing what products like it actually sold. For any product we know its category, roughly where it sits in the sales rankings, and how often its rank has been dropping. Those three facts place it in a group — a cell — alongside every other product we have measured with the same profile. The range is the middle half of that group: a quarter of them sold less than the low end, a quarter sold more than the high end, and half landed in between.

So 120–340 units/mo does not mean "this product sells somewhere between 120 and 340". It means "products in this position typically sell between 120 and 340, and this is one of them." That is a claim about a population, not a prediction about an individual. A product can and will land outside the range — one in four does, by construction.

The width tells you something too. A narrow range means products in that position behave consistently. A wide one means they do not, and no amount of modelling can make that product's outcome predictable.

How We Calculate It

1. We place the product in a cell using its category, its sales rank band, and its rank-drops band — how frequently its rank has been falling, which is the closest public proxy for sales frequency.
2. We check whether that cell is a trustworthy anchor. It qualifies when it holds at least 5 distinct products and fewer than half of them sit against the measurement floor. A cell mostly made of floor readings cannot tell us anything, because we would only be measuring the limits of what Amazon publishes.
3. Where the cell is trustworthy, it is the reading. The range is that cell's observed middle half, taken directly from the products in it.
4. Where the cell is floored or too thin, we discard it rather than lean on a censored number, and rebuild the estimate from the product's own activity: its rank drops and its review count, valued at rates measured for its specific category. Each rank drop mechanically implies at least one sale, so this path has a real floor rather than a guessed one.
5. Where only review data survives, we report "Likely stalled" rather than a number. Reviews lag sales badly, so a review-only signal describes the past, not the present.

Displayed figures are always snapped downward to a round number, so the estimate errs toward understating rather than overstating. Very high readings are capped at 2K+.

Confidence

| Rating | What It Means | |--------|---------------| | High | The cell is a trustworthy anchor, and the product's own rank-drop and review activity independently agree with it and with each other. Three signals pointing the same way. | | Medium | Either the cell is sound but the product's own activity does not corroborate it, or the estimate was rebuilt from rank drops because the cell was unusable. | | Low | Thin evidence. Treat the range as a rough bracket only. |

A trustworthy cell never produces a Low rating — if the population is solid, the reading is at least Medium even when the product's own signals disagree. Disagreement lowers the confidence label; it never moves the range.

Limitations & Caveats

• It describes a population, not this product. Everything above follows from that. A product with unusual demand for its rank position will be misestimated, and nothing in the data would reveal that.
• Variation families distort review counts. Where a listing rolls up reviews from many sizes or colours, its review count overstates its own performance. We detect this and drop reviews from the calculation when it happens, but the screen is not perfect.
• Rank drops are a proxy. A rank drop usually means a sale, but rank moves for other reasons too, and not every sale produces a visible drop.
• Category coverage varies. The estimate rests on measurements taken per category. Categories where we have sampled less deeply produce wider ranges and lower confidence — which is the honest outcome, not a defect.
• Very deep ranks fall out of scope. Beyond a certain rank there is no usable signal and no estimate is offered.
• Seasonal products are estimated against their current position. A Christmas product measured in July sits in a low-rank cell and will be estimated accordingly. Check Seasonality before reading a single month's figure as typical.

Related Metrics

• Demand Stability: How consistent the sales rank has been week to week. Monthly Demand tells you how much; Demand Stability tells you how reliably.
• Sell Price: What those units have been selling for. Demand times price is the revenue picture.
• Stock Depth: How the demand estimate translates into weeks of supply across sellers.
• Seasonality: Whether this month's demand is representative of the year.