Going from 10 to 20 is a 100% jump that means far less than a 10% jump on a larger base

What it is

Going from 10 to 20 is "+100%"; going from 1,000,000 to 1,100,000 is "+10%." The first number is ten times the second as a percentage, yet its business meaning is far smaller. The denominator of percentage change is the base value; as the base shrinks toward the noise floor, percentage stops being a measure of change and becomes an amplifier of noise.

Why it happens

The distortion is built into the definition: change rate equals delta divided by base, so a smaller base magnifies the same delta. Two effects stack up on small bases. First, random noise is comparable in magnitude to the base itself (a handful of users moving in or out swings the rate by several points or doubles it), so the percentage jitters wildly. Second, the ratio loses stability—today's +50% may flip to -30% tomorrow because the denominator is bouncing around single digits. This is why growth percentages for new features, new product lines, and small channels are the most common source of false narratives: the number is huge, the information content approaches zero. The presentation fix is dual-value display (absolute delta plus percentage, or base plus percentage) and flagging changes whose base falls below a reliability threshold as "insufficient sample," preventing the percentage from traveling alone.

Where it stops holding

"Small" is not an absolute value but a matter of statistical confidence: for a product with a million daily active users, a thousand-user swing is a small base; for a seed-stage product, a hundred daily actives is normal. Thresholds should follow each metric's historical variance, not a global constant. Small-base percentages remain legitimate in specific contexts: early readings from growth experiments (explicitly labeled as low-confidence), or compliance-mandated fixed disclosure. Cross-metric comparison is especially dangerous—concluding that "new channel grew 300%" is healthier than "main channel grew 5%" compounds the distortion twice (both base and scale differ); comparison should return to absolute contribution.

Applying it

• Show percentage change alongside the absolute delta by default; never write a percentage when only the absolute delta is visible.
• When the base falls below the metric's reliability threshold, replace the percentage with an "insufficient sample" notice or a grayed-out presentation.
• Verification: pull the five metrics with the largest percentages from the dashboard and check their bases; if the base is comparable to the noise level, it is a distortion instance—switch to dual-value display.

Related

• Same group: U7.05.1 A single value means something only against a baseline · U7.05.2 Period-over-period suffers cyclic interference; year-over-year removes cycles but reacts late · U7.05.3 The baseline choice can flip the same metric into a rise or a fall · U7.05.5 The comparison period's statistical scope must match the current period's
• Nearby: U10.04.4 Check whether uncertainty, missing proportion, and sample size are labeled · U7.03.2 Needs a comparison baseline and a direction of change
• Search terms: small base percentage · low base effect · relative change distortion