Beyond the Average User: Percentiles and Inclusive Product Design

Key takeaway (TL;DR)
The normal distribution, or Gaussian bell curve, is a familiar model for variation across a population. In product and interaction design, however, an uncritical reliance on the mean can create the flaw of averages. No actual user is exactly average in height, arm length, reading speed, device performance and proficiency simultaneously. JVDS emphasizes that effective commercial systems should not impose a single fixed solution on an imaginary average user. They need adjustable, tolerant and inclusive structures informed by percentiles, potentially from the first to the ninety-ninth, and by combinations of different variables.

1. The trap beneath the bell curve: the average person falls apart

Teams may assume that designing to the mean will satisfy 68% of users or a comfortable majority. That does not follow mathematically or ergonomically. Four ideas help explain variation in real populations:

• Variation in a single normal distribution: the mean plus or minus one standard deviation covers approximately 68% of a normally distributed measure, such as height. The lower and upper 16% still differ substantially from that central range.
• Different measurements do not move together: physiological and behavioral characteristics vary across dimensions. Someone of average height may have above- or below-average arm length, seated height or hand width.
• The all-purpose average person disappears: as the number of dimensions grows from one to ten, the joint probability that someone falls within every average band rapidly approaches zero.
• Adjustment is the practical response: universal design is not about fixing one size. It uses adjustable structures to cover broad ranges such as P5–P95 or even P1–P99.
• The historical lesson of the US Air Force cockpit: in the late 1940s, the Air Force encountered a troubling frequency of aircraft crashes. Cockpits had been built around average measurements from thousands of pilots, including height, seated height, arm length and chest circumference. Researcher Gilbert Daniels reexamined ten core dimensions across 4,063 pilots and found none who were close to average across all ten measures, within the stated 30% band around the mean. The supposed average person did not exist. The military abandoned fixed cockpit dimensions in favor of adjustable seats and pedals, now familiar in cars, and crash rates fell sharply.
• Shoe sizes illustrate the joint-probability problem: suppose adult male foot length in a region follows a perfect normal distribution with a mean of 260 mm. Manufacturing only a 260 mm shoe would not make 68% of people comfortable merely because their foot lengths fall near the mean. Width, arch height and instep depth also vary, and the probability of being average across every dimension declines geometrically.
• The misleading average response time: an enterprise-software dashboard might report an average API response time of 200 ms. Further investigation can reveal that 80% of simple read-only queries take 10 ms, while 5% of critical transfers and batch exports take 8,000 ms or frequently time out. Celebrating the average conceals the problems that important customers experience every day.

2. The average-user problem in digital design

Online products, enterprise dashboards and mobile interfaces frequently create exclusionary experiences when teams rely too heavily on averages:

2.1 The fictional standard viewport

A front-end team may point to an average visitor resolution of 1920 × 1080 and an average network speed of 50 Mbps. It then fixes typography, column widths and first-screen height to a 1920-pixel canvas and loads an uncompressed 10 MB promotional video. Field sales staff using a 13-inch, 1440 × 900 laptop encounter clipped content; customers on weak train connections remain stuck on a blank initial screen.

2.2 A single workflow that erases human variation

Forms often assume steady hands, visual acuity of 1.5 and plenty of time. Consider instead a mother holding a baby while using a phone, an older adult with stiff fingers and presbyopia, or a logistics worker making an urgent report in a moving truck. Tiny buttons and pale, thin type based on the capabilities of an average healthy young adult can exclude all of them.

3. A four-dimensional comparison for inclusive system design

The following matrix helps product and engineering teams move beyond averages and support a more varied population:

4. Full-spectrum inclusion in JVDS practice

In commercial digital platforms and modern UI/UX design, JVDS uses flexible engineering to challenge the imaginary average user:

• Test both ends of the P5–P95 range: prototype validation does not rely only on nominal average conditions. Core workflows run in two contrasting environments: a five-year-old budget Android phone with simulated 3G and the largest text setting, and a professional 4K display on a high-end fiber connection. A system must deliver value acceptably at both ends before launch approval.
• Create an adaptive workspace: one enterprise application need not impose one rigid layout. For complex industrial logistics and financial terminals, we design viewport-adaptive grids. An advanced operator can arrange eight dense monitoring columns on a 4K screen, while a branch employee can switch to a larger-type, focused transaction view. Explore JVDS design work for approaches that accommodate substantial individual differences.
• Investigate the tail behind the average: in experience audits, JVDS helps clients examine the slowest 5% of transaction logs and the 5% of form fields with the highest abandonment, then follows up with the people affected. Eliminating those severe points of friction can produce NPS increases exceeding 30% and recover substantial hidden commercial losses.

5. Frequently asked questions

Q1: Why is the median often more useful than the mean for evaluating user experience?

A few extreme outliers can heavily distort the mean, while the median is more resistant. If nine users complete a form in thirty seconds and a tenth takes 3,000 seconds after answering a phone call, the mean rises to 327 seconds, potentially suggesting that the form is extremely difficult. The median remains thirty seconds and better represents the typical completion. Reviewing the median together with P90 usually gives a more informative assessment.

Q2: Does inclusive design sacrifice efficiency for most users to accommodate a small group with disabilities?

No. Addressing one person's persistent constraint can also help many people experiencing a temporary or situational limitation: solve for one, extend to many. A comfortable thumb zone designed for a person with one arm also helps someone holding a subway rail while replying to an email. High-contrast type designed for low vision helps a delivery rider reading in bright sunlight. Inclusion strengthens the resilience of the whole system.

Q3: How can we show that a feature designed for an average persona may fit no real customer well?

Use a multidimensional persona comparison. Ask the product manager to select a supposedly standard feature, then compare five actual major customers across screen size, network setup, business concurrency and employee age. If none fits the average assumptions and each encounters constraints in different dimensions, the limits of the average-user model become concrete.