The dominant procurement narrative in retail commerce, when the conversation turns to demographics, is approximately the following: the retailer has a target customer in mind, the target customer has a defined age and income profile, and the design of the storefront should be calibrated to that customer’s preferences. The narrative is intuitively compelling, widely repeated in the trade literature, and — on the data I have available from a portfolio of more than one hundred retail sites at Praxis Digital 3D — substantially incorrect in the specific ways that matter for the design and engineering of the storefront itself.
The Demographic Behaviour project was an eighteen-month effort to assemble, normalise and analyse the behavioural data across the portfolio with a view to identifying the differences in retail behaviour that actually correspond to demographic variables, as distinct from the differences that the trade literature assumes correspond to them. This post is an account of the project’s methodology, the data set, and the findings that have, in my subsequent work with retail clients, repeatedly informed both the architecture and the conversion strategy of new storefronts.
The methodology, in some detail
The data set comprised the behavioural logs of one hundred and seven retail sites across categories including fashion, homewares, garden equipment, beauty, automotive accessories, footwear and outdoor leisure. Each site contributed approximately twelve months of data, with the contribution period staggered such that the aggregate data set covered the period from January 2023 through June 2024. The total volume of session-level records exceeded forty-seven million; the unique-visitor count, after deduplication on the basis of the cross-site identifiers available, exceeded eleven million.
The behavioural variables collected included: the visitor’s session duration; the depth of the visitor’s catalogue exploration measured in product-page views per session; the channel of the visitor’s arrival on the site (organic search, paid search, social, direct, email, referral); the device type the visitor used; the proportion of the visitor’s catalogue exploration that was accomplished via category browsing as distinct from search and as distinct from filter use; the conversion outcome of the session; and, where the conversion completed, the average order value and the number of items in the order.
The demographic variables available for cross-tabulation against the behavioural variables included age band, gender (where the visitor had self-reported it on a previous interaction with the site), and the visitor’s broad geographic location. The age bands were of necessity coarse — the most reliable inference available from the data placed each visitor in one of five bands: under 25, 25 to 34, 35 to 44, 45 to 54, and 55 and over — and the inferences themselves were derived from a combination of self-reported data, retail account age, and demographic models supplied by the sites’ analytics platforms. The accuracy of the inferences for any individual visitor was, in the great majority of cases, modest; the accuracy in aggregate, against populations of several hundred thousand visitors per site, was sufficient to support the comparisons that the project was undertaking.
The first finding: search versus filter
The most decisive finding of the analysis, and the one that has, in my subsequent practitioner work, most consistently informed the architecture of new storefronts, concerns the channel by which different age bands accomplish their catalogue exploration. The dominant pattern, replicated across categories with very few exceptions, was that visitors under 25 accomplished the great majority of their exploration via the site’s search interface, while visitors over 45 accomplished the great majority via category browsing and filter use. The proportion of search-driven exploration declined approximately monotonically with age; the proportion of filter-driven exploration increased correspondingly.
The practical implication for storefront design is considerable. A retailer whose target customer profile is concentrated in the under-25 band is investing the great majority of its UX budget in the wrong place when that budget goes to filter design; the filter is, for the visitor population in question, a feature that the visitor will interact with infrequently if at all. Conversely, a retailer whose target customer is in the over-45 band is investing the wrong way when the budget goes to search-experience improvements; the visitor population in question performs few searches and the improvements are, on the conversion data, recovered slowly if at all.
The mid-range bands (25 to 44) showed a more even distribution between the two modes, with the cross-over occurring approximately at age 35 in the cross-tabulated data. The implication for retailers serving these mid-range bands is that the budget should be split, with a non-trivial proportion going to each interface; the proportion is, in practice, sensitive to the category, with fashion and beauty showing a younger search-skew than the demographic alone would predict and homewares showing an older filter-skew.
The second finding: session-depth divergence
The second finding, which has not, to my knowledge, been previously documented in the trade literature, concerns the relationship between session duration and conversion rate across age bands. The conventional position is that longer sessions correlate with higher conversion rates, and the position is, on the data, broadly correct as a population-wide observation. The position is, however, decisively incorrect when the data is disaggregated by age band.
For visitors under 25, the relationship between session duration and conversion rate was approximately U-shaped: very short sessions converted at a low rate, very long sessions converted at a low rate, and sessions in the middle range (approximately three to seven minutes) converted at the highest rate. For visitors over 45, the relationship was approximately monotonic: longer sessions consistently converted at higher rates, with no observable upper bound within the duration ranges available in the data. The mid-range bands showed intermediate patterns, with the U-shape becoming progressively flatter as the age band increased.
The interpretive implication, which the data does not directly establish but which the qualitative interviews conducted alongside the project broadly support, is that the under-25 cohort treats long sessions as evidence of indecision and abandons; the over-45 cohort treats long sessions as evidence of considered exploration and proceeds to purchase. The design implication is that the storefront’s conversion strategy should be calibrated to the age band the retailer is targeting; for younger audiences, the design should facilitate decision rather than exploration, while for older audiences, the design should support extended consideration rather than streamlined checkout.
The third finding: device and channel interaction
The third finding concerns the interaction between device type and arrival channel, and the way the interaction varies across age bands. For visitors under 25, mobile-device sessions arriving via social channels converted at approximately the same rate as desktop sessions arriving via direct channels; the channel was, in this cohort, a non-decisive variable for the conversion outcome. For visitors over 45, the channel was decisive: mobile-device sessions arriving via social channels converted at approximately one-third the rate of desktop sessions arriving via direct channels.
The implication is that the conventional position — that social-driven mobile traffic is, in some sense, an inherently lower-converting traffic class — is correct only for the older demographic bands and is essentially false for the younger ones. The retailer’s investment decisions in social channels should reflect the demographic composition of the audience on the channel; for younger audiences, the channel produces conversion outcomes broadly comparable to the better-performing channels, while for older audiences, the channel is genuinely a substantial conversion discount.
The fourth finding: returns and AOV
The fourth finding concerns the post-purchase behaviour, and specifically the relationship between average order value and return rate across age bands. The data showed that visitors under 25 placed orders with a higher AOV but a return rate approximately twice that of visitors over 45; the net contribution per session to the retailer’s revenue was, after adjusting for return-related costs, approximately equivalent across the two extremes of the age distribution.
The finding is not entirely surprising — the trade literature has, for some years, noted the higher return rate among younger consumers, particularly in fashion — but the size of the offset, against the AOV advantage, is rarely articulated in the trade literature with the specificity that the data permits. The retailer optimising for revenue per session should, on this data, treat the AOV advantage of the younger cohort with some scepticism; the additional revenue is, in non-trivial part, a transient inflation that the returns process subsequently deflates.
What the analysis did not establish
Two questions that the trade literature treats as settled were, on the data available to the project, more open than the literature suggests.
The first concerns the prevalence of mobile commerce among older demographics. The conventional position is that the under-25 cohort is overwhelmingly mobile and the over-45 cohort is overwhelmingly desktop; the data showed, for the period covered, that the desktop preference of the older cohort was decisively weaker than the conventional position holds, with mobile sessions accounting for approximately fifty-eight per cent of older-cohort sessions in the most recent quarter of the data and rising. The conventional position appears to have been correct circa 2018; by 2024 it was, on the data I have available, no longer correct.
The second concerns the responsiveness of older demographics to social channels. The conventional position is that the older cohort is unresponsive to social-driven traffic and that retailers targeting older customers should invest principally in email, search and direct channels. The data showed a more nuanced picture: the older cohort responded poorly to social channels for first-purchase acquisition, but responded comparably well to social channels for repeat-purchase engagement once an existing customer relationship had been established. The implication is that the social channel is misallocated when used as an acquisition channel for the older cohort and is well-allocated when used as a retention channel; the conventional position fails to make this distinction.
The implication for storefront design, considered carefully
The aggregate implication of the four findings is that the storefront design should be substantially more demographic-aware than the conventional procurement process treats it as being. The dominant procurement pattern, in which the design agency presents a single visual and interaction design that the retailer accepts as appropriate for its target audience, is producing storefronts that are calibrated to the imagined customer rather than to the actual customer; the calibration error is, on the data, decisive for the conversion outcome.
The recommendation that has emerged from the analysis, and which I now make routinely in scoping conversations with new clients, is that the storefront’s interaction architecture should be calibrated to the dominant age band of the audience as identified in the existing analytics, with explicit accommodation for the secondary age bands rather than an attempt to produce a single design that serves all of them equally. The accommodation may take the form of a search-prominent interface for younger-skewing categories, a filter-prominent interface for older-skewing categories, or a configurable interface in which the visitor’s first interaction reveals the appropriate mode. The choice between the three is sensitive to the category and the audience composition; the requirement that one of the three be made, rather than treating all visitors as undifferentiated, is the operative point.
(For a closely related view focused on the under-25 cohort specifically, see Why Gen Z bounces off your filters and millennials don’t. For the related question of how the analysis informed the design of the recommendation widget across the portfolio, see Content Recommendation Widget.)
An advisory close
The Demographic Behaviour project produced, in some sense, a single recommendation: that the design and engineering of retail storefronts should be informed by the actual behavioural patterns of the audience rather than by the demographic narratives that the trade literature treats as established. The recommendation is not, in any abstract sense, a novel one; the novelty, if there is one, is in the specificity of the patterns that the data revealed and the directness with which those patterns map to architectural decisions about the storefront itself.
It is recommended that any retailer commissioning a new storefront, or contemplating a substantial redesign of an existing one, conduct an analysis of the existing site’s behavioural data along the lines described above before the design work commences. The analysis is, in resource terms, modest; the design decisions that the analysis informs are, in commercial terms, decisive. The retailers I have worked with who have made the analysis a precondition of the design work have, on the conversion data, consistently produced better outcomes than those who have treated it as an afterthought.