Why benchmarks are tempting and dangerous
The appeal is obvious. A click-through rate (CTRShare of impressions that result in a click.) or a conversion rate (CVRShare of visits or clicks that convert.) on its own is a number without a story, and a benchmark promises the story: above average, below average, room to improve. Boards ask for context, agencies are judged against it, and a single published figure can settle an argument in a meeting faster than any analysis. The danger is that the figure settles the argument without earning the right to. Published benchmarks are averages over businesses that differ from yours in product, price, market, season, maturity and, most of all, in how they define and collect the metric. A conversion rate counted on sessions in one tool is not the same quantity as one counted on users in another, and a return on ad spend (ROASRevenue generated per unit of ad spend.) with a long attribution window is not the same as one with a short window, yet a benchmark table presents them as a single column.
The second danger is that a benchmark becomes a target. A target set from an external average inherits every one of its definitional mismatches and none of its context, and a team chasing it will either game the metric or conclude that it is failing when it is not. Goodhart's law describes the general pattern: a measure that becomes a target tends to stop measuring what it was chosen for. The core metrics guide explains what each common metric measures; this section is about what a published value of it can and cannot support.
Where benchmarks come from
Every benchmark has a sample, a method and a motive, and the three decide what the number means. The table lists the common sources and the questions to ask of each before quoting it.
| Source | How the figure is produced | Typical strengths | Typical weaknesses |
|---|---|---|---|
| Platform or vendor reports | Aggregated from the vendor's own customer accounts | Large samples; consistent collection within the tool | Only that vendor's customers; definitions fixed by the tool; a commercial reason to look good |
| Agency reports | Aggregated from the agency's client base | Practitioner context; often split by industry | Small and self-selected samples; clients who stay are the ones doing well |
| Industry surveys | Marketers self-report their figures | Covers businesses no vendor sees | Self-reported, unverified, and skewed towards those willing to answer |
| Academic and research studies | Controlled measurement on a defined population | Transparent method; stated limitations | Narrow populations; often dated by the time they are read |
| Aggregators and listicles | Compile figures from the sources above | Convenient | Sources mixed without reconciling definitions; dates and samples lost in the copying |
| Peer groups and sharing circles | Companies exchange figures directly | Comparable businesses; definitions can be agreed | Tiny samples; confidentiality limits detail |
None of these is dishonest by nature, and several are careful. The point is that the headline figure is the least informative part of any benchmark report, and the methodology section is the most. A figure whose source cannot be traced to one of these rows, with a date and a sample, is a rumour.
How this site labels figures
This site publishes no benchmark figures of its own, and the rule is deliberate. Where a page quotes a figure from elsewhere, it carries a sourced label with the publisher, the date and a link, so that a reader can check the sample and the definition. Where a page needs numbers to show how a method works, they appear inside a labelled illustrative example, invented for the page, and the label says so. Where a dataset or scenario is generated from stated rules, it is labelled modelled. The data library sets out the four labels and what each permits; for benchmarks the consequence is simple. An illustrative number may be used to learn a method and never to judge a result. A modelled number may be used to practise and never to set a target. A sourced number may be quoted with its source, and even then only after its definition has been matched to yours.
The labelling also applies to figures readers bring here. When a guide says "compare this with your own history", it means a history computed from a definition in your own KPI dictionary, enforced the way the metrics layer guide describes, at a stated grain, from a named source. The pages in this section keep every KPIA metric tied directly to an objective. in words rather than numbers for that reason: the method transfers, the figures do not.
The guide
The guide to using industry benchmarks is the working method for the moment a benchmark lands on your desk. It covers what a benchmark can and cannot tell you, a table of what to check in a methodology section and why, how to match the publisher's definition of a metric to your own before any comparison, why the comparison against your own history comes first, how to turn a gap against a benchmark into a hypothesis rather than a verdict, and how to report benchmarks to people who will act on them. It quotes no figures, so it can be read for the method without the method being confused with a result.
A furniture retailer and a vendor's benchmark report
A furniture retailer's marketing lead receives a vendor report stating that the retailer's search conversion rate sits well below the published median for home and garden retailers. The managing director asks why. Rather than defend the figure, the lead opens the report's methodology page and finds that the median is computed on clicks across accounts of every size, over a period that includes a seasonal peak the retailer does not share, and with the vendor's own conversion counting rather than the order system's. The retailer's figure, by contrast, is computed on sessions from the analytics tool with returns removed. Recomputed on the vendor's definition, the retailer's rate is close to the median, and the difference that remains is explained by the retailer's deliberate choice to bid on broad research terms that convert later through the showroom. The lead reports the comparison with both definitions side by side and proposes a test on the research-term landing pages, which is the only part of the gap the team can act on.
Building your own benchmarks
The benchmark that drives decisions is your own metric's history: its distribution over time, by segment, with the conditions recorded. It shares your definitions, your tracking, your market and your season, so a gap against it means something. Build it from the KPI dictionary's definitions so that every comparison is like for like. Record the range, not only the average, because a metric that moves between a low and a high every month cannot be judged on one month's value. Segment by the dimensions decisions are made on, channel, market, device, product line, because an overall average hides the differences that matter, and the KPI trees guide shows how a change in one segment propagates to the overall figure. Annotate the history with what happened: launches, price changes, tracking fixes, budget shifts, so that a future reader knows which movements were caused and which were noise.
External benchmarks then take their proper place. They are a sanity check when a metric is new and has no history, a prompt to investigate when your figure sits far outside a published range on a matched definition, and an argument for attention when a board needs one. They are not targets, and a team that treats them as targets will spend its effort explaining the gap rather than closing the one that matters.
