What the library holds
Most analytics guidance is written as prose, and prose is a poor place to keep a template or a table of sample rows. The library separates the reusable artefacts from the explanations so that each can be maintained on its own. There are three kinds of resource. The measurement templates are documents a team fills in: a measurement plan that runs from objectives to a tracking specification, the UTM and naming conventions that keep campaign data readable, and a KPIA metric tied directly to an objective. dictionary that fixes what each metric means. The practice datasets are small synthetic tables, with their schemas and generation method described, that let a reader practise the SQL, cohort and reporting techniques the guides teach. The benchmarks section is deliberately different: it contains no figures of its own, only an account of where published benchmarks come from and a guide to using them without being misled.
The library sits alongside the measurement cluster, which teaches the method behind each template, and the reporting and BI cluster, which teaches the analysis the datasets are for. A reader who arrives from a guide will usually want one resource; a reader who arrives here first should read the labelling rules below before anything else, because the labels decide how much weight any table on this site can bear.
How resources are labelled
Every table, figure and sample in the library is labelled with one of four statuses. The labels are not decoration; they are the contract between the page and the reader about what the content is. A figure without a label should be treated as illustrative until proven otherwise, and the site's own validation rejects unlabelled examples on published pages.
| Label | What it means | Where it appears | What you may do with it |
|---|---|---|---|
| Sourced | The figure or definition is taken from a named, linked publication, with the date and sample stated | Help-centre definitions, documented platform behaviour, cited studies | Quote it with its source and date; check the source has not changed |
| Calculated | Derived by a stated formula from sourced or synthetic inputs, with the formula shown | KPI dictionary formulas, exercise answers on dataset pages | Reproduce it; challenge the formula, not the arithmetic |
| Modelled | Produced by a generation rule or assumption set that the page describes | Synthetic datasets, scenario tables built from stated assumptions | Use it to practise or to test a method; never cite it as evidence about the market |
| Illustrative | An invented, labelled scenario that shows how a method works | Every labelled Example block on the site, filled template examples | Learn the shape of the reasoning; do not transfer the numbers |
Two consequences follow. First, nothing in this library is a benchmark. The datasets are modelled, the filled examples are illustrative, and the KPI formulas are calculated from whatever you feed them. Second, when a resource is updated the label travels with it: a sourced definition that loses its source becomes illustrative, and the page says so. The benchmarks hub explains how these labels apply to figures published elsewhere, which is where most of the confusion in marketing measurement begins.
Templates
The measurement templates exist because the hardest part of measurement is rarely the tooling. It is agreeing what the business wants to know, which numbers would answer it, how those numbers will be collected and who will keep the whole arrangement honest. The measurement plan template runs that sequence as a set of tables: objectives and the decisions they serve, the questions each raises, the KPIs and targets that answer them, the segments reports will cut by, the events and parameters that must be collected in GA4Google's event-based analytics platform. or an equivalent, and the owners and cadence that keep the plan alive. It follows the structure of the measurement plan framework, with a KPI tree beneath each objective so that a moving number can be explained rather than merely reported.
The UTM and naming convention templates are the smallest resource in the library and the one most often copied. They fix the controlled vocabularies for UTMURL tags identifying traffic source. source and medium, a campaign name pattern that encodes market, objective, theme and period, and a matching pattern for ad account campaign names so that platform reports and analytics reports describe the same thing in the same words. The KPI dictionary closes the loop: one row per metric, with definition, formula, grain, data source, owner, direction and caveats, so that two dashboards cannot disagree about what "conversion rate" means without someone noticing.
Datasets
The practice datasets are synthetic. That word is doing precise work: each dataset was generated from rules the page describes, with a fixed seed so the files are reproducible, and with known patterns planted so that exercises have answers. Nothing in them comes from a real account, a real customer or a real campaign, and the pages say so in their first paragraph because the alternative, a reader quoting a modelled conversion rate as if it were market evidence, is exactly the error the labelling rules exist to prevent.
The sample campaign dataset is a daily table of spend, impressions, clicks, conversions and revenue by channel and campaign, with a seasonal pattern, a tracking gap and a creative change planted for the reader to find. The sample e-commerce dataset is three related tables of customers, orders and order items, built so that cohort retention, average order value and repeat purchase exercises produce clean, explainable results. Both are small enough to load into a spreadsheet and structured enough to query with the patterns from the SQL guide. The datasets hub lists exercises in order of difficulty and is honest about what synthetic data cannot teach, chiefly the mess: the duplicate rows, broken tags and renamed campaigns that real data always contains and generated data only contains where someone chose to put them.
Benchmarks
The benchmarks section contains no site-authored benchmark figures, and that is a decision rather than an omission. Published benchmarks come from vendors summarising their own customers, from surveys of self-reported performance, from agencies describing their client base and from aggregators compiling all of the above. Each has a sample, a definition set and a date, and most readers see only the headline number. The hub explains these sources and why their figures vary so widely, sets out the labelling this site applies to any figure it quotes from elsewhere, and introduces the guide to using industry benchmarks, which is a method for reading a benchmark's methodology, matching its definitions to yours, comparing against your own history first and turning any gap into a hypothesis rather than a verdict.
The practical advice throughout is to build internal benchmarks: a metric's own distribution over time, by segment, with the conditions recorded. Those are the comparisons that lead to decisions, because they share your definitions, your tracking and your market. External figures are useful for sanity checks and for arguing that something is worth investigating; they are rarely useful as targets.
A subscription box company adopting the library
A subscription box company with a marketing team of four has three dashboards that disagree about monthly
conversion rate: one reports 2.1 percent, one 2.8 percent and one 3.4 percent. The analyst starts with the
KPI dictionary, discovers that one dashboard divides orders by sessions, one by users and one by clicks,
and fixes a single definition with its grain and source. She then fills the measurement plan template for
the quarter's two objectives, which exposes that the "trial to paid" question has no event behind it, and
adds trial_started and subscription_activated to the events table with a named owner. Before the new
events ship, a junior colleague practises the cohort query on the synthetic e-commerce dataset, where the
planted retention pattern gives a known answer to check against. When the board asks whether 2.8 percent
is "good", the team answers with the company's own twelve-month range by channel rather than a vendor's
headline, and proposes a landing page test as the hypothesis for the gap.
Using the library with the analytics guides
Each resource here pairs with a guide that owns the reasoning behind it, and the pairing is the intended way to use the library. Fill the measurement plan template while following the measurement plan tutorial, which explains why each table exists and who signs off on it. Copy the naming conventions after reading the naming conventions and UTM guide, which explains how analytics tools read the parameters and why case and separators matter. Build the KPI dictionary with the metrics layer guide open, because the dictionary is the human-readable face of a metrics layer and the governance rules are the same. Work through the datasets with the SQL for marketers guide for query patterns and the data quality guide for the habits that real data demands.
The frameworks behind the templates, the measurement plan and the KPI tree, are summarised on the framework pages and applied in the KPI trees guide. Read the framework for the idea, the guide for the method and the template for the document; in that order the library saves time, and in any other order it tends to produce well-formatted plans that nobody can explain.
