Every agency I have worked in kept a private wiki. Specs pasted from platform help centers, a benchmarks tab someone updated in 2022, a glossary the new hires were told to read and never did. The wiki was always out of date and always behind a login. The public alternatives were worse, either a vendor's blog with a lead form on every definition or a "complete guide" written to rank rather than to be right.
So we built the one we wanted and put it at agencyechelon.com/reference. It is free, it carries no advertising, and the text is licensed CC BY 4.0, which means you can quote it, teach from it, paste it into a deck and cite it.
Here is what is in it as of today.
501 glossary terms. Each has its own page, a plain definition, the aliases people actually use, related terms, and a citation line. The entries are written to settle arguments. CPM says what it is and what it is not, and points you to vCPM when the distinction matters. Incrementality explains why a platform's attributed number and a holdout result rarely agree. Terms are grouped by what they are used for: buying, measurement, targeting, creative, infrastructure, search, practice and privacy.
22 platform profiles. Google, Meta, TikTok, LinkedIn, Amazon, Pinterest, Reddit, Snapchat, X, The Trade Desk, StackAdapt, Netflix, Roku, Walmart Connect and the rest. Each one answers the same questions in the same order: what it sells, how it is bought, how it is measured, what the minimums are, what changed lately.
Benchmarks by channel. CPM, CPC, CTR, conversion rate and cost per lead for thirteen channels, with the source behind every number and the date it was reviewed. Every figure is a range, because an average CTR hides more than it shows, as we have argued before.
Ad specs. Dimensions, ratios, durations, file limits and character counts for fifteen platforms and channels on one page each. Specs change several times a year, so every page carries its review date and a link to the platform's own documentation.
29 answer-first questions. What is a good ROAS. How much do Google Ads cost. What is a good CPC for LinkedIn ads. The answer comes first, in the first paragraph, with the number, and the reasoning follows. They are written for a person with a plan due tomorrow and for the AI answer engines that increasingly read on that person's behalf.
18 comparisons, 8 calculators and 5 CSV templates. The calculators cover CPM, CPC and CTR, CPA and ROAS, break-even ROAS, reach and frequency, CAC against LTV, budget pacing and the sample size a lift test needs. The templates are a media plan, a campaign naming convention, a launch QA checklist, a media RFP question list and a measurement plan. They download as CSV with no email address in exchange.
There is also a directory of 117 vendors across DSPs, SSPs, verification, identity, consent, CTV, retail media and workflow, each described in one neutral line. Nobody paid to be listed and nobody can. And there are pages on the standards that move the money (OpenRTB, VAST, ads.txt, sellers.json, TCF, GPP, the MRC viewability definition) and the rules that constrain it.
For the machines, there is a glossary.json, a platforms.json and an llms.txt. If you are building something and want the whole catalog, take it.
The reference now sits under every post on this blog too. When a note uses a term that has an entry, the first mention links to it, and each term page lists the posts that used it. Both directions were built by hand, then automated, and the automation is capped at eight links per post so the prose stays prose.
Why give it away. Two reasons, and both are selfish enough to be believed. First, we use it. A shared reference that is public has to be kept right, which is the only reliable way I know to keep a reference right. Second, the trade is better when the definitions are shared. Half of the arguments I have sat through between agencies and clients were two people using the same word for different things.
While we are listing free things, here are the other five, because they belong together.
The Audience Finder searches 950,000 third-party audiences from 287 data providers, plus the native segments inside Google, Meta, LinkedIn, TikTok and Amazon. Filter by industry, objective, size, geography, provider and CPM. Shortlist, export a CSV, hand it to your DSP.
The Ad Mockup Generator turns one square image and four lines of copy into 39 placements across every major channel, rendered to current specs, in the browser, with nothing uploaded to us. We wrote about it last week.
And three regulation registers, each a country-by-country record of the law as it stands: artificialintelligenceregulations.com for AI, cryptoregulations.net for digital assets, and digitalprivacyregs.com for data privacy across all 197 jurisdictions. The reference's rules page summarizes the rules that touch advertising. The registers carry the full record by jurisdiction.
All six are linked from the homepage and from the footer of every page on this site.
If you find an error, send the source and we will fix it. If a term is missing, tell us which one. The library is reviewed on a schedule, and the review date is printed on every page so you can judge for yourself how fresh a number is.
Quick answers
Is the Agency Echelon Reference free to use?
Yes. Every page is free to read with no account, no email gate and no advertising. The text is licensed CC BY 4.0, so you can quote it, teach from it or reuse it with attribution, and glossary.json, platforms.json and llms.txt are published for software and AI systems.
How current are the benchmarks and specs in the reference?
Each page prints its review date. Benchmarks were reviewed September 17, 2026 and specs September 23, 2026, and both link to the platform documentation or study behind each number. Treat the figures as starting points and confirm in the platform before committing a plan.