Methodology

How each profile is built

The only data source for a profile is the agency's own website. For each one we read the home page and, where they exist, the "about" page, the contact page and the legal notice or privacy policy, and from there come the head office, markets, services, platforms, minimum, clients with a logo or case study, founding year and size. Every field with data stores the exact URL it was read from, and the profile shows it in its sources section along with the date that page was read, which for the 221 published profiles runs from September 29, 2026 to October 11, 2026.

Whatever the website does not say stays empty and is shown as "Not declared". The head office is not inferred from the domain, the language, the phone prefix or the clients, and nothing is taken from third-party directories, professional networks or rankings; if a cleanup finds an inferred value, it is emptied and the verification note says so. Clutch, Sortlist and LinkedIn were used only to discover that the agency exists and to store its profile URL as a link; no data was copied from them, no description, no size, no reviews.

A head office is marked only when the website declares an address on one of those pages, and the profile stores, when the page allows it, a short verbatim quote in the original language, so it can be checked without opening the site. When the website lists several offices without naming which one is the head office, the field stays empty and the agency is listed through its markets. Markets are only the countries the website names; when it talks about Latin America without detail, the profile says LATAM with countries not specified, and when it talks about Europe without naming countries, nothing is listed.

A complete profile shows every field with the URL it came from, and all profiles are listed under the same criteria.

Publication condition

A profile is published if it has an assigned company type, if the head office or the markets have a cited source URL, if the website was accessible when last read, and if it declares operations in at least one covered country (the 7 countries with their own page or any other Latin American country). Today 228 of 274 loaded profiles meet the data condition and 221 are published: 7 declare operations only outside the covered countries and would have no page to appear on, and the other 46 stay in the database without being shown until the website declares a head office or countries of operation, or until the agency claims the profile with an address.

Company types form a closed list: agency, agency with its own software, software, creator marketplace, talent representation, UGC provider and multi-service network. Agencies and agencies with their own software go in the main list; software and platforms, talent and UGC, and multi-service networks go in separate blocks on the same page, because they are not hired the same way. Services are also a closed vocabulary of 5 values (influencer marketing, UGC, talent management, performance and software) and only those the website declares are marked.

Order and updates

Lists are shown by default with the featured agencies first, each with the "Featured" badge, and then the rest in alphabetical order within each block, with no score or ranking; with the sort selector they can be viewed from A to Z, and featuring an agency does not change the content of its profile or its data, which always come from its source. Each profile is re-read against the agency website every 6 months; a profile not revalidated in 9 months shows "data not verified since" that date, and after 12 months with no revalidation or claim it leaves the lists, though it stays reachable by URL. Corrections arrive at listings@influencerbench.com or through the profile button and are answered within 5 business days.

Benchmarks: rules for publishing a figure

The benchmarks section publishes CPM and cost per result by segment (vertical, markets and period) under these rules. Today 2 segments computed on campaigns provided by an agency are published (details on the benchmarks page), and both carry the "estimated" label because they fall short of the strict rule on three points stated right there: the spend was paid in local currency and converted at the documented exchange rate, the cost basis is declared by the contributor per segment because it is not yet loaded row by row, and in one of them a single client concentrates almost all observations. Whatever does not reach even that says "No proprietary data" and is not replaced by an estimate or by a third-party figure presented as our own.

• Provenance: only closed, executed campaigns with measured spend and views. No proposals, quotes or scenarios. A row whose only estimate is the exchange rate, because it was paid in local currency and converted to dollars at the documented rate of the period, enters marked as "estimated", and that label travels with the figure everywhere.
• Unit of observation: one closed campaign in one market and one period. A multi-market campaign is split by market; if it cannot be, it is labeled "mixed market" in the title.
• Deduplication: client by campaign by market by period counts once, even if it appears in more than one file. "Total" rows that add up months already counted do not count.
• Cost basis: each segment declares whether the CPM was computed on creator fees or on the client's total spend, and whether it includes paid amplification. Different bases are not mixed, and while the source base does not have that field loaded row by row, the basis is declared by the contributor per segment and the page says so in those words.
• Currency: USD, with local currency converted at the documented exchange rate of the campaign date, published on this page once there is a segment.
• Result metric: a cost-per-result segment counts a single type of result (sign-up, install, lead, purchase) and is published under that name, never as plain CPA. Only results measured by the client or by the campaign's own tracking.
• Statistic: median and p25 to p75 range with percentiles by the inclusive method (linear interpolation). Never an average. With fewer than 8 campaigns the full range is published and marked "small sample".
• Minimum per segment: 5 closed campaigns and 2 different clients. A single-client segment is not published.
• Period and version: each segment carries its year range, and each publication is a dated version that distinguishes observed period, recalculation and editorial revision. Changing the date without new data is not presented as new data.
• Confidentiality: no figure for an identifiable client is published. Aggregates by segment are, with the number of observations and clients and, when one client concentrates most of the observations, it is stated; views, spend and client names never go next to a CPM.

When a market has no own sample, the benchmarks page says so in those words and, if it cites a public source, does it with a link and a date, never presenting a third-party figure as the directory's own.

Who contributes data today: Ramped provides the directory with the rows of its Benchmarks Library (campaigns 2023 to 2026) and the directory recomputes them under these rules; the Ramped Index on getramped.com is cited separately as an external source. Any agency or brand can contribute its own campaigns through data@influencerbench.com and they enter at the next review with the same treatment.