Paid Media Experiment
Designs and reads out A/B tests for ad accounts, and refuses to call an underpowered result a winner just because it looks good early.
Overview
Paid Media Experiment is the experimentation skill behind /ads experiment in the
Claude Ads suite, for designing and reading out A/B tests
on ad accounts.
What it covers
- States the decision, causal hypothesis, treatment, control, randomization unit, primary metric and stopping rule up front.
- Checks platform constraints, overlapping experiments, conversion lag, seasonality and interference before sizing.
- Calculates sample size and duration from declared assumptions, disclosing approximations.
- Changes one decision surface at a time unless the design explicitly estimates interactions.
- For readout, verifies assignment integrity and data completeness before estimating effect; never repeatedly peeks and stops on a favorable result.
Part of Claude Ads
This skill is dispatched automatically by the main ads conductor for /ads experiment
requests. See the full Claude Ads entry for installation.
Installation
This worker skill installs automatically as part of the Claude Ads plugin, there's no separate step for it.
- Install the plugin once, from inside Claude Code:
/plugin marketplace add AgriciDaniel/claude-ads /plugin install claude-ads@ai-marketing-hub-claude-ads - This skill runs on its own whenever the
/adsconductor needs it; you don't call it directly. - See the Claude Ads entry for the full setup and platform list.
Related assets
Frontend Design
Generates distinctive, production-grade UI that avoids generic "AI slop," with bold typography, cohesive color, and precise motion, implemented as real code.
Graphify
Turns a codebase, plus its docs, PDFs, images, and video, into a queryable knowledge graph that a coding agent can traverse instead of grepping raw files, and never guesses silently: every edge is tagged EXTRACTED, INFERRED, or AMBIGUOUS so you know what was found versus inferred.
Caveman
Makes your coding agent answer in a stripped down, high signal style that keeps every technical fact and cuts filler, saving output tokens on chat style questions.