agentleFS
narenaryannarenaryan/agent-skills/skills/marketing/marketing-incrementality/SKILL.md

marketing-incrementality

Skill0 starsChanged 41 days agoMITView on GitHub

Use when attributed ROAS drives budget decisions, campaign holdout lift is disputed, or experiment groups diverge after logging, joins, filters, or exposure-based segmentation.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Verify outcome coverage, spend separation, contamination, and cohort integrity below before interpreting lift.

    line 15
  2. Distinctive:

    No detected SRM does not prove unbiased outcome coverage.

    line 42

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Marketing Incrementality

Attribution allocates touchpoint credit; incrementality estimates effects of changing marketing. Randomize eligibility before exposure; compare assigned groups, including people never exposed. Valid assignment does not guarantee valid analyzed cohorts.

Design the decision

  1. Specify intervention, eligible population, control, decision, and horizon. Additional spend and channel shutdown estimate different effects.
  2. Randomize stable users/accounts where isolation is feasible; otherwise consider randomized non-overlapping geographies. Analyze uncertainty at the assignment level, not as independent visits.
  3. Predeclare outcome, guardrails, decision-relevant effect, power assumptions, duration, conversion lag, exclusions, and fixed-horizon or valid sequential analysis. Check feasibility against baseline variability.
  4. Verify outcome coverage, spend separation, contamination, and cohort integrity below before interpreting lift.
  5. Analyze intention-to-treat across channels. Report absolute/relative lift, interval, incremental spend, and margin/operating-cost assumptions. Positive estimates alone do not justify scaling.

Campaign execution or changes require authorization.

Diagnose cohort integrity

For user-level tests, compare distinct assignment-unit counts with the configured allocation using the design-appropriate SRM test. Trace counts through processing stages. Allocation ramps require design-specific analysis.

First divergenceInvestigate
AssignmentBucketing, unstable IDs, ramp timing
Logging/joinsVariant-dependent loss, duplicate/drop joins
Filter/triggerTreatment-dependent inclusion, absent control counterfactual logging

Localize by time and pre-treatment segment. An exposure/click filter can select on treatment effects. Recover retained raw outcomes and reprocess; if missing outcomes prevent valid target-population estimation, fix collection and rerun. Matching counts through arbitrary deletion/reweighting does not recover missing outcomes.

Arithmetic check

600 versus 500 purchasers in groups of 10,000 imply +1 percentage point, +20% relative, and 100 incremental treatment purchasers; $5,000 additional spend implies $50 each. Unequal groups require rates and population scaling. Estimates alone establish neither significance nor profitability.

Incremental ROAS = incremental revenue / incremental spend; near-zero denominators make ratios unstable.

Sources, worked counterexample, and validation.

Pitfalls

  • No detected SRM does not prove unbiased outcome coverage; balanced losses can still distort lift.
  • An SRM alert identifies a symptom, not its cause; user-count thresholds do not transfer to few-cluster geo designs.
  • Repeated fixed-horizon peeking inflates false positives; safety monitoring is not permission to declare an early winner.
  • Non-significance is not zero effect; spillovers and delayed conversions limit interpretation.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.