agentleFS
narenaryannarenaryan/agent-skills/skills/marketing/marketing-geo-experiments/SKILL.md

marketing-geo-experiments

Skill0 starsChanged 41 days agoMITView on GitHub

Use when measuring campaign lift by city, DMA, region, or market because user-level randomization is unavailable or cross-channel sales matter.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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 targetable geo grain across relevant campaigns.

    line 21

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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 Geo Experiments

Geo is the assignment unit: a million rows across 12 geos still gives 12 clusters. Choose assignment and estimator separately: synthetic control can estimate randomized tests; without random assignment, the result is quasi-experimental.

Select a design

SituationPathGate
Exposure can be assigned across non-overlapping targetable regionsHoldback (new), go-dark (existing off), or heavy-up (existing increase); randomize within pairsRespect geo/pair assignment in inference
Randomization is infeasible but treated geos have a credible donor poolPre-specified synthetic controlLabel quasi-experimental; check pre-fit, donor contamination, placebos, and sensitivity
Few usable geos, no credible controls, or material interferenceRedesign or use another methodDo not launch a test that cannot change the decision; null is not “no effect”

Plan before changing spend

  1. Specify intervention, eligible geos, comparator, one primary KPI and conversion lag, guardrails, and decision threshold. Separate tested spend change from total-channel ROI.
  2. Verify targetable geo grain across relevant campaigns. Isolate budgets/bidding; shared portfolios can move spend between cells. Track planned and actual spend by geo.
  3. Use geo-by-period history to compare assignments and duration. Predeclare alpha/power and report them with MDE. For pairs, reserve separate pre-periods for matching and power to limit overfit. If MDE exceeds the decision effect, redesign. In randomized tests, pre-fit affects precision/power, not validity; synthetic control requires credible pre-fit.
  4. Predeclare assignment, KPI, guardrails, window, exclusions, interference checks, inference, and inconclusive action. Keep campaigns stable; log regional shocks and spend deviations before viewing lift.

Read at the assignment level

  • Randomized: use original assignment in inference; respect cluster/pair design and small geo count. Purchases/daily rows are not randomized units.
  • Synthetic control: require credible pre-fit, placebo/sensitivity checks, and named assumptions. Poor fit or contaminated donors means inconclusive, not a post-hoc tuned control.
  • Report contrast, actual spend change, lift with uncertainty, conversion maturity, and guardrails. Keep lift and spend change visible separately; near-zero spend denominators cannot support iROAS.
  • Apply only predeclared exclusions. Shocks/leakage may invalidate the contrast; do not drop geos after viewing outcomes.

Hypothetical: a 24-DMA video test shares one portfolio budget, so treatment spend may reduce control spend. Isolate budgets first. Pre-test power gives a 16% MDE; finance would act on an 8% lift. Do not launch. Add geos or extend a mature window, then rerun power; proceed only if the MDE can resolve the decision. A later +4% estimate with an interval spanning zero remains inconclusive.

Primary sources and worked examples: SOURCES.md.

Pitfalls

  • Choosing matched geos after observing test outcomes biases the comparison.
  • Neighbor spillover, travel, national promotions, or shared bidding can erase the intended contrast.
  • One pre-period fit or synthetic-control specification does not establish robustness.

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.