# Choose a test

Use this when choosing how to learn about or improve acquisition. The result is
a justified test and channel, not a campaign configuration. If the channel and
purpose are already agreed, address consequential gaps and proceed to preparation.

## Find the uncertainty worth testing

Connect the customer's problem, the offer and the action they can actually take.
Use [market](market.md) when evidence is missing. Distinguish a delivery check,
an interest probe and a test of acquisition economics: each supports different
claims. Choose an observable outcome that serves this decision, not a universal
"new paying customer" target. Qualified enquiries may suit a long buying cycle;
clicks can test response but do not establish customer value.

Consider relevant ways to resolve the uncertainty, including preparation or
customer research before paid traffic. Compare a small set of credible options,
not every channel in the catalog. A requested channel is a starting point: explain
evidence against it before spending, without silently replacing the founder's goal.

## Compare channel, audience and market

Use current research, past campaigns and available channel methods. Compare:

- audience intent and reach in the intended geography/language;
- continuity of the offer, destination and useful action;
- material and product preparation, time, cost and ability to handle responses;
- observable outcomes, attribution limits and the buying/learning window;
- access and operational readiness, checked through the relevant skill.

An available integration is not a reason to choose a channel. Conversely,
incomplete account setup is a preparation cost, not evidence the audience is
wrong. If a different market is worth considering, explain the opportunity,
adaptation costs and comparable evidence; do not mix unrelated country costs or
silently broaden targeting to make the forecast attractive.

Ground each option in evidence or a stated hypothesis. To assess short-video
acquisition, use `egbe-os-organic-trend-watch` for TikTok/Reels examples, audience
response and possible content adaptations. Research a gap when the answer could
change the choice or test plan; reuse sufficient existing evidence instead of
requiring a new collection.

Competitor ads and funnels can suggest a test. Distinguish what was actually
observed, why it might transfer to this audience, what will be adapted and why.
Keep the source market, channel and destination with the borrowed component;
explain cross-market or cross-channel adaptation rather than presenting it as
observed evidence in the new context.
Activity or longevity does not prove profitability. Partial screenshots, an
unvisited checkout or a teaser do not establish a complete successful funnel.
Reuse useful evidence, not another company's promises or unverified proof.

## Make the test feasible and interpretable

Explain the proposed mechanism: why this audience, offer and channel might
produce the outcome. Use relevant costs and demand estimates with their sources,
coverage and uncertainty. Check numerical inputs against the cited values, units
and conversion denominators; combine funnel rates only when their stages and
populations match. Label adapted estimates as assumptions, not measured source
values. Forecasts are scenarios, not promises or pre-approved keyword/bid
settings. If useful estimates are unavailable, propose a bounded measurement
instead of inventing a rate.

Choose budget, duration and checkpoints from the question, plausible cost per
useful action, available resources and risk. Explain what the affordable sample
can and cannot distinguish. Check whether the spending cap or deadline would
arrive before the needed evidence, including conversion delay. An overly early
stop can reject a useful approach by chance; an unreachable success criterion
cannot support a useful test. For forecast-based numeric checkpoints, estimate
the observations affordable by each checkpoint and the risk of stopping a
plausibly useful test, exposing uncertainty in the rates and model. A weak design
cannot be repaired by silently extending spending. No fixed daily budget or
universal winner/dead threshold substitutes for this reasoning. Zero media cost
does not mean zero founder effort or preparation cost.

Recommend the test, why it beats the relevant alternatives, and what would change
the choice. Record concrete assumptions and agreed limits in the existing plan,
or `egbe-docs/marketing/plans/<id>/plan.md` if a new cross-channel plan is needed.
Alternatives are not cumulative spending commitments. Continue with
[run-test](run-test.md) and the relevant channel method.