A test budget should buy enough evidence to make a decision, not a promise that the campaign will “work.” Start with the outcome you need to observe and the amount you can afford to lose while learning.
Calculate a decision budget
Write down: the target action; the approximate cost range based on your own history or a clearly labeled planning assumption; the number of meaningful observations needed; and the stop condition. A simple planning model is:
Learning budget = expected cost per observed action × planned observations.
This is a planning ceiling, not a forecast. If there is no credible cost range yet, run a smaller discovery test and define what signal would justify another increment. Do not invent a benchmark from another market, season, or objective.
Worked example with hypothetical numbers
Suppose a team can allocate $600 for a two-week lead test and has a planning assumption of $30 per completed inquiry. That budget could produce about 20 inquiries at that assumed rate, but it does not guarantee 20 qualified leads. Set an additional quality check: what share receives a response, meets service criteria, and reaches the next business stage? If the landing page is not tracking correctly, stop and repair measurement before spending the full amount.
For Egypt COD campaigns, a submitted order is not the final economic result. The decision may need delivered, retained orders and return costs. For seasonal activity, record the date window and market so the test is not compared casually with an off-season period.
Define stop, continue, and change
Stop if tracking is broken, the offer is unavailable, or the spend cap is reached. Continue only if the agreed evidence is accumulating within the cap. Change the creative, audience, or page one at a time when the diagnostic points to that part of the journey. If a test produces too little data, mark it inconclusive instead of declaring a winner.
Budget for a decision threshold, not a round number
Estimate the spend needed to observe a plausible number of eligible outcomes using an explicitly stated cost range, then choose the maximum learning loss the business can accept. Define what happens at the cap: stop, revise the offer, fix tracking, or extend only if evidence meets a prewritten condition. Do not infer success from one conversion or failure from a short quiet period. Separate the cost of traffic from the value of the decision the test unlocks, and label all forecast inputs as assumptions until replaced with actual account data.
Optional AI-assisted work brief — Learning budget scenario
Using the cost range and outcome rate supplied, show low, middle, and high spend scenarios, the likely number of eligible outcomes, and the decision each scenario could support. Mark every input as observed or assumed. Define the stop, repair, and extend conditions before calculating; do not frame a forecast as a result guarantee.
Experiment thread: buy one decision
1/
A test budget should buy an answer to a business question—not a round-number spend target.
2/
Define the eligible outcome and the maximum loss before launch.
3/
Estimate a low, base, and high scenario. Mark every input as observed or assumed.
4/
Write the stop, repair, and extend rules in advance.
5/
A forecast is a planning range, never a result promise.
AI disclosure: AI tools assisted with drafting. All figures are hypothetical and not a performance benchmark.