Microsoft Ads experiments: eligibility and test timing

Microsoft Advertising’s 30 September 2026 product update confirms general availability of optimization experiments for Search, Shopping, Audience and Performance Max campaigns. They let advertisers compare a proposed change with an existing campaign before applying it more widely.

For ecommerce teams, this is a way to test a defined advertising decision. It is not evidence that a particular bid strategy or creative will increase sales.

Leave time for a meaningful comparison

The setup guide describes a control campaign and a copied treatment campaign, with traffic and budget split between them. Microsoft recommends four to eight weeks overall: typically two to four weeks for stabilisation, followed by two to four weeks of measurement.

That makes this a planning tool for the coming holiday season, not a dependable way to find a winner before next week’s 6–7 October sales window. Choose a test whose timing fits the decision you actually need to make.

Define the decision before starting

Write down one hypothesis, the change being tested and the result that would justify adopting it. For a store, that might mean testing a bid target while tracking order value and acquisition cost together. Confirm that purchase tracking reflects the orders you intend to count.

Microsoft recommends a 50/50 split. Its guide also says to exclude the initial stabilisation period when reviewing results and allow delayed conversions to arrive. Once the experiment starts, its split cannot be edited.

Keep a dated record of promotions, stock changes and tracking changes that could affect interpretation. A discount that starts halfway through the test can change the business context even when the platform reports a statistical result.

Apply business judgement to the result

Microsoft provides a results comparison and options to apply a treatment to the original campaign or create a new campaign. Review the financial consequence as well as the platform’s label: more orders may not be useful if the additional advertising cost and discounts consume their margin.

An inconclusive test is a reason to review the evidence and test design. It does not establish that both approaches perform equally. Keep the original objective visible when deciding whether to adopt, continue or end a test.