You have a Microsoft Advertising campaign that is doing something. Maybe you imported it from Google Ads and let it run. Now you have a change in mind: a different bid strategy, new ad wording, a tighter audience, a new landing page. But your budget is small, the campaign is live, and if the change goes wrong you will not know until the money is spent. On September 30, 2026, Microsoft announced that optimization experiments are generally available for Search, Shopping, Audience and Performance Max campaigns. Microsoft Advertising experiments let you try the change on a copy of the campaign, compare the two side by side, and then decide. This post walks through a first test you can set up in your own account, what to leave alone while it runs, and when fixing your tracking should come before testing anything.
Key Takeaways
Microsoft creates an experiment campaign from an eligible base campaign, splits budget and traffic between the two, and compares them on one results page.
One change gives you a clear answer, and edits to the base campaign during the test can undermine the comparison.
A result is only as good as the conversions behind it. If your conversion goal is missing or measures the wrong action, fix that first.
Try the change on a copy of your campaign instead of the live one
According to Microsoft's September 30, 2026 announcement, optimization experiments are controlled A/B tests for bidding, targeting, creative and more. A unified results page compares the experiment with your original campaign. When a test wins, you can apply it to the original campaign or use it to create a new one.
Microsoft's experiment documentation describes the mechanics. You choose an eligible base campaign, name the experiment, set start and end dates, pick a traffic split, and change settings on the experiment campaign. You can test bidding, targeting, ads, creative, URLs and other copied campaign settings. When the test ends, the experiment can be promoted into the base campaign, graduated into an independent campaign, partially merged, or ended.
That structure protects you in one specific way: the original campaign keeps running while the copy is tested. It does not promise a clear answer. A small or unstable campaign can end a test with no winner, and an inconclusive result is still information you can use.
Get a safe first test ready in about ten minutes
You can prepare the whole test before you create anything. Open Microsoft Advertising and find the Experiments area. Then work through this list:
- Pick one campaign with steady, relevant traffic.
- Confirm it is an eligible Search, Shopping, Audience or Performance Max campaign. Microsoft's documentation says shared-budget campaigns cannot be used as the base campaign.
- Write down one proposed change, such as a new bid strategy, an ad message, a targeting adjustment or a final URL.
- Write down the one business metric that decides the test, such as qualified leads or purchases.
- Confirm your conversion goal records the action you care about.
- After you create the experiment, review the copy for anything that did not carry over, and confirm its start date, end date, traffic split and budget settings.
- Check the Experiments area in your own account to confirm your campaign is eligible, since access can differ by account.
Do all of that before the experiment starts serving. Writing the change and the metric down first keeps you from picking a favorite number after the results arrive.
Pick a traffic split that fits your budget
It is easy to assume an experiment must be 50/50. It does not have to be. Microsoft's documentation allows traffic split values from 1% to 99%. A small advertiser who is nervous about a bid-strategy change may start with a smaller share going to the experiment, while the original campaign keeps most of the traffic. A smaller share also means the test needs more time to collect enough comparable data, so the choice is a trade between risk and speed.
Microsoft also describes two ways of assigning traffic. In a traffic-based experiment, relevant searches are randomly assigned to the experiment or the original campaign, so the same person may see both versions on different searches. In a cookie-based experiment, a person stays assigned to one source. Microsoft says cookie-based tests may produce more accurate comparisons but may take longer to build statistically significant data. For a small account, that extra wait is worth considering before you choose.
Whichever you choose, avoid changing the split while the test is active. Treat it as set at launch.
Start with the change that is easiest to read
A small advertiser should begin with one clearly defined change that can be measured against the existing campaign. Ad copy is usually the simplest first test, because it does not alter targeting, bidding or the landing-page experience. Bid strategy is a reasonable test once conversion tracking works and the campaign has enough conversion activity for the strategy to use.
A landing-page URL can be tested too, but check the copied URLs carefully after the experiment is created, because the experiment copies campaign settings and entities rather than your website.
| Change | Good first test? | Check before you start |
|---|---|---|
| New ad wording | Usually the simplest | The experiment copy has your ads |
| New bid strategy | Yes, once tracking works | Enough conversions for the strategy to use |
| Targeting adjustment | Possible; changes who sees the ad | Audience settings copied correctly |
| New final URL | Possible, with care | URLs on the copy point where you intend |
If you are still deciding whether Bing deserves ad budget at all, our post on whether Bing is worth it for a small business covers that question separately from testing.
Read the result without fooling yourself
Microsoft's results page puts the experiment next to the original campaign. Read it against the business metric you wrote down, not against clicks alone. A higher click-through rate does not prove a better business outcome, so look at qualified leads, purchases or cost per lead.
A fair reading needs enough comparable traffic and conversions, a stable original campaign, and no unrelated changes during the test. How long that takes depends on your traffic. Set a fixed period long enough to cover normal weekly behavior, leave the base campaign alone, and stop early only for a safety or tracking problem. If the test ends with too little data to separate the two versions, treat it as inconclusive rather than forcing a winner. No longer test period can create conversions that do not exist.
Before you apply a winner, check lead quality or sales outcomes, not only the platform's numbers. Then choose how to apply it: promote it into the original campaign, graduate it into its own campaign, partially merge it, or end the experiment.
Fix tracking first if the account is too thin to test
Some accounts should not start with an experiment. If conversion tracking is missing, broken or measuring the wrong action, a test cannot tell you anything useful. If traffic is sparse, or you cannot leave the original campaign alone for the test period, the better first job is measurement and cleanup. A specialist cannot manufacture enough demand to make a low-volume test conclusive, and you can open the Experiments area and inspect it yourself without paying anyone.
Help earns its fee when the hard part is not clicking "create experiment." That means choosing one clean question, checking the Microsoft UET tag and conversion goals, reviewing what copied correctly, keeping the control stable, and judging qualified leads instead of surface numbers. If your account does not have enough reliable data for a useful test, the sensible recommendation may be to fix measurement or simplify the campaign first.
Are you ready to run an experiment?
Pick an answer to begin.
1. What traffic split values can a Microsoft Advertising experiment use?
2. During a running experiment, what should you do with the original campaign?
3. Your conversion goal is not recording the action you care about. What comes first?
Frequently Asked Questions About microsoft advertising experiments
What are Microsoft Advertising experiments?
They are controlled tests that compare an experiment campaign, copied from an eligible base campaign, with the original campaign.
Which campaigns can use them?
Microsoft lists Search, Shopping, Audience and Performance Max campaigns. Check your own account's Experiments area to confirm your campaign is eligible.
How much traffic goes to the experiment?
You choose. Microsoft's documentation allows a traffic split from 1% to 99%.
Can I test bidding and ad copy?
Yes. Microsoft describes testing bidding, targeting, creative and other campaign changes.
Can I test a landing page?
A final URL or related campaign setting may be testable. Review the copied URLs and entities after the experiment is created.
Can I edit the original campaign during a test?
Avoid unrelated edits to it while the experiment runs, because they can undermine a fair comparison.
Moving Forward
Microsoft Advertising experiments are a safer way to change a live campaign: copy it, change one thing, give the copy a share of traffic, and compare the two side by side. Your first test needs only a steady campaign, a working conversion goal, one written change, one business metric and a hands-off original campaign.
Not every account is ready. If your tracking is shaky or your traffic is thin, fix measurement first and treat a muddy result as a reason to simplify, not to spend more. When a test does produce a clear difference, check lead quality before you apply it.
If you would rather have someone set up the test and read it with you, Web Leveling can check your tracking, design the one question worth asking, and keep the original campaign steady. Our pay-per-click work covers setup and interpretation, and you can reach us through our contact page. If your account is too thin for a useful test, we will say so before you spend on one. We work with small and medium businesses across the country and overseas.
Terms
Microsoft Advertising words in this post
Tap a term to see what it means.
Optimization experiment. A controlled A/B test that compares a changed copy of a campaign with the original.
Base campaign. The original campaign an experiment is copied from.
Experiment campaign. The copy where you change a setting to see its effect.
Traffic split. The share of the base campaign's budget and traffic given to the experiment, from 1% to 99%.
Cookie-based experiment. A test where a person stays assigned to one version, which may be more accurate but slower.
Graduate. Turn a finished experiment into its own independent campaign.
