Microsoft Advertising is now opening up Performance Max experiment types in beta, with upgrade and uplift experiment options available to advertisers. On the surface, that sounds like a minor platform update. In practice, it points to something more significant: PMax on Microsoft is maturing into a channel where structured, evidence-based testing is actually possible.
For advertisers already running Performance Max on Google, or those considering Microsoft Advertising as part of a broader paid search mix, this matters. The ability to run controlled experiments rather than making blind campaign transitions is one of the most practically useful features any automated campaign type can have.
Why PMax and experimentation have always been uneasy partners
Performance Max was built around automation and consolidation. The model asks advertisers to hand over channel allocation, audience targeting, and creative selection to the platform's machine learning. The trade-off for that autonomy is performance - but measuring that performance against a controlled alternative has always been difficult.
Without a proper experiment framework, the only way to evaluate PMax was to run it and look at the account-level results. That approach is riddled with problems. Seasonal variation, external factors, budget shifts, and changes elsewhere in the account all contaminate the data. You end up drawing conclusions from noise.
Google introduced campaign experiments for Performance Max some time ago, which allowed advertisers to split traffic and compare PMax against standard campaigns in a more controlled setting. The fact that Microsoft Advertising is now introducing equivalent functionality - including upgrade and uplift experiments specifically - signals that the industry recognises this as a baseline requirement, not a premium feature.
What upgrade and uplift experiments actually test
The two experiment types serve different strategic purposes. An upgrade experiment is designed for advertisers migrating from existing campaign types to Performance Max. Rather than cutting over entirely and hoping for the best, you can run the new PMax campaign against your existing structure and let the data guide the transition.
An uplift experiment tests whether adding Performance Max on top of existing campaigns produces measurable incremental performance. This is the more nuanced question. PMax is often accused of cannibalising traffic from search and shopping campaigns rather than generating genuinely new conversions. An uplift test attempts to quantify what is actually incremental and what is just reallocation of demand that would have converted anyway.
Both experiment types depend heavily on how you define success. If your conversion tracking is not clean - if you are measuring form submissions without any quality filter, or relying on browser-based tracking that is degraded by consent refusals - the experiment will give you a clean-looking result that is built on bad data. The testing framework is only as useful as the signals feeding into it.
The case for treating Microsoft Advertising more seriously
For many UK lead generation advertisers, Microsoft Advertising sits in the background - absorbing a fraction of the budget, generating a portion of leads, and rarely receiving the same strategic attention as Google. That approach is understandable given the volume differences, but it does mean that Microsoft campaigns are often under-optimised relative to their actual contribution.
The arrival of PMax experiment types is another step in Microsoft Advertising's broader push to close the feature gap with Google Ads. For advertisers who have been reluctant to move Microsoft campaigns to Performance Max precisely because there was no way to test it properly, that barrier is now lower. The question is whether the Microsoft PMax product is mature enough to justify the test - and the only way to find out is to run one.
Microsoft audiences skew older and have higher household income profiles on average in many sectors. For certain B2B and financial services advertisers, that composition means Microsoft can punch above its traffic weight on lead quality. Running a properly structured PMax experiment on that audience base is a different proposition to running one on a broad Google inventory.
How to structure a PMax experiment you can actually learn from
The mechanics of setting up an experiment matter less than the conditions around it. Before you start, be clear on what question you are answering. Are you testing whether PMax drives better cost per acquisition than your existing campaigns? Whether it generates incremental volume? Whether it performs at equivalent efficiency to justify a full migration? Different questions require different setups and different evaluation windows.
Keep everything else stable during the test period. Do not change budgets significantly, do not run promotions that skew conversion behaviour, and do not adjust your standard campaigns in ways that affect the control arm. This sounds obvious, but in practice most accounts have ongoing optimisation activity that contaminates experiment results if you are not disciplined about it.
Set a minimum test duration that accounts for machine learning ramp-up. PMax campaigns need time to exit the learning phase before their performance stabilises. Cutting an experiment short because the first two weeks look bad is one of the most common ways to draw the wrong conclusion from a properly structured test.
What this means for cross-platform PMax strategy
Running Performance Max on both Google and Microsoft in parallel creates coordination challenges. The campaigns will compete for some of the same queries. Attribution will get messy if you are using last-click or even data-driven attribution models that do not account for cross-platform journeys. If you are using GA4 as your primary analytics layer, Microsoft click data will be underrepresented unless you have properly configured your UTM parameters and confirmed that Microsoft auto-tagging is not overwriting them.
The broader point is that expanding PMax to a second platform increases the importance of your measurement infrastructure. You need to know which platform is actually driving conversions, at what cost, and with what lead quality downstream. That means clean tracking, consistent UTM conventions, and ideally some form of offline conversion import that connects ad clicks to CRM outcomes rather than just form submissions.
Microsoft Advertising adding PMax experiment functionality is genuinely useful news for advertisers who have been sitting on the fence about committing to the format. It removes the binary choice between full adoption and complete avoidance. But the experiment framework is a tool, not an answer. The discipline required to run a clean test - in tracking, in account management, and in how you interpret the results - is exactly where the real work sits.