Campaign structure

Closely collaborating with engineering and the product team, I balanced the technical feasibility of updating components, some with older architecture that was significantly more expensive to update, with improvements benefiting users most. I set principles that we all agreed to for when it was okay to reuse what we have vs. when we needed a net new framework. I strived to avoid future tech debt as much as possible. In certain cases, I came up with temporary solutions to unblock us while still working towards our north star vision for reusable, scalable components.

Creating a new Pmax Campaign

Since we were building many components from the ground up and setting new standards, I leveraged this as an opportunity to push for a complete visual redesign of the wizards we use for creating ad campaigns. Historically, each campaign type was designed and implemented in a silo, resulting in many different inconsistencies —everything from interaction patterns for selecting locations to visual design of the wizard itself. Some campaigns were not even a wizard but one single, extremely long page full of stacked inputs.

I designed a new expandable card system using progressive disclosure to highlight the most important inputs at each step. Each card was a component added to a new global design kit for Microsoft Advertising, ensuring everything we were building could be easily ported over to new campaign types or refresh the older ones.

This phase involved a lot of competitive research, not just Google Ads’ Performance Max but similar structures for a simplified creation UX, like Snapchat’s “Instant Create,” that we referenced to understand the evolving competitive landscape. I combined that with a survey of all of Microsoft Advertising’s varying workflows, the user research study I ran, and data from several metrics dashboards we have in order to ideate through options for visual design treatments, the ideal order of steps in the wizard, and organizing all the inputs.

I first conducted a two‑stage research study to understand how advertisers approach campaigns, beginning with a competitive analysis of Google Ads, Microsoft Ads, and Facebook Ads.

I confirmed that advertisers consistently begin with goals, not campaign types. Many described thinking in terms of the sales funnel—awareness, engagement, conversions—before selecting formats. Participants mostly preferred Google for its guided flow that “feels like each Q/A builds to create a story” while experts preferred Facebook’s direct flow. Nobody preferred Microsoft’s UX.

I collected this information to ideate through some ideas for a better workflow, then tested 3 wireframe prototypes with varying degrees of progressive funneling in order to evaluate advertiser mental models.

The results were split evenly between the super wide funnel and showing more options upfront, and there wasn’t a business or technical reason to pick either direction. The consistent theme I saw though was that number of clicks was irrelevant and the progressive step-by-step guide resonated well. So I created a high fidelity prototype combining the learnings so far to evaluate in a follow-up study.

It was a massive success. The new goals strongly aligned with real advertiser intent, and first impressions of the updated layout, clarity, and visual design were unanimously positive. Participants found the guided flow easy to use, informative, and confidence‑building, with ad previews adding meaningful value. The only real friction point was that the “Skip goals” path wasn’t discoverable, which I addressed in the final design. Overall, the new experience outperformed Google & Meta, with 8 of the 13 participants choosing Microsoft’s new UX as their preferred option.

The biggest problem to solve for asset groups was how to connect the assets and audience. Pmax, optimized for performance, functions in a way where AI will use the advertiser’s input for audience as a signal rather than a strict definition for targeting. AI will even create assets for you if you have that feature enabled, but audiences will always expand to show ads to consumers likely to convert.

I explored design options for a framework of adding multiple simple audiences vs. single complex ones, and interaction patterns that compared the audience inputs directly in the page vs a side experience. While the inline versions were more discoverable and easier to add/edit, it felt too heavy in the page and added too much complexity for the requirement of saving these inputs in a library of audience. I ended up deciding on a framework of one “audience” made up of several segments that can be created and edited in a side panel.

My point of view at this stage was that since customers were very confident about their audience, let’s start with that. We should use these audience signals for asset recommendations, which addresses our core challenges. My next step was to A/B test two prototypes to validate this hypothesis as well as understand which model I had best matched the advertiser’s mental model.

My hypothesis confirmed, I devoted a design sprint to optimizing the workflow so we can use the customer’s URL to suggest audiences, then use the audience selection to suggest assets, showing samples of what the ads would look like. This aligned with what the participants in the study told us and made for a streamlined flow that automated most of the process of creating the asset group for them.

However, while reviewing with the leadership team, I got severe pushback about the prominence of audience signals. LT’s point of view was very strongly that PMax should seem smart, and asking advertisers for their audience might downplay the strength of automation from a marketing perspective. As I pushed back on that with the UXR, the engineering team also provided an estimate for the new pipeline of data we would need in order to actually make asset recommendations for a specific audience. It was insanely expensive, with a timeline that wouldn’t be feasible for our pilot anyway.

We moved towards the pilot and then GA with audience signals under the assets, but interestingly enough, I have heard a lot of feedback after the launch about the order of this that continues to confirm my original hypothesis. As the world of AI has evolved, our leadership team is now discussing switching the order of steps in Pmax so that we start with targeting, then move to assets. My main takeaway from this: I picked my battle too soon.

I didn’t make any major structural changes, since advertisers were happy with the existing workflow, but I did address the major visual design flaws that existed, making it easier to leverage those extensions that they already created. I also focused on adding value with the ad preview. At the time this was a source of major ambiguity for advertisers since they had no idea where extensions even go in ads. This was a huge win for both the advertiser that wants to see how their assets are used, as well as for Microsoft since it encourages more assets to be added to ads, thus more assets for AI to work with when the ad serves.

With my work on Pmax, I created the foundation for guided campaign creation that now powers consistent, explainable AI across the product—driving adoption, confidence, and measurable business impact

back to list