A MILLION ADS

Bringing clarity to an interface managing millions of dynamic audio variants

As Product Design Lead at A Million Ads, I was the sole designer working to overhaul the core editing interface used by creative producers to build and sign off dynamic audio ads for some of the world's biggest brands.

ROLE
Product Design Lead

TIMELINE
3 weeks

CHALLENGE
A Million Ads’ proprietary platform, Studio, powers dynamic audio advertising by personalising audio content (e.g. based on variables such as location, weather, time, date). Their core editing interface, Arrange View, had major usability issues.

Defining Success

Producers were spending up to three days editing a single dynamic ad. We needed to reduce that time significantly. Working with the PM, we set ourselves an ambitious target of cutting production time in half, to around a day and a half. We defined three things we needed to see from producers to consider the redesign a success:

Producers able to complete the editing of an ad without assistance from the product team
A measurable reduction in the moments where producers got confused or lost context
The most time-consuming tasks, such as checking variants and setting rules, feeling faster and more intuitive

Discovery

Before moving into high-fidelity designs, I sketched early concepts to pressure-test my assumptions. The core tension was how to surface controls producers needed constantly, such as variant selection, rule settings and filtering, without eating into the timeline space. The timeline was sacred. This led me to think carefully about progressive disclosure: what should always be visible, and what could be tucked away without losing context. From this, I identified five key challenges to solve:

Visualising audio with many variations
Showing which ad version is playing
Building confidence in rule logic (e.g. weather, location)
Making rule testing quick and easy
Editing individual audio blocks (e.g. trim, volume)

The side-bar problem

My initial instinct was to house the controls in a sidebar panel, a familiar pattern I expected producers to adopt naturally. However, producers were unanimous: the sidebar was eating into the horizontal space they needed for the timeline.

Reluctantly, I moved the controls to a horizontal bar above the timeline instead. It wasn't my preferred aesthetic call, but the reaction it received was immediately more positive.

I tested the redesigned interface with five internal creative producers at A Million Ads, using high-fidelity screen mocks rather than full flows to keep things realistic but focused.

Key learnings

Producers treat the timeline as sacred. Any design that competes with it, even a familiar pattern like a sidebar, will be rejected
Internal users are valuable proxies when external users aren't available, particularly when they share the same pain points
Progressive disclosure is essential when the underlying product is inherently complex. Not everything needs to be visible at once
Low-fidelity sketching has limits with highly complex software, but it's still a valuable tool for pressure-testing assumptions before committing to higher fidelity work

Prototype

Below are some selected screens from the prototype I tested with participants. I used high-fidelity screen mocks (not full flows) to keep things realistic but efficient. For example, when exploring audio variants, I tested:

Is a dropdown the right interaction?
Is drag-and-drop clear?
Do users need search/filter?
Should we limit visible variants?

Results

~2 days average production time per ad, down from up to 3 days — roughly a third faster than before (Target: 50%)
4 support requests per week from internal producers, down from 6 — a 33% reduction in help requests
Rule setting was significantly improved, with producers describing it as cleaner and easier to manage
Managing large numbers of dynamic audio variants remained a challenge and warranted further investigation

Next steps

While the redesign made a meaningful impact, it was the beginning of a longer journey rather than a finished solution. The variant management problem remained unsolved and needed dedicated time to explore properly. I would also have liked to test with external, client-side producers, as our validation was limited to internal users throughout. Competing priorities prevented a further iteration cycle, but the groundwork was laid for what a more comprehensive redesign could achieve.