ChatGPT × TikTok Ads Manager: An Ad Agent that Small Businesses can Trust with Their Budget
Role
Tools/Skills
Team
Duration
Overview
Turning a complex ad platform into a conversation a first-time advertiser can finish
ChatGPT × TikTok Ads Manager puts TikTok's ad platform inside a ChatGPT conversation. Ads Manager is powerful, but it assumes you already know the system. In ChatGPT, a small-business owner describes their business and ends up with a published campaign. The challenge is trust: first-time advertisers worry about overspending, so the agent has to earn their confidence before they publish.
Core Flow
From uploading website link to published campaign.
Final review step
Video and campaign details together, before publishing.
Edge cases
Payment, video failure and targeting states
My role: I owned one core flow, the final review step and every edge case. On this early-stage product, I tested designs with small-business owners to validate designs, and work directly with pm and engineers to deliver efficiently.
Design Goal
A first-time advertiser should publish a campaign without learning the platform
In testing, small-business owners worried about spending too much and wanted to know what the agent could do before they trusted it.
User goal
Go from a website link to a published campaign with confidence.
Business goal
Grow new users who actually spend(NCA) by helping more small advertisers launch from inside ChatGPT.
Challenge
In a chat, users can say anything at any step, so the flow can't assume an order
Pilot users iterated heavily before publishing, asking for the campaign draft before the video or changing product details after seeing the result. I had to design for the unpredictable.
Hypothesis: seeing the finished video and the budget in one place lowers the fear of overspending.
If users can see the finished video and the campaign details, including budget, in one place before publishing, they will feel in control and be more likely to publish.
Core Flow
Decision
One final review step shows the video and campaign details together
Because steps arrive in any order, users lost track of their results. After generating several videos or making edits, they had to scroll back up to find the latest one, and the PRD hadn't planned for it, so I came up with the final publish step.
AI pratice
I prototyped it to make the case.
I built fast AI prototypes and showed the product team the difference. I also considered reposting the video at the end of the conversation, but it would clutter the chat and repeat what users had already seen
Generated in Claude Design, a single final step that shows the video and the campaign details, including budget, together.
Impact
The design is shipping.
88%
completion rate
+TBD
New net cash active users
Improved
CSAT score
Reflection
What this taught me about designing AI products
Observe real behavior, not assumed behavior
Testing showed users iterate and move back and forth, so I designed for how they actually behave, not the happy path we assumed.
Business outcome is important for product decisions
Knowing the goal behind the redesign, growing new net cash active users, shaped what I prioritized and how I made the case.
Learning about OpenAI's native App SDK components
This was a new design environment, so I learned the SDK's native components and translated TikTok concepts into ChatGPT's design language.
Communicate with engineers to find solutions beyond technical constraints
With so many dependencies, we challenged each other until we found the best option. Engineers agreed to build the final review step even though it cost more.