AI Transformation Workflow for Performance Ad Research and Planning

When competitor research, marketing analysis, creative planning, and image production moved separately, the team had to transfer the same material and explain the same context every week. I saw this as a broken flow between stages, not an individual productivity problem.

I created one data flow in which research fed analysis, and analysis fed ad briefs and production prompts. I drew a clear boundary around the system: people remained accountable for what was actually made and launched.

Every Wednesday at 1:00 AM, the routine researches competitor advertising and market signals, produces a weekly marketing analysis, and records it in Notion.

At 3:00 AM, it reads the accumulated analysis and generates six ad briefs with prompts ready for image production.

Automation covers competitor research, marketing analysis, six ad briefs and production prompts, and the Notion record.

The team retained ownership of hypothesis prioritization, image production and selection, brand and quality review, campaign execution, and budget decisions. Campaign results then become inputs for the following week’s automated analysis and planning.

Repeatable research and drafting are automated; people remain accountable for final decisions about brand, quality, and budget.

I accumulated weekly competitor research and advertising analysis in a Notion database so previous observations and new hypotheses could be tracked together.

I linked analysis and ad briefs in the same operating database so production outcomes could feed the next week’s strategy.

Beyond advertising automation, I built agents for meeting notes, data analysis, and strategy planning. I standardized their roles and output formats so the work of one agent could become the input to the next.

This goes beyond using isolated tools: I operate a portfolio of agents that connects recurring work from one stage to the next.

Rather than stopping after a single generation, I aligned each agent’s role and output format so its result could be used directly in the next task.

After people reviewed and selected the generated analysis and briefs, we applied them to performance-ad production and campaign execution. This created a weekly creative portfolio spanning different messages, visual directions, and product hypotheses.

I evaluated AX by whether live campaign results informed the following week’s analysis and planning, not by the number of documents generated.

Weekly competitor research and analysis

Wednesday 1:00 AM

Ad planning and prompt generation

Wednesday 3:00 AM

Ad briefs

6 per week

In live operation

6 months

Creative originating in the automated workflow was used in Facebook user-acquisition campaigns on iOS and Android.

In the latest week of January 2026 campaign data, the two campaigns generated 260 installs, a combined CPI of KRW 1,491, and D1 ROAS of 10.2%.

The iOS campaign delivered a CPI of KRW 1,204, 24.8% below the KRW 1,600 target, while the Android campaign reached 18% D1 ROAS, 3.6 times its 5% target. I used this difference to steer the next iOS work toward install efficiency and the next Android work toward early revenue efficiency.

These figures are early efficiency signals from the latest week of January 2026. They do not establish incremental lift from AX or long-term ROAS.

Combined installs across two campaigns

260

Combined CPI

₩1,491

Combined D1 ROAS

10.2%

Android ROAS versus target

3.6×