South KoreaProfessional services2024 – ongoingDream Insight
DMAX AI platform and generative AI in ad production
Dream Insight cut ad planning-to-analysis time to about a third with generative AI and its DMAX tool
Problem
An advertising agency spent long hours on image and video production and on collecting and cleaning campaign data by hand. Low-performing ad creatives and keywords were found slowly.
Approach
Dream Insight built DMAX, an in-house deep-learning tool that predicts and optimizes ad performance, and added generative AI for images and video. In 2026 it finished an end-to-end AI platform, DMAX AI, covering planning, production, operation and performance analysis. Some features already run on its own campaigns while others are still being tested. The company also ran five 8-hour generative AI training sessions for staff and plans to sell the platform to outside firms later.
Results
Takeaway
The agency tested the tools on its own campaigns before offering them to clients. The company reported time savings only for the 2024 setup and gave no figures for the 2026 platform.
Categories
- Size
- Small business
- Industry
- Professional services
- Business function
- Marketing and salesProduction and quality
- Innovation type
- Process innovation
- AI technology
- Generative AIPredictive analytics and ML
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Reports the company's statements that it uses generative AI for ad images and video and applies big-data AI from planning to evaluation. Its in-house DMAX tool runs a prediction and learning loop for ad performance. The company says the time from planning to performance analysis fell to about one-third; it gave no cost-reduction figure.
Reports that the DMAX AI platform build is finished and under function-by-function testing. Some functions already run on the company's own campaigns, automating manual data gathering and flagging weak ad creatives and keywords. The company says it saw productivity gains but gave no numbers; the only figures are five 8-hour training sessions.
Full details are in the original sources.
