Halving Costs, Gaining 40% Speed: YouJing and MiMan Boost AI Marketing with Volcano Engine EMR

AI is rapidly deepening its role in marketing, with the Chinese AI marketing technology market projected to grow from $6.8 billion in 2025 to $9.7 billion in 2026, a year‑over‑year increase of over 40%. This surge brings massive compute pressure, as each audience segmentation and attribution task processes exponentially larger data volumes.

YouJing Technology: Halving Compute Cost, Speeding Core Tasks 40%

YouJing Technology, a high‑tech advertising solutions provider, originally ran its core data pipeline on EMR Serverless Spark Java Runtime, handling more than 1,000 Spark jobs daily. Weekly compute consumption reached about 60,000 CUH, and a critical weekly task took roughly four hours, delaying decision‑making.

The company’s goal was clear: reduce resources and latency without altering SQL, schedules, or production stability. After fully integrating the Bolt engine, weekly CUH dropped from 60,000 to 30,000—a 50% reduction in resource consumption and cost—while the average duration of core Spark SQL tasks shortened by about 40%, compressing a four‑hour job to roughly two hours. The migration required zero code changes and was seamless for downstream processes.

MiMan Technology: Zero‑Migration, Boosting Marketing Tasks 20%+

MiMan Technology, a data‑plus‑AI full‑stack intelligent marketing service provider, also relied on EMR Serverless Spark Java Runtime for over 1,000 daily Spark jobs covering tag aggregation, cross‑screen recognition, audience selection, and performance analysis. As brand partnerships expanded, data volume and EMR resource usage surged, and some core SQL jobs took 5–10 minutes each, slowing campaign pacing.

After adopting Bolt, overall resource consumption fell by more than 60%, and the duration of several key SQL jobs decreased by 20%, enabling faster audience segmentation and attribution. The Bolt engine worked plug‑and‑play within the existing stack, requiring no architectural changes.

EMR Component Bolt: Native Execution Boosts Spark Efficiency

Bolt is ByteDance’s open‑source native execution engine designed for enterprise‑grade data analysis. It now powers 100% of PrestoSQL and 85% of SparkSQL workloads internally at ByteDance. As a built‑in acceleration feature of Volcano Engine EMR Serverless, Bolt allows users to switch to the Spark Bolt Runtime without modifying their Spark usage patterns.

The migration follows a three‑step, controllable process: (1) simulate on production‑identical datasets to verify SQL compatibility, result consistency, and performance gains; (2) perform a traffic‑weighted gray rollout while monitoring resource use and job stability, with the ability to roll back instantly; (3) automatically revert to Spark Java Runtime if compatibility or stability issues arise, ensuring uninterrupted data delivery.

Numerous service providers similar to YouJing and MiMan have leveraged Bolt to achieve cost savings and speed improvements without rebuilding data pipelines or replacing compute frameworks. Bolt transforms Spark from a “general‑purpose executor” into a “native‑accelerated” engine, delivering higher throughput and lower latency with fewer resources, directly converting compute efficiency into business value.

Integrated Bolt‑enabled Volcano Engine EMR Serverless Spark is already deployed across internet, advertising, e‑commerce, and fintech sectors, delivering tangible results in AI marketing, audience selection, and attribution scenarios. As Bolt expands to support Presto, Flink, and other frameworks, the zero‑migration, native‑acceleration capability will become available to more enterprises, helping them unlock the full value of data in the AI era.