Tec-Do used the ECCV 2026 conference in Malmö, Sweden, to bring academic and commercial attention to a growing challenge for AI-powered marketing: enabling agents to reason across images, video, text and other evidence before making commercial decisions.

On September 9, the Chinese technology company led the second Multimodal Reasoning and Slow Thinking in the Large Model Era (MARS2) Workshop, focused on multimodal reasoning and agentic commerce. The workshop brought together researchers from institutions including Tsinghua University, the University of Oxford, Nanyang Technological University and Seoul National University.

The event explored the gap between AI systems that can perceive multimodal information and systems capable of performing reliable, multi-step reasoning. That distinction is becoming increasingly important as AI agents move into marketing and commerce workflows where decisions may require evidence from multiple sources.

MARS2's focus reflects a broader development in AI marketing: agents are being positioned to do more than generate content or retrieve information. Commercial applications increasingly require systems to interpret advertising creative, understand video sequences, identify relevant evidence and connect those observations to marketing decisions.

The workshop's multimodal reasoning challenge tested those capabilities across three tracks: Multimodal Advertisement Comprehension, Video Temporal Grounding, and Marketing Strategy Decoding and Conversion Analysis. The competition attracted 64 teams and more than 1,060 submissions, according to Tec-Do.

The results exposed a distinction between so-called System 1 tasks, centered on perception, and System 2 tasks involving deeper reasoning. While models generally performed well on perception-oriented workloads, performance declined when tasks required multi-step reasoning, temporal causal attribution and other higher-order analysis.

One experiment cited by Tec-Do found that adding an audio-event timeline with cross-modal temporal alignment improved localization performance by 16.7 points. By comparison, increasing model parameters from 4 billion to 8 billion produced only marginal gains in that test.

The findings suggest that scaling model size alone may not solve the reliability challenges facing agentic marketing systems. How an agent collects, aligns and verifies evidence can be equally important.

Top-performing challenge solutions reportedly included Proposer-Critic dual-model validation, coarse-to-fine localization and adaptive token allocation. These approaches point toward architectures in which AI systems check evidence and reasoning rather than simply producing an answer from a single inference process.

That could be particularly relevant to agentic commerce, where marketing agents may eventually influence product discovery, campaign decisions, content interpretation and other commercial activities. Traceability becomes more important when an AI recommendation needs to be understood or validated by a human team.

Tec-Do is also positioning its MARS2 initiative as a bridge between academic research and commercial deployment. The company says its M-CAR benchmark and codebase have been open-sourced to encourage further research and reproducibility.

Tec-Do additionally cited performance by its Tec-Chi models in SuperCLUE advertising, marketing and overseas marketing video-understanding benchmarks, and said its Tec-Chi multimodal models and Navos marketing multi-agent platform serve more than 100,000 advertisers across more than 200 countries and regions. These figures are company-reported rather than independently verified.

Top Insights

• MARS2 focused on multimodal reasoning for agentic commerce.
• The challenge attracted 64 teams and more than 1,060 submissions.
• System 2 reasoning remains harder than perception-focused AI tasks.
• Cross-modal evidence alignment produced a 16.7-point improvement in one test.
• Validation and evidence verification emerged as key techniques among leading solutions.
• Open benchmarks could help connect academic AI research with commercial marketing problems.

Get in touch with our MarTech Experts.