Document Type
Conference Material
Abstract/Description
Marketing automation is moving beyond rule-based systems and prompt-driven generative tools toward agentic AI: large language model (LLM) based agents that plan, act, and adapt under high-level objectives. Existing analyses are largely conceptual and commercial platforms are proprietary, leaving the community without reproducible baselines. We contribute three artifacts. First, a compact taxonomy organized around an augmentation–automation-autonomization continuum and a Perceive–Reason–Act(PRA)cycle that distinguishes autonomous agents from rule-based automation. Second, MarketingAgent, an open-source ReAct agent with six tools spanning SQL access, RFM segmentation, XGBoost churn prediction, campaign and email generation, and budget allocation. Third, the Marketing Agent Benchmark (MAB), a 15-task suite on UCI Online Retail II and IBM Telco Churn covering customer analytics, churn prevention, campaign design, and budget optimization. As an initial baseline, MarketingAgent completes 15/15 MAB tasks at 7.5 reasoning steps and 219.4 seconds per task on average, reaches 0.84 AUC and 0.79 high-risk precision for churn prediction, and segments 5,878 customers via RFM. Code, tasks, and evaluation scripts are released at https://github.com/shraddhakhadka/ MarketingAgent.
Department
Computer Science and Information Systems
First Page
761
Last Page
767
DOI
https://doi.org/10.1145/3838457.3838570
Date
9-2026
Conference Title
IoTML 2026: Proceedings of the 2026 6th International Conference on Internet of Things and Machine Learning
Citation Information
Thapa, Prashidda; Khadka, Shraddha; Sharma, Pragyan; and Rai, Ana, "Agentic AI for Autonomous Marketing: Taxonomy, Benchmark, and an Open-Source Agent" (2026). Student Publications. 5.
https://lair.etamu.edu/cose-student-publications/5
