MktAI Assistant: AI‑Driven Marketing Data Query and Insight Platform

In a data‑driven marketing environment, traditional “write SQL‑run‑build dashboard” workflows are inefficient. The MktAI Assistant was developed to combine DATA+AI techniques, enabling self‑service queries and insight generation.

Challenges

Low‑efficiency manual data requests.

Difficulty extracting actionable insights from massive datasets.

Slow query performance and poor interpretability.

Solution Overview

The assistant is built on AI Studio, bsp, OneService data services, and Java applications. It uses LLMs as the brain, augmented with memory, skill planning, and tool usage. Core capabilities include tool calling, prompt engineering, and calibration.

Architecture

Agent = LLM × Memory × Skill Planning × Tool Use. The system calls external APIs (function calling) to fetch up‑to‑date data, applies Chain‑of‑Thought (CoT) prompting for transparent reasoning, and returns results in user‑friendly formats.

Prompt Engineering

Prompt structure (LangGPT) consists of Role, Constraints, Workflow, and Initialization. Constraints enforce data fidelity and avoid hallucination. CoT prompts guide step‑by‑step reasoning.

Evaluation Metrics

Stability = frequency of consistent answers.

Accuracy = frequency of satisfactory answers.

Case Studies

1. Self‑service query in “速爆” scenario – single‑item queries completed within 1 minute, higher data freshness, and richer summaries.

2. Insight analysis in “秒杀” scenario – AI‑generated attribution reports with white‑box reasoning, improving decision speed and interpretability.

Results

Metric

MktAI Assistant

Decision 360

Response speed

✅ Faster (≤1 min)

Longer due to large tables

Data freshness

✅ Higher (real‑time API)

Delayed offline tables

Insight quality

✅ Structured, explainable

❌ Limited

Conclusion

The MktAI Assistant demonstrates how AI agents can streamline marketing data access, improve insight quality, and foster deeper collaboration between data scientists and business users.

## Profile
- Author: qiaoyu
- Version: 1.0
- Language: 中文
## Goals:
- Identify user intent and route to appropriate tool
- Extract parameters, invoke log query tool
- Analyze and present results clearly
## Skills（需要说明每个技能调用什么工具、用户关键词是什么、模型该参考什么、给出什么样的回答、few shot示例）
1.技能1
2.技能2
## Constraints
1. 在调用相关API工具后，你需要解释结果，用markdown加粗标记重点以清晰的语言表述给用户。
2. 不要编造数据和事实。
## Workflow
1. 首先, xxx
2. 然后, xxx
3. 最后, xxx
## Initialization
作为MktAI助理，我擅长获取数据并进行诊断分析。我将用清晰和精确的语言与您对话。请告诉我您想要问的问题，我将竭诚为您提供分析结果.