AWS re:Invent 2025 - Agentic Analytics in Action (AIM295)

Summary of AWS re:Invent 2025 - Agentic Analytics in Action (AIM295)

Transition to Agentic AI in Analytics and Data Science

• Over the past year, the presenter has observed a convergence of classical AI, data science, and "geni" into a new paradigm of "agentic AI".
• This is expected to lead to a reimagining of most decision-making engines built over the last 15 years from an "agent AI" standpoint.
• The presenter outlines 5 key developments that have accelerated this trend in analytics and data science:

1. Reasoning Capabilities

• Agentic applications can now execute complex, multi-step analyses, proactively identifying and investigating secondary and tertiary questions.
• This allows them to provide more comprehensive and insightful answers.

2. Agentic Workflows

• Agentic workflows have become mainstream, moving beyond narrow AI (N8) approaches.
• These workflows leverage domain knowledge graphs and ontologies to infuse context and enterprise-specific information.

3. Deterministic Performance

• The combination of large language models (LLMs), multi-context prompt servers, and evaluation frameworks enables deterministic performance without hallucination.
• Agentic applications can provide specific, reliable answers to questions by triggering fine-tuned models for particular tasks.

4. Compliance and Security

• Deploying agentic applications in enterprises requires addressing compliance and legal requirements.
• Configurable guardrails and monitoring allow compliance teams to oversee the operation of these agentic systems.

5. Context Engineering and LLM Ops

• Context engineering and LLM operations (telemetry, monitoring, etc.) are critical for driving the performance of deployed agentic applications.

Real-World Agentic Analytics Use Cases

The presenter showcases four examples of agentic analytics applications deployed in Fortune 500 companies:

1. On-Demand Customer Analytics

• Agentic applications can execute complex customer analytics queries end-to-end, providing visualizations and insights integrated into CRM systems.
• These applications handle both structured and unstructured data, and can adapt to regional and language-specific contexts.
• Compliance teams can monitor the questions asked and responses provided by these agentic systems.

2. Agentic Marketing

• Agentic applications can automatically enrich lead information, generate personalized marketing responses, and scale lead generation and outreach.
• These systems continuously learn from user interactions and adapt the content and tone of communications.

3. Agentic Segmentation

• Agentic segmentation workflows can rapidly analyze data from multiple sources, create new personas, and identify drivers and product affinities for different segments.
• This replaces manual, time-consuming segmentation processes that were previously required.

4. Agentic Forecasting

• Agentic forecasting applications can incorporate multiple stakeholder opinions, review historical assumptions, and simulate the impact of disruptions or changes.
• These systems provide transparency and human-in-the-loop control over the forecasting process.

Industry Outlook and Expectations

• According to a Forrester survey of 400 analytics organizations, most expect human-in-the-loop agentic systems to be widely deployed within the next year, with autonomous agentic analytics following in the next 2 years.
• Agentic analytics is expected to significantly improve the speed and cost-effectiveness of routine analytics tasks, potentially halving the time and cost compared to current approaches.
• The presenter suggests that fully autonomous "AI analysts" capable of performing complex analytics at scale may emerge within the next 2-3 years, following the initial adoption of human-in-the-loop agentic systems.