Intelligence Efficiency: Routing Each AI Task to the Best Model to Maximize Value
Webinar Speaker: Bill Franks, Research Fellow
Date: Tuesday, October 27, 2026
Time: 9:00 a.m. PT / 12:00 p.m. ET
Assessing AI’s impact on an organization based on token usage is rapidly losing favor. Instead, focus is rightly being placed on the value driven by that token usage. To achieve this value, it is critical to move beyond optimizing the capabilities of each individual model. Instead, focus on matching each task to the model that best meets that task’s cost, speed, and performance requirements. That leads to intelligence efficiency, which is effectively turning compute, models, data, and context into business impact.
Many frontier AI models are very good at complex tasks but come with a high price tag. They don’t make sense for simpler tasks. Open models are great for simple tasks but aren’t as powerful when deeper reasoning and analysis is needed. The reality is that different models occupy different points on the cost, quality, and latency curve. Each model’s position on the curve is changing all the time, and so the only way to select models well is to do it as tasks are requested. Successfully doing this is how intelligence efficiency can be achieved.
With the rapid pace of evolution of today’s AI models, and the increasingly important business decisions they are being applied to, it is more critical than ever to optimize price and performance effectively and to maximize intelligence efficiency.
Join experts from TDWI and Snowflake to learn about the latest trends in intelligent model routing, the methods behind the process, and how to rethink how AI is being applied in your organization.
Attend to learn:
- The benefits of continuously choosing the best model for each task
- How to define acceptable trade-offs once and apply them repeatedly when routing each task
- A feedback loop where each model’s output is evaluated for quality by another model, with the feedback influencing future decisions
- Why and how to measure business value instead of token usage as a sign of success