Salesforce Decode
Salesforcedecode
Back to questions
AgentforceExpertmodelslatencyquality

Evaluate model selection per agent task for latency and quality balance

Real World Scenario

Tier-1 FAQ agent uses same large model as complex billing disputes; simple intents feel sluggish and costs remain high.

Expected Answer

• Task-tier model routing: fast model for classification and FAQ, larger for dispute reasoning • Measure quality delta on golden set before downgrading model tier • Latency SLA per channel: chat vs email vs async • Escalate model tier mid-session when confidence low on first model • Track token usage per intent for optimization targets • Avoid model proliferation unmaintainable — standardize tiers • Review new platform model releases against eval harness

Follow-Up Questions & Answers

Click to expand — each follow-up includes a direct, interview-ready answer

Main difference: use case and scale. Task-tier model routing: fast model for classification and FAQ, larger for dispute reasoning. Measure quality delta on golden set before downgrading model tier. Pick based on your integration pattern and team capability. One model for all intents wastes money and time — architect tiered inference like microservices sizing. Validate against production-scale constraints and long-term platform evolution.

Architect Perspective

One model for all intents wastes money and time — architect tiered inference like microservices sizing.