Amazon Cuts Ticket Classification Costs 98% With Strands Decider
Amazon replaced Claude Sonnet 4.6 with its own decision-making model, Strands Decider, for sorting internal tickets into 20 categories, and cut processing costs by 98%.
1 source 6h ago
Reporting from 1 source: GIGAZINE.
Why it matters
Classification work that only needs a label, not generated text, can be done far more cheaply by a small decision model than by a large language model, and Amazon's own numbers put the gap at roughly 44 times the cost.
Key Facts
- Amazon replaced Claude Sonnet 4.6 with Strands Decider for classifying internal tickets into 20 categories.
- Strands Decider E4B is a 4-billion-parameter model based on Gemma 4.
- Input tokens per ticket fell from 6554 to 2009 and output tokens from 170 to 1.
- Cost to process one million tickets dropped from $22,200 to $500, a 98% reduction.
- Average Lambda execution time fell 74% after the migration to Strands Decider E4B.
The tuned Strands Decider E4B, a 4-billion-parameter model based on Gemma 4, handles one million tickets for $500 instead of $22,200 and shortened average Lambda execution time by 74%.
Amazon's internal ticket system sorts customer escalations, capacity requests, automated alarms, and security issues into 20 categories. Claude Sonnet 4.6 handled that classification until Amazon decided a model that outputs a probability and confidence score per option fit the job better than a large language model built for text and image generation.
The tuned Strands Decider E4B returns only the option, not JSON, which shortens prompts and drops output tokens per ticket from 170 to 1. Input tokens fall from 6554 to 2009. Strands Agents cautions that a decision model may not replace an LLM when labels need explanatory text, the catalog changes weekly, or useful context cannot fit in a few thousand tokens.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google