VorkLab

LLM workflow optimization / implementation

Choose a model for each step

Route routine extraction and summaries to Haiku, with validated fallback to Sonnet.

Choose the model for the step

Haiku 4.5

JSON field extraction
Short summaries of retrieved facts

Sonnet 4.6

Other LLM steps
Fallback when validation fails

Route by step type and document sizeFailed schema or business-rule check: Sonnet
No paid classifier call.LLM API total: $19,004 / month

Route routine steps with validation fallback

JSON field extraction and short summaries of retrieved facts moved to Claude Haiku 4.5. Routing is deterministic, based on step type and document size. It does not add a paid model call to classify the task. Other steps use Claude Sonnet 4.6.

If schema validation fails or a business invariant is violated, the step falls back to Sonnet 4.6. Access controls and confirmation requirements for sensitive CRM actions are the same for both models.

Call measureBeforeAfter
Incoming tasks485,000485,000
All LLM calls, including retries and fallbacks1,183,400979,700
Sonnet 4.6 calls1,183,400785,700
Haiku 4.5 calls0194,000
Calls per incoming task, rounded2.442.02

Haiku counts include attempts followed by fallback. They do not reveal the fallback rate or the number of unique tasks handled entirely by Haiku.

Interpreting the API total

The diagram compares total model API spending across the combined intervention. Checkpoint reuse and context changes also affect that total. It is not an isolated measurement of routing savings.

Combined results and evaluation

See the shared financial calculation, rollout checks and scope of the reported results.

Read the evidence behind the case
Agentic Code-Level Cloud & LLM Optimization
Request a free billing review

Accepted projects have no upfront service fee. Fixed fee: 30% of verified net savings over 12 agreed months, after approved transition and additional operating costs.