Teach the model by example: supply 2-3 input to output samples, then have it apply the pattern to your task.
Prompts / Techniques / Few-Shot Exemplar Pack for Insurance Claim Triage
Few-Shot Exemplar Pack for Insurance Claim Triage
Builds labeled few-shot examples that teach a model to classify and route insurance claims consistently.
ROLE: You are a senior claims-operations prompt engineer who builds few-shot exemplar sets for high-volume insurance triage.
CONTEXT: A team must auto-route incoming claims of type [CLAIM_TYPE] into the buckets [BUCKET_LIST] using only the fields [AVAILABLE_FIELDS]. Edge cases described: [KNOWN_EDGE_CASES].
TASK:
1. Write 6 few-shot examples, each as Input -> Reasoning (one short line) -> Label, covering at least 2 ambiguous edge cases.
2. Make the examples span every bucket and include 1 deliberate near-miss pair that differ by a single decisive field.
3. Derive a concise labeling rubric inferred from the examples.
4. Append a final 'NEW CLAIM:' slot for the live input.
CONSTRAINTS: Use realistic but fully fictional names and policy numbers. No protected-class attributes may influence labels. Keep each reasoning line under 20 words. Examples must be internally non-contradictory.
OUTPUT FORMAT: Markdown with a 'Rubric' section, an 'Exemplars' numbered list, and a 'Template Tail' code block ready to paste before a new claim.