Practical patterns for LLM scoring in n8n workflows

Описание и применение

Внедрение LLM‑оценки в n8n: структурированный JSON, правила‑первый, LLM‑второй, обработка ошибок и логирование причин.

Задача

Integrating LLM scoring into n8n workflows while keeping deterministic, auditable rule-based logic and handling model failures.

Результат

Lead qualification becomes auditable, cost-effective, and resilient to LLM failures, with structured JSON output and logged reasons enabling review.

Рабочий процесс

• 1. Run deterministic rule chain first.
• 2. Pass data to LLM node with JSON schema request.
• 3. Receive structured score, verdict, and reasons.
• 4. Use IF node to route based on verdict.
• 5. On error, set default score and reason via Set node.
• 6. Log score and reasons to storage for audit.
• 7. Continue downstream actions (CRM update, alerts).

Доказательства

Never let the model write you a paragraph you then parse with string matching.

Источник

Put the scoring step behind your existing rule chain, and put an IF node after it.

Источник

Every one of those cases must land somewhere sensible: error branch to a Set node that writes a default score and a reasons: ["model_unavailable"] flag.

Источник

Scoring a lead against a rubric is not frontier-model work. A small, fast, cheap model with a prompt that contains your actual rubric — with examples — beats a big model with a vague prompt.

Источник

Write score and reasons to your log store every single run.

Источник