Design a scenario for sorting and routing requests. Return JSON {"scenario_name":"...","trigger":"...","ai_step":"...","routing":"...","action":"..."}. ai_step — application classification (department/topic/priority) + contact extraction + confidence assessment; routing - rule to whom to send by category, and indicate that applications with low confidence go to manual verification by a person (branch 'doubtful -> person'); action — create a card from the person in charge and confirm it to the client.
Applications are letters, forms and messages where the client wants something. The task of automation is to accept the application, understand its essence and deliver it to the right person without manual sorting. Scheme: trigger - a form has been filled out or a message has been received; AI step - classify the request (department, topic, priority) and pull out contacts; routing step - select the recipient by category; action - create a card in the responsible person’s to-do list and confirm to the client that the application has been accepted. Mechanics: AI works as a dispatcher - reads free text and puts it on shelves, and routing simply looks at the label. Leverage: always send the client an instant confirmation “received, we’re working on it” - this eliminates half of the repeat requests. Insight: ask AI to return not only the category, but also confidence in it; submit applications with low confidence for manual review rather than into the general flow. Second trick: add a priority deadline so that the urgent doesn’t drown. A typical rookie mistake: trusting AI to make the final decision on controversial applications without a fallback. Always start the “doubtful -> person” branch, otherwise a rare, strange case will go the wrong way and quietly get lost.
Unlock access to submit solutions for instant AI review.