Email Triage Agent
Build an agent that reads unstructured support emails, understands the intent and tone, then decides the category, priority, and which team to route each message to — with zero human help.
Challenge Overview
This challenge simulates the front line of a support operation — an AI-powered triage agent that clears the shared inbox before the team logs on. You will work through a batch of 6 incoming emails, one at a time.
What makes it agentic is that nothing is pre-labelled. For each email the agent has to read the raw language, infer the sender's real intent and emotion, and make three judgment calls: the category, the priority, and the correct routing queue. It is the unstructured-text decisioning that separates a true agent from a simple rules bot — and misrouting an angry customer or a production outage has real consequences.
Routing Logic
Decide the category first, then route. Note: an angry complaint always goes to Escalations — not to the team it mentions.
Spam / promotional junk
→ Spam BinAngry / threatening complaint
→ EscalationsBilling / invoice dispute
→ Billing TeamBug, outage, how-to question
→ Tech SupportPricing / demo / buying intent
→ Sales TeamPriority Signals
Production down, charged twice, legal threat, churn risk
HighStandard request needing a human reply
MediumFYI, low-impact, no urgency, or spam
LowStep-by-step Instructions
1. Open the test inbox and read the first email in full — sender, subject, and body
2. Infer the intent: is it billing, technical, sales, a complaint, or spam?
3. Read the tone and urgency to set the priority (High / Medium / Low)
4. Pick the routing queue — remember angry complaints go to Escalations
5. Submit your decision and move to the next email
6. Process all 6 emails to get your triage accuracy score
What you will learn
✓ Intent classification from unstructured natural language
✓ Sentiment and urgency detection to drive prioritisation
✓ Decision routing — mapping understanding to an action
✓ Handling edge cases (a complaint that mentions billing still escalates)
✓ Agentic thinking: perceive → reason → act, autonomously
✓ Real tool exposure: UiPath Communications Mining, Power Automate AI Builder, Automation Anywhere IQ Bot, Blue Prism Decipher
Common mistakes
• Routing an angry complaint to Billing instead of Escalations
• Marking a production outage as Medium priority
• Treating a "you won a prize" email as a real sales lead
• Flagging a calm how-to question as High priority
• Confusing a pricing enquiry (Sales) with a billing dispute (Billing)
Complexity
Intermediate · Agentic
Language understanding, sentiment, and autonomous routing decisions.