Stop asking 'who handles this?' for every inbound message

Create a workflow 'AI inbox triage' triggered when a new inbound conversation arrives in any tracked channel (email, web chat, contact form, social DM via webhook). Goal: every message reaches the right human within 5 minutes with appropriate context attached, replacing the manual ops triage.

Configure AI step that classifies the inbound message into one of: sales-inquiry (interested in buying), support-question (existing customer issue), billing-question (payment/account), partnership (BD/integration), spam/non-actionable, or escalation-required (legal threat, security, abuse). Output: intent_label + confidence + extracted entities (mentioned products, mentioned competitors, urgency signals).

Configure multi-split routing into branches: sales-inquiry → AE task + sales-inbox; support-question → support queue (linked customer); billing-question → finance queue; partnership → BD task; spam → archive silently; escalation-required → on-call alert. Each branch gets the full message context attached.

On the sales-inquiry branch: try to match sender's email/domain to an existing Account or Lead. If match, surface their history (prior touches, opened deals, account tier). If no match, create lead with auto-enrichment trigger.

Configure AI write step on the sales branch: draft a personalized initial reply acknowledging the inquiry, referencing any past interactions if matched, offering specific next steps (book a demo / talk to specialist / send pricing). Saved as draft on the AE task; AE reviews, edits if needed, sends. Reply-SLA target: 5 minutes from inbound.

Create AE task with: original message, AI classification + confidence, matched account/lead context, AI-drafted reply ready-to-send. Assign by territory or round-robin. Priority: HIGH for known accounts, MEDIUM for unknown. SLA: 5-minute first response.

On the support-question branch: route to support queue with customer context (account tier, recent product usage, recent tickets). If customer is on a paid plan, flag as priority; if free-tier, route to FAQ-first response template.

Configure Slack step that fires when AI classification confidence is below 60% OR when the message is escalation-required type. Posts to #ops-triage channel with the message and AI's best guess at routing, asking for human review. Don't let edge cases fall through.

Validate and publish. Monitor: classification accuracy (sample 5% of messages weekly, human-rates whether the AI got it right — target: 90%+), first-response SLA, mis-routed-message rate. Tune the AI prompts when accuracy dips.