Asana AI Request Triage and Routing

Overview

This flow automates request triage in Asana using Azure OpenAI. It polls an Asana intake project; for each new task it fetches the full description, asks Azure OpenAI to classify the request (type, priority, effort, owner, due date) with a confidence score, and then routes it: confident items are assigned, dated, moved to a section, commented, and announced in Teams; low-confidence items are moved to a manual-triage section and a human is pinged in Teams.

Why it matters: Manual triage is a bottleneck. AI classification routes the bulk of requests instantly and consistently, leaving only the ambiguous ones for a person — and every decision is recorded as a comment on the task for an audit trail.

As-built status: Built API-first as a Dataverse workflow record. Ships Off. Flow Checker = 0 errors / 1 warning (the warning is just "This flow is off"). All three connections are bound and green in the designer.

Use Case

An operations or project-management team runs a high-volume Asana intake project. Requests arrive as tasks with free-text descriptions. Rather than a person reading and routing each one, the flow classifies and routes automatically, escalating only the cases the model is unsure about. Audience: Operations, Project Managers.

Flow Architecture

Environment Variables

Connectors & Connections

Connector	API name	Actions used
Asana	shared_asana	OnTaskCreatedV2 GetTaskV2 AddCommentV2
Azure OpenAI	shared_azureopenai	ChatCompletions_Create_2024Feb15Preview
Microsoft Teams	shared_teams	PostMessageToConversation
HHTTP	http	PUT /tasks/{gid} POST /sections/{id}/addTask

Customization Guide

Almost every realistic variant of this flow can be implemented by changing environment variable values. A few cases require small edits inside the flow definition — those are called out explicitly below.

Tune the threshold

Raise flowlibs_TriageConfidenceThreshold to send more borderline items to humans; lower it to auto-route more aggressively.

Routing rules

Edit flowlibs_RoutingGuidance to describe your teams/owners; the model uses it to choose owner_email.

Resolve owner to gid

Add an Asana ListUsersV2 + Filter step to map the AI's owner_email to a user gid for a guaranteed-valid assignee.

Effort to due date

Change Compute Due Date to derive the due offset from effort (e.g. XL = 10 days) instead of the model's due_in_days.

Persist a record

Add a Dataverse CreateRecord to log every classification for reporting / model-quality review.

Richer Teams card

Swap the HTML message for an Adaptive Card with approve/reassign buttons.

Key Expressions

The flow is intentionally light on Power Fx / WDL gymnastics — the heaviest expressions are the branch-name concatenation and the approval outcome check. They are listed below in the order they appear in the flow.

@triggerOutputs()?['body/gid']

@outputs('Build_OpenAI_Messages')

@first(outputs('Classify_With_Azure_OpenAI')?['body']?['choices'])?['message']?['content']

@greaterOrEquals(float(coalesce(body('Parse_Classification')?['confidence'], 0)), variables('varConfidenceThreshold'))

@formatDateTime(addDays(utcNow(), int(coalesce(body('Parse_Classification')?['due_in_days'], 3))), 'yyyy-MM-dd')

@{concat('Bearer ', parameters('EnvVar_AsanaPAT'))}

Customize & download