PRDFeature Bet 1

Atom Triage Autopilot

Policy-Governed Agentic Triage + Routing + Action Execution

Turn Slack/Teams into a true "front door" by doing agentic triage, routing, and safe resolution for common IT requests---with policy gating, approvals, and full audit trails.

Target Segments

  • Primary: Mid-market to enterprise (2,000-10,000 employees)IT service desks with Slack/Teams-first culture and strong MTTR/SLA pressure
  • Secondary: HR/FinanceFor routing + approvals once IT proves value

Why Now

  • Competitors have limitations: JSM is Teams-OR-Slack not both; Freshservice Teams has enterprise constraints; BMC Teams support is 1:1 IM only
  • Atomicwork already supports Atom across Slack/Teams/email/portal and has Azure AD automations (password reset, group joins)
  • Atomicwork emphasizes audit logging and auditable decision paths for AI
  • Buyers switching to reduce MTTR needs fast-to-value capabilities without massive replatform

Objectives

1

Reduce Time-to-Correct-Assignment

Triage time for Slack/Teams-originated requests

2

Reduce Agent Touches

For top request types via policy-gated actions

3

Improve Routing Accuracy

Reduce "ping-pong" reassignments

4

Maintain Enterprise Governance

Explainability, auditability, human-in-the-loop

User Personas

Primary

Employee Requester

Quick help in Slack/Teams without filling forms

Primary

Agent

Clean, correctly routed tickets with context + suggested next steps

Primary

Service Desk Manager

SLA compliance, reduced backlog, predictable load

Primary

IT Admin

Safe automation, governance, minimal setup time

Functional Requirements

Core capabilities required to deliver on the stated objectives.

1P0
FR-001

Slack + Teams DM intake for Atom

User DMs Atom, session starts, user authenticated

2P0
FR-002

Context enrichment from IdP

Dept/role/location and manager chain available to routing logic

3P0
FR-003

Intent classification + confidence

System outputs intent (Question/Request/Incident-like) + confidence, logged

4P0
FR-004

Category/subcategory suggestion + confidence

Category suggestion + confidence generated and visible to agent

5P0
FR-005

Routing suggestion with "why"

Top 3 evidence signals and confidence shown for queue/assignee suggestion

6P0
FR-006

Ticket creation + in-channel confirmation

User receives ticket ID and ETA expectations in Slack/Teams

7P0
FR-007

Agent override + feedback capture

Override reason (optional) is logged when agent changes category/route

8P0
FR-008

Shadow Mode evaluation dashboard

Routing accuracy + confusion matrix generated from historical tickets

9P0
FR-009

Policy controls: allow/deny actions + thresholds

Policy enforced immediately and logged when admin updates

10P0
FR-010

Audit logs for AI decisions + actions

Time-stamped audit event stored for any AI suggestion or action

11P1
FR-101

Channel-based intake (mentions in #help)

Atom responds in thread and captures context when mentioned in channel

12P1
FR-102

Dedupe suggestion + merge workflow

System suggests duplicates; agent can merge in 1 click

13P1
FR-103

Built-in low-risk actions (3-5)

Reset/unlock/join group/request access with success/failure posted and logged

14P1
FR-104

Approval cards in Slack/Teams

Approver gets card when action requires approval; decision logged

15P1
FR-105

On-call escalation integration

PagerDuty/Opsgenie can be paged per policy when urgent incident intent detected

16P2
FR-201

Multi-department routing with data boundaries

HR tickets never expose HR fields to IT agents; enforced by RBAC

17P2
FR-202

Multi-language triage

Classifier supports top N languages with target F1

18P2
FR-203

Auto-triage learning loop

Overrides feed a retraining queue with human review

Success Metrics

MetricBaselineTargetChange
DeflectionTBD (pilot baseline)+5-10 percentage points
MTTRTBD (pilot baseline)15-30% reduction
SLA ComplianceTBD (pilot baseline)+5-15 percentage points
Cost per TicketTBD (pilot baseline)10-25% reduction
Routing Top-1 AccuracyShadow Mode baseline>=80% for top intents

Rollout Plan

1-3

Internal Alpha

Use internal tickets and synthetic cases. Validate audit logs, policy gates, connector safety.

4-9

Design Partner Pilot

Shadow Mode (suggest-only) for all requests. Enable auto-execution for 1-2 low-risk actions only after accuracy thresholds met.

10-15

Beta

Expand to channel intake + dedupe. Add 3-5 actions. Add on-call integration for incident-like intents.

16+

GA

Routing Top-1 accuracy >=80% for top 10 intents. Reopen rate not worse than baseline. No Sev1 governance incidents.

Want the full artifact?

Email me to request a PDF of this complete PRD with all requirements and specifications.

Request PDF