## The Challenge
A Fortune 500 financial services company came to us with a critical problem: their IT service desk was drowning. With 2,400 open tickets and mean time to resolution (MTTR) sitting at 4 days, engineers were spending more time triaging alerts than actually fixing infrastructure.
## The Now Assist Architecture
We built a three-tier agentic workflow using ServiceNow's Now Assist platform, connected to Datadog for telemetry ingest.
### Tier 1: Intelligent Triage Agent
Now Assist analyzes every incoming ticket using its built-in LLM capabilities to:
- Classify severity (P1-P4) based on system impact
- Route to the correct team without human assignment
- Draft an initial diagnosis using historical incident data
### Tier 2: Autonomous Remediation Agent
For the 60% of tickets that match known patterns, the agent executes the fix directly:
``
javascript
// ServiceNow Flow Designer — Agentic Remediation Flow
const agentFlow = {
trigger: "incident.created",
conditions: ["category == 'infrastructure'", "confidence_score > 0.85"],
actions: [
{ step: "diagnose", source: "datadog_telemetry" },
{ step: "match_runbook", catalog: "remediation_library" },
{ step: "execute_fix", approval: "auto_if_low_risk" },
{ step: "verify_resolution", wait: "5m" },
{ step: "close_ticket", with_summary: "ai_generated" }
]
};
``### Tier 3: Escalation with Context
The remaining 40% that need human eyes arrive at the engineer's queue pre-populated with:
- Root cause hypothesis (90% accurate after training)
- Relevant Datadog graphs embedded in the ticket
- Suggested fix from the runbook library
## Results After 90 Days
| Metric | Before | After | Change |
|--------|--------|-------|--------|
| Mean Time to Resolution | 4 days | 19 hours | -80% |
| Tickets requiring human touch | 100% | 38% | -62% |
| Agent burnout score (survey) | 6.2/10 | 8.1/10 | +31% |
| Cost per resolved ticket | $47 | $11 | -77% |
## The Key Configuration Wins
### 1. Datadog Webhook → ServiceNow Integration
Every Datadog monitor that fires creates a ServiceNow incident with full context — no copy-pasting of error messages.
### 2. Now Assist Skills Training
We fine-tuned Now Assist on 18 months of the client's historical incidents, giving it domain-specific knowledge of their infrastructure quirks.
### 3. Confidence Gating
We never let the agent auto-resolve unless its confidence score exceeds 0.85. Below that, it still helps but requires a human approval click.
## What This Means for Your IT Org
An engineer freed from routine ticket triage isn't just "more productive" — they're working on the strategic infrastructure improvements that actually move the business forward. The 80% resolution time improvement is the headline, but the real win is what your senior engineers do with those recovered hours.