Mtell Agent Builder — Turning Alert Noise Into Actionable Decisions

_____________________________________________________________________________________

Case Study Snapshot:

Mtell's alert workflows were overwhelming reliability engineers with too many KPIs, unclear causality, and buried context—leading to alert fatigue, missed events, and slow time-to-resolution. I redesigned Agent Builder around a core principle: users need clarity, not clutter. The solution elevated the few KPIs actually driving each alert, made the AI's reasoning transparent, and gave experts control to tune thresholds, weights, and operating states without black-box mystery.

The shipped product transformed Mtell from an alert stream into a decision system. Key design moves included contribution-driven KPI ranking ("why this fired"), progressive disclosure for decision-ready summaries, and guardrails like confidence indicators, audit trails, and confirmation gates for consequential actions. Results were significant: 70% reduction in triage time, 80% fewer false positives, and 50% faster alert resolution—helping teams act faster with fewer mistakes.

_____________________________________________________________________________________

Context:

Mtell sits in AspenTech’s reliability and predictive maintenance portfolio, where teams monitor equipment health and respond to alerts that can impact uptime, safety, and production. Agent Builder was designed for reliability engineers, operations, maintenance, plant managers, and analysts who need to move from signal → diagnosis → action fast.

Problem:

• Alert workflows were slow and cluttered. Users faced too many KPIs and charts, unclear causality, buried context, and weak prioritization—leading to:

• Missed events and slow triage

• mixed results and error-prone decisions

• False positives and alert fatigue

• Longer time to resolution and higher operational risk

Approach:

We redesigned the experience around a simple principle: users need clarity, not clutter. The solution was to elevate the few KPIs that truly drove an alert, make the reasoning transparent, and still give experts control to tune the system.

• Key design moves

• KPI ranking + contribution analysis (what’s driving this alert)

• Progressive disclosure (decision-ready summary first, detail on demand)

• Thresholds, grouping, and filters to isolate what matters

• Standardization via the Unified design system (consistent color mapping by sensor role, units, annotations, and chart behaviors)

What shipped:

Agent Builder as both:

• A configurable tool users could tune, and

• An insights UI that explained the agent’s output in a decision-ready format.

Core capabilities:

• Contribution-driven KPI ranking and “why this fired” clarity

• Drill-down path: KPI → cause → sensor → action

• “View Agent” transparency: show reasoning, rules, and allow edits

• Configuration controls: KPI selection, weights, thresholds, time windows, operating states

• Mixed autonomy: some steps automated; others suggest-only or auto-tune with confirmation

Guardrails:

We avoided black-box AI with:

• Observed vs. modeled comparisons

• confidence indicators and explanations

• audit trail + data lineage

• Permissioning and controlled editability

• Automation constrained to safe zones; confirmation required for consequential actions

Outcomes:

• Triage time reduced by 70%

• False positives reduced by 80%

• Alert resolution time reduced by 50%

• Reliability improvements that reduced downtime and helped avoid incidents

Why it mattered:

Agent Builder turned Mtell from an alert stream into a decision system. By ranking what matters, showing why it matters, and enabling safe tuning, we reduced noise, increased trust, and helped teams act faster—with fewer mistakes.