Transforming a 12,000-student university IT helpdesk—reducing ticket volume by 65%, cutting resolution time from 4 days to 14 minutes, and freeing IT staff for infrastructure projects.
Ticket Volume
Avg Resolution Time
Student Satisfaction
Key Outcomes
65% ticket reduction achieved in first full academic year deployment
Action execution (doing tasks automatically) is 40% more effective than instruction delivery
24/7 AI availability resolves 35% of tickets outside business hours
Transparent AI limitations maintain user trust better than overreach
IT staff transition from reactive support to proactive infrastructure investment
Dartmouth College deployed an AI helpdesk system that handles common IT requests—password resets, software installations, Wi-Fi troubleshooting, VPN access, and account provisioning—through a conversational interface integrated into the student and staff portal. The system uses RAG over Dartmouth's IT knowledge base and service catalog, resolving 65% of tickets automatically and reducing average resolution time from 4 days to 14 minutes.
Dartmouth College is an Ivy League research university with approximately 12,000 students and 4,500 faculty and staff. The Office of Information Technology (DartIT) manages IT services for the entire campus, handling thousands of support tickets per semester covering everything from email access to research computing. Legacy helpdesk processes required IT staff to manually process routine requests, creating bottlenecks during high-demand periods like the start of academic terms.
Every August and January, ticket volume spikes 400% as returning students need password resets, new students need account setups, and faculty need access to new course tools. With a fixed-size IT team, these spikes meant 3–5 day resolution times and frustrated students at the worst possible time in the academic calendar.
+400%
Semester Start Ticket Spike
Ticket volume surge during first two weeks of each semester when IT staff capacity is fixed.
4 days
Average Resolution Time
Time from ticket submission to resolution during peak periods—critical when students can't access course materials.
78%
Repetitive Ticket Types
Proportion of tickets that were identical or near-identical requests IT staff had answered hundreds of times before.
AGIX Technologies built an AI helpdesk layer that connects to Dartmouth's Active Directory, service catalog, and knowledge base to handle IT requests autonomously. The system can execute self-service actions (password resets, software license provisioning, VPN account creation) rather than just providing instructions—resolving issues instantly rather than guiding users through complex processes.
IT Knowledge RAG Engine
Semantic search over Dartmouth's entire IT documentation library, service catalog, and 3 years of resolved tickets to answer any known question instantly.
Identity System Integration
Direct API connection to Active Directory and Duo MFA enables the AI to verify identity and execute account actions (password resets, group membership) without human involvement.
Service Catalog Automation
Common service requests (software installs, VPN access, email forwarding) are fulfilled automatically via integration with the IT service management platform.
Escalation Intelligence
Complex tickets are automatically classified and routed to the appropriate IT specialist team with a full context summary, eliminating the triage step that previously added 1–2 days to resolution.
Multilingual Student Support
Support for 12 languages serving Dartmouth's international student population, with culturally adapted communication styles.
IT Staff Dashboard
Real-time visibility into AI resolution rates, common failure modes, and knowledge gaps helps IT managers continuously improve both the AI system and underlying IT processes.
Ticket Volume
Reduction in tickets requiring human agent handling across the full academic year
Avg Resolution
Down from 4 days during peak periods—students get help before their next class
Student CSAT
Student satisfaction with AI helpdesk—higher than 84% human-only baseline
IT Project Capacity
IT staff freed to work on infrastructure and security projects previously deferred
"Students submit a ticket at 2am on a Sunday before a Monday morning exam and it's resolved in 12 minutes. That simply wasn't possible before. The AI never sleeps, never has office hours."
Chief Information Officer
Dartmouth College Office of Information Technology
Student submits request via portal or email
The student submits a request in natural language. The AI classifier identifies the request type from 50+ categories and extracts key entities: the student's NetID, the affected service (email, VPN, software), and any error messages. Urgency is scored based on academic calendar context (exam week = high urgency).
Action Execution vs Instructions Only
Systems that tell students what to do have 40% lower completion rates than systems that execute the action directly. Dartmouth's AI does both, depending on the request type.
Campus Calendar Awareness
The system knew exam periods, add/drop deadlines, and semester starts and auto-escalated tickets during high-stress academic moments, maintaining trust at critical times.
IT Staff as Curators, Not Just Operators
IT staff were given a knowledge management interface to add solutions and flag errors. Staff ownership of the knowledge base quality drove faster improvement than any automated process.
Always-On 24/7 Availability
Students submit IT tickets outside of business hours—weekends, late nights—and university IT can't staff 24/7. AI coverage during off-hours alone resolved 35% of previously deferred tickets.
Transparent Routing to Humans
When the AI can't help, it says so explicitly and provides an estimated human response time. Honesty about limitations maintained student trust more than pretending the AI could handle everything.
Every AI system has constraints. Here's what to know before building something similar.
Physical Hardware Issues Require On-Site Support
Broken laptop screens, malfunctioning peripherals, and network infrastructure problems cannot be resolved remotely—the AI routes these to on-site support immediately.
Novel Software Not in Knowledge Base
When a professor adopts new research software mid-semester, there's a 48-72 hour gap before the AI has documentation to draw from. Edge cases like this still require human specialists.
FERPA-Sensitive Requests
Requests involving student academic records or sensitive personal data require human review to ensure FERPA compliance—the AI cannot make privacy judgment calls autonomously.
Institutional Policy Interpretation
Complex questions about IT policy exceptions (e.g., can a student use a non-standard VPN client for research?) require human policy judgment.
Explore the services, industry solutions, and intelligence types that power this system.
Common questions about building ai it helpdesk systems like the one deployed at Dartmouth College.