65% Fewer Tickets.
14-Minute Resolutions. Zero Downtime.
Agix built Dartmouth College's AI IT Helpdesk, a RAG-driven support engine integrated with Active Directory, campus identity systems, and 500+ service types that resolves 63% of tickets automatically and cuts average resolution from 4 days to 14 minutes, 24 hours a day, 7 days a week.
One of America's oldest universities. 16,500 people who all need IT help at the same moment.
Founded in 1769, Dartmouth is an Ivy League research university in Hanover, NH, 12,000+ students, 4,500 faculty and staff, and one IT team serving them all. Support needs spike 400% at semester starts and collapse during breaks, a demand curve no human staffing model can absorb. Dartmouth's brief to Agix was precise: build an AI layer that handles high-volume, repetitive IT support at scale, without touching the complex work that still requires expert human judgment.
400% ticket spikes. 4-day wait times. A 78% repetition rate the team was too busy to eliminate.
Dartmouth's IT challenge wasn't complexity, it was volume. The same questions, asked thousands of times, by students whose exams and deadlines made the urgency real and the wait time unacceptable.
Semester start spikes overwhelmed a team sized for average demand, not peak demand.
Every August and January, 12,000 returning and incoming students arrive simultaneously, each needing VPN access, NetID activation, email setup, software installs, and dorm network help within 72 hours. Ticket volume spikes 400% above the annual average. The IT team, sized for steady-state operations, had no mechanism to absorb it. Resolution times stretched to 4 days. Students preparing for the first week of classes waited days for access to the tools they needed to attend them.
78% of tickets were repetitive questions IT had answered hundreds of times, and couldn't stop answering.
Password resets, VPN configuration, email access, software license activation, and Wi-Fi onboarding accounted for nearly 80% of all tickets, questions with well-documented, consistent answers the IT team had resolved thousands of times before. The knowledge existed. The documentation existed. The bottleneck was delivery: every ticket still required a human to read it, find the right answer, and send it back. Expert engineers were spending their days on questions a first-year student could answer with a knowledge base and ten minutes of reading.
Zero after-hours coverage left students without support during the moments that mattered most.
Students study, submit assignments, and sit for exams at 11 PM on a Tuesday and 2 AM on a Sunday. IT does not. A student who can't access VPN the night before an exam, or whose Dartmouth email stops working hours before a graduate school application deadline, had no recourse until business hours resumed, sometimes 12–16 hours later. The gap between student urgency and institutional availability was responsible for the highest-anxiety support tickets and the lowest satisfaction scores in every post-resolution survey the IT team ran.
Seven stages. One conversation. From "I can't access VPN" to problem solved, in under 14 minutes.
Student Request → Intent Classification → Identity Verification → Knowledge Retrieval → Action Execution → Resolution / Escalation → Feedback & Learning, every stage compounds on the last, building a resolution engine that gets faster and more accurate with every interaction.
Six integrated layers that turn a student's frustration into a resolved ticket, with or without a human in the loop.
Each layer operates independently and as part of a unified resolution context; passing a live ticket object that accumulates intent signals, verified identity, knowledge hits, action outcomes, and escalation flags from first message to final resolution.
IT Knowledge RAG Engine
A retrieval-augmented generation engine trained on Dartmouth's entire IT knowledge base, 500+ service documentation articles, 3 years of resolved ticket history, system configuration guides, and the Dartmouth IT service catalog. When a student asks "how do I set up VPN on my new Mac," the RAG engine retrieves the three most relevant resolution paths from the knowledge base, ranks them by context match, and generates a response calibrated to the student's specific device and Dartmouth's current VPN configuration, not a generic answer from the internet.
Identity System Integration
Direct integration with Dartmouth's Active Directory, NetID identity system, Duo MFA, and student directory, so the AI can verify identity, check account status, and take authenticated actions on behalf of the student. When a student reports they can't access VPN, the system doesn't just return instructions: it looks up their account, confirms their NetID is active, checks their Duo enrollment, and either fixes the issue directly or pre-populates the escalation ticket with verified identity context so the human engineer starts from a complete picture, not a blank screen.
Service Catalog Automation
Automated resolution of Dartmouth's 500+ catalogued IT service requests, from password resets and software license provisioning to VPN access restoration and email account unlocks. For each service type, the system executes the resolution workflow autonomously when identity is verified and the action is within the pre-approved automation boundary. Service types outside that boundary trigger a pre-populated escalation to the right IT specialist, with identity verification and diagnostic context already attached, cutting human handling time from 47 minutes to under 8.
Escalation Intelligence
A routing layer that knows when it can't help, and what to do about it. Every unresolvable ticket is triaged by category, urgency, and requester context before being assigned to the right specialist queue. A student who can't access their exam portal the night before finals is routed differently to a faculty member requesting a new software license. Priority scoring weighs academic calendar context (exam periods, registration deadlines, semester starts) and flags surge conditions to the command center before human engineers notice the volume climbing.
24/7 Multilingual Student Support
Dartmouth's international student population brings IT support needs in 12 languages, submitted at all hours and with context that varies dramatically by home country and device configuration. The AI handles all inbound volume regardless of language or time zone, with language detection and localised response generation built into the first-pass classification stage. 35% of all AI-resolved tickets are submitted outside business hours, a previously unserved population that had no option other than waiting for the next business morning.
IT Command Center & Knowledge Loop
A real-time operations dashboard giving IT managers visibility across ticket volume, AI resolution rate, SLA compliance, escalation types, and emerging knowledge gaps. The knowledge gap detection layer identifies categories where the AI is escalating more than expected, a signal that the underlying documentation is incomplete or outdated, and surfaces them as ranked improvement opportunities. Teams that action the top three knowledge gaps per month see measurable AI resolution rate improvements within 30 days.
A student portal that resolves. A command center that sees everything.
The student-facing helpdesk delivers instant, conversational IT support through web, mobile, and email, with related action suggestions, service tiles, and resolution confirmations that show the student exactly what happened, not just that something happened. Behind it, the IT Command Center gives Dartmouth's operations team a live view of ticket velocity, AI resolution rates, SLA status, and knowledge gap rankings, updated in real time, available without a data request.
The same AI intelligence extended to course enrollment, understanding prerequisites, availability, and deadlines in a single conversation.
Following the IT helpdesk deployment, Dartmouth extended the platform to Course Enrollment, an AI agent that answers "which courses can I enroll in?", checks eligibility in real time, and walks students through the full registration process without a single advisor touch.
What happened when 63% of IT tickets stopped needing a human to resolve them, and the humans got their capacity back.
Measured across Fall 2024 – Spring 2025 academic year, against the same period prior year. 18,842 total tickets processed.
Total ticket volume fell 65% year-over-year, from 53,619 to 18,842, as the AI resolved tier-1 requests automatically and its feedback loop progressively eliminated the knowledge gaps responsible for repeat tickets. The reduction wasn't just about speed: it was about deflection. Students who received an immediate resolution in their first interaction didn't need to submit follow-up tickets, call the helpdesk, or walk to the IT office. A resolved ticket at 11 PM is worth three unsubmitted ones at 9 AM.
Down from a peak-period average of 4 days (96 hours) before AI deployment. The 14-minute figure includes both AI-resolved tickets (averaging 4 minutes) and human-handled escalations (averaging 38 minutes), weighted by volume. The improvement is most dramatic during semester start periods, where the AI absorbs the 400% volume spike entirely, keeping resolution times flat while human engineers focus exclusively on the complex, judgment-intensive tickets that the AI correctly escalated.
Up from 84% in the prior year, measured via post-resolution surveys sent to every ticket submitter. The satisfaction improvement reflects two changes: first, the elimination of wait time as the primary complaint driver; second, the quality of AI-generated responses, which consistently provide step-by-step resolution instructions specific to the student's device and Dartmouth configuration rather than generic advice that requires the student to interpret and adapt it. Students consistently rated AI responses higher on "solved my problem without needing to follow up" than prior human responses.
The hours recaptured from tier-1 ticket handling were reallocated to infrastructure projects, security upgrades, and research computing support that had been deferred for two consecutive academic years due to capacity constraints. IT engineers who previously spent 60–70% of their week on repetitive helpdesk tickets now spend that time on work that requires their expertise. Dartmouth's IT team did not shrink, it got three times more done.
Our IT team used to dread September. Fourteen thousand students arriving in three days, each one needing something, none of them willing to wait. Now we dread nothing. The AI absorbs the entire spike. Our engineers spend those first two weeks of semester doing real infrastructure work instead of resetting passwords and restoring VPN access. Ticket volume is down 65%. Student satisfaction is at 92%. The team is happier than it's been in years.
Four decisions that made a 65% ticket reduction possible, and made the remaining 35% faster to resolve.
Resolution over instructions, the AI acts, it doesn't advise
The critical architectural decision was to build an AI that executes resolutions, not one that returns documentation. An AI that tells a student "follow these 7 steps to reset your VPN" is marginally better than a knowledge base. An AI that detects the account issue, restores access, and sends confirmation is a fundamentally different product. The Identity System integration was the prerequisite, without verified identity and the ability to act on behalf of the student, the system would have been another chatbot. With it, 63% of tickets never needed a human at all.
Academic calendar awareness, urgency isn't uniform
A student who can't access email during finals week has a different urgency profile than the same student on a Tuesday in February. The escalation intelligence layer is calendar-aware: it applies surge scoring during registration windows, exam periods, and semester starts, automatically elevating priority on tickets submitted during high-stakes academic moments. This means students in crisis get escalated faster even when overall volume is high, and the IT manager can see the surge building in the command center before the phone starts ringing.
Always-on coverage converts deferred demand into resolved tickets
Before deployment, 35% of after-hours submissions were either abandoned before resolution or required next-business-day follow-up, creating a Monday morning backlog that compressed the start of every week. The AI converts that deferred demand into immediate resolution: 35% of all tickets are now resolved outside business hours, none of which require human follow-up. The compounding effect is that Monday morning backlog effectively disappeared, and the IT team starts every week from zero rather than from a queue that built overnight.
Transparent escalation builds trust, students know exactly when a human is coming
The AI never pretends it has resolved something it hasn't. When a ticket exceeds the automation boundary, the system tells the student precisely what it's done so far, which team their ticket has been assigned to, what priority level it carries, and when they can expect a response. This transparency, absent from the pre-AI experience, where tickets entered a queue with no acknowledgement, was the single most cited driver of the satisfaction improvement. Students reported higher satisfaction for AI-escalated tickets than for pre-AI human-resolved ones, simply because they knew what was happening.
What powers this system.
Conversational AI
Natural-language student support, intent classification, and identity-verified action execution, turning an IT request into a resolved ticket without human intervention.
RAG & Knowledge AI
Retrieval-augmented generation over institutional knowledge bases, IT documentation, service catalogs, and resolved ticket history, delivering institution-specific answers, not generic web results.
Agentic IT Automation
Identity-verified action execution across Active Directory, MFA systems, and service catalogs, resolutions that close the ticket rather than returning instructions for the student to follow.
IT Operations Command Center
Real-time visibility into ticket velocity, AI resolution rates, SLA status, escalation types, and knowledge gaps, giving IT managers the data they need to act before issues become incidents.
Higher Education AI
Sector-specific AI for universities; from IT helpdesk automation to course enrollment agents and student services, built for the institutional complexity and data sensitivity of higher education.
FERPA-Compliant AI Deployment
AI systems built inside institutional data boundaries; no student data leaves the campus environment, no third-party model training on protected records, full audit trail compliance.
Common questions about deploying AI IT support in a FERPA-regulated higher education environment.
The entire system operates within Dartmouth's institutional data boundary. Student identity data, ticket content, and resolution history never leave the campus environment, there is no transmission to external model providers and no use of student records for third-party model training. The RAG knowledge base is built exclusively from Dartmouth IT documentation, not from student data. When the system retrieves or acts on student account information (identity verification, account status checks), it does so through Dartmouth's own Active Directory and identity APIs with the same access controls that govern human IT staff. Full audit logs are maintained for every AI-executed action, satisfying FERPA audit trail requirements and enabling the same post-resolution review that applies to human actions.
The system uses retrieval-augmented generation rather than fine-tuning, which means Dartmouth's IT knowledge is stored in a structured, searchable knowledge base that the model queries at inference time, not baked into model weights through training. This has two significant advantages: first, the knowledge base can be updated in real time as IT documentation changes, without model retraining; second, the system can cite its sources, giving IT staff the ability to verify every response against the underlying documentation. The initial knowledge base was built from Dartmouth's IT service catalog, 3 years of resolved ticket history, system configuration documentation, and policy guides, a corpus that took 6 weeks to structure and index at project start. New documentation is added continuously by the IT team through an admin interface that requires no AI expertise.
Adoption reached 78% of all IT interactions within the first semester, driven almost entirely by response speed rather than promotion. Students who submitted a ticket through the AI assistant at 11 PM and had their issue resolved by 11:14 PM told other students. Word-of-mouth propagation during orientation week, when the AI absorbed the semester start spike that would have produced 4-day wait times under the prior system, created a strong first impression that persisted through the year. The IT team ran a single communication at the start of semester pointing students to the new system. Beyond that, the product sold itself: a 14-minute resolution versus a 4-day wait is not a difficult preference to form.
Yes, and Dartmouth has already done it. The Course Enrollment AI Agent is a direct extension of the same platform architecture: a domain-specific knowledge base (course catalog, prerequisite data, enrollment policies), identity integration (student profile, academic standing, registration history), and a conversational resolution layer that answers enrollment questions and walks students through the registration process. The expansion required domain knowledge indexing and a new service catalog definition, it did not require rebuilding the core platform. The same approach is applicable to financial aid inquiries, housing requests, registrar services, and library research assistance. Each deployment shares the core identity and knowledge infrastructure, reducing the per-domain implementation cost significantly.
Dartmouth went from kickoff to live deployment in 14 weeks. The project ran in four phases: knowledge base construction and indexing (weeks 1–6), identity system and service catalog integration (weeks 5–9), UI deployment and staff training (weeks 8–12), and phased launch with monitoring (weeks 12–14). The longest phase is always knowledge base construction; the quality of IT documentation directly determines early AI accuracy, and institutions with well-maintained service catalogs move faster than those that need documentation work done in parallel. We typically recommend a pre-engagement documentation audit to identify gaps before the clock starts. Most universities with a structured IT service catalog and documented policies can expect a 10–16 week implementation timeline, with AI resolution rates above 50% within the first 30 days of live deployment.
Ready to give your IT team their time back, and your students the support they need at 2 AM?
Most projects go from kickoff to deployed AI system in 8–16 weeks. Let's talk about what a 65% ticket reduction, 14-minute average resolution, and 3× IT capacity could mean for your institution.