Back office ops · Production

Syracuse University deploys Claude to all students, faculty, and staff; builds Clementine AI course search and agentic data platform

The problem

Syracuse's traditional classroom model had been unchanged for over a century, preventing personalized learning at scale. Decades of accumulated institutional data lacked defined rules and relationships, making it unsuitable for AI-powered applications without extensive preparation.

First attempt

An initial multiple-choice practice exam built with Claude scored students only marginally above average because students clicked through answers without genuinely engaging with the material. The university's prior keyword-based course search returned irrelevant results for common queries.

Workflow diagram · grounded in source
1
University-wide Claude deployment
trigger
“In October 2025, Syracuse gave every student, faculty member, and staff member a Claude license and focused on training, governance, and the integration of AI into the classroom.”
2
Institutional data preparation
integration
“recording departmental meetings where staff debated the rules governing their own data and feeding the transcripts into Claude so it could learn the governing logic”
3
Agentic platform orchestrates data queries
ai_action
“Claude orchestrates specialized agents that run Data Analysis Expressions (DAX) queries against Microsoft Fabric or retrieve data through custom MCP connections”
4
Student submits assessment response
trigger
“students now receive a term and type in an answer”
5
Claude grades and identifies weaknesses
ai_action
“Claude grades their response and identifies where they could be stronger, based on actual words Rubin has used in class”
6
Clementine natural language course search
ai_action
“Clementine uses Claude Opus 4.6 to query millions of rows of enterprise data, so a student can ask for an AI class on Wednesday afternoons and get results contextualized to their schedule and career goals”
7
Donor reports delivered in minutes
output
“gives deans access to donor reports filtered by geography and giving history in just minutes”
Reported outcome

Exam scores jumped 12 points after the redesigned assessment.
Student peak daily active users grew 394%, staff 214%, and faculty 146%, with 123 of 139 university leadership members active. Work that previously took hours now takes minutes, and donor reports that previously took two weeks are now delivered in just minutes.

Reported metrics
Exam score improvement12 points
Student peak daily active users growth394%
Staff daily active users growth214%
Faculty daily active users growth146%
Show all 9 reported metrics
exam score improvement12 points
student peak daily active users growth394%
staff daily active users growth214%
faculty daily active users growth146%
university leadership active users123 of 139
budget planning task timehours to minutes
donor report turnaround time (before)two weeks
donor report delivery time (with Claude)just minutes
AI workshop attendance200 to 500 participants per session
Reported stack
ClaudeClaude Opus 4.6Claude CodeMCPClementineMicrosoft FabricMicrosoft 365ConfluenceJira
Source
https://www.anthropic.com/customers/syracuse
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Exam scores jumped 12 points after the redesigned assessment.

What tools did this team use?

Claude, Claude Opus 4.6, Claude Code, MCP, Clementine, Microsoft Fabric, Microsoft 365, Confluence, Jira.

What results were reported?

Exam score improvement: 12 points; Student peak daily active users growth: 394%; Staff daily active users growth: 214%; Faculty daily active users growth: 146% (source-reported, not independently verified).

What failed first in this deployment?

An initial multiple-choice practice exam built with Claude scored students only marginally above average because students clicked through answers without genuinely engaging with the material.

How is this back office ops AI workflow structured?

University-wide Claude deployment → Institutional data preparation → Agentic platform orchestrates data queries → Student submits assessment response → Claude grades and identifies weaknesses → Clementine natural language course search → Donor reports delivered in minutes.

Related back office ops cases