---
title: "AI Case Study: Harvard's $0 Teaching Assistant That Never Sleeps"
description: "5 source-backed AI claims from AI Case Study: Harvard's $0 Teaching Assistant That Never Sleeps, with key statistics, context, and the original AI Adopters…"
url: "https://kbanc.com/claims-library/harvard-ai-teaching-assistant-case-study"
source: "https://aiadopters.club/p/ai-case-study-harvards-0-teaching"
date: "2025-01-23"
topics: ["implementation", "strategy", "tools"]
generated: "2026-08-31"
---

# AI Case Study: Harvard's $0 Teaching Assistant That Never Sleeps

By Kamil Banc | January 23, 2025

## Claims

1. **One Query Every 42 Minutes** (source summary): Harvard Business School's ChatLTV answered 3,112 student questions across 14 weeks, averaging one query every 42 minutes.
2. **Privacy Drove Adoption** (source summary): Making private chats the default drove 68% of students to engage with the AI system.
3. **Citations Built Trust** (source summary): Adding source citations to every response led 40% of users to rate answers high quality.
4. **Dashboard Spotted Strugglers** (source summary): A dashboard tracking student questions helped identify 12 struggling students before they fell behind.
5. **Four-Week Tech Stack** (source summary): Harvard built ChatLTV using Azure OpenAI, Pinecone vector storage, and Slack integration over four weeks.

## Evidence

### Quote
> "It's not about replacing experts—it's about amplifying their impact." - Kamil Banc

### Key Statistics
- **3,112 questions answered (one every 42 minutes)**: ChatLTV handled 3,112 questions from 250 MBA students over 14 weeks at Harvard Business School.
- **68% student adoption**: Usage jumped when private chats became the default, addressing student privacy concerns about visible questions.
- **40% high-quality ratings**: After source citations were added to every response, 40% of users rated answers 4 or 5 out of 5.
- **70% reduction in administrative emails**: The author notes this figure is not explicitly cited but is presented as plausible based on broader industry benchmarks.

## Context
The article presents a four-week implementation playbook Harvard followed: inventorying institutional knowledge, selecting a technical architecture (Azure OpenAI for security, Pinecone for vector storage, Slack as the interface), designing a constrained prompt template requiring source citations, and launching with a small test group while monitoring usage patterns. Practitioners can replicate this by starting with tools their teams already use, defining strict content boundaries in prompts, and defaulting to private chats to encourage candid questions. The author emphasizes that the system's value came not just from answering questions but from surfacing learning patterns—revealing gaps like 'founder market fit' confusion and identifying struggling students early. One caveat: the 70% email-reduction statistic is acknowledged by the author as inferred from industry trends rather than directly measured.

## Source
- Original: [AI Case Study: Harvard's $0 Teaching Assistant That Never Sleeps](https://aiadopters.club/p/ai-case-study-harvards-0-teaching)
- Cite: kbanc.com/claims-library/harvard-ai-teaching-assistant-case-study

## Primary Evidence
- [Professor Jeffrey Bussgang](https://www.linkedin.com/pulse/ai-professor-harvard-chatltv-jeffrey-bussgang-oiaie) (linkedin.com; supports claims 1, 5)
- [Pinecone](https://www.pinecone.io/) (pinecone.io; supports claim 5)