Goldman Sachs Black in Business × NYU Stern: 97.7% AI Adoption by Design

Context

The Goldman Sachs Black in Business program, delivered in partnership with NYU Stern, teaches entrepreneurship to a cohort of 265 Scholars running real businesses. As AI Graduate Assistant, I embedded structured AI systems into the learning and operational workflows of the program: prompt systems first, then production web applications.

Problem

Most enterprise AI rollouts hit 30-60% usage and decay within 90 days. Here the users were non-technical small business owners, the stakes included academic integrity and sponsor brand standards, and the tools had to scaffold thinking without doing the work for people.

Approach

Design for adoption from day one: module-specific prompt libraries aligned to the curriculum, evidence-tagging and decision checkpoints to prevent shortcut behavior, and AI outputs framed as reflection material rather than answers. Then graduate from prompt cards to interactive production tools Scholars use directly inside the curriculum.

Frameworks

Adoption Science (novelty decay, workflow embedding)Curriculum-Aligned AI DesignNon-Punitive UX for Low-Financial-Literacy UsersData Minimization & Governance

Implementation

  • •Designed module-specific prompt libraries for customer discovery, persona design, and channel selection
  • •Built and shipped the SMART Goal Builder: a production web app (React, Vercel Edge Functions, LLM-guided conversational flow) that walks business owners from a vague aspiration to a polished SMART goal
  • •Built and shipped the BiB Financial Benchmark Tool: P&L margin comparison against benchmarks for 37 industries, built from data aggregating ~4.5M privately held U.S. businesses, with AI-powered industry matching
  • •Engineered the benchmark dataset from 34 enterprise industry PDFs with a Python extraction pipeline; 185 of 185 spot-checked values matched source exactly
  • •Designed session-isolated, zero-persistence architecture so Scholar financial data never leaves their device
  • •Added "check the AI response" guidance to every AI output, teaching verification habits alongside the tools

Outcomes

  • ✓97.7% adoption: 259 of 265 Scholars used the Module 1 AI prompts in their assignments
  • ✓96.5% rated the prompts helpful; zero Scholars rated them "not at all helpful"
  • ✓Zero negative-intent non-use: the six non-users cited operational friction, not AI rejection
  • ✓89% of Scholars now use AI daily or weekly in their actual businesses
  • ✓Observed non-use patterns matched the failure modes predicted in my published SSRN research on AI adoption

Learnings

  • →Adoption is engineered through workflow embedding, not announced through training sessions
  • →Non-technical users adopt AI when it scaffolds their thinking instead of replacing it
  • →Universal adoption AND universal satisfaction is achievable when tools are curriculum-native
  • →Field results validating your own published framework is the strongest research signal there is