# cornell-ai-hub-100k-unidentified-payments-2026-06-15

## Veille

Case study published by the **Cornell AI Innovation Hub** (June 15, 2026): how a two-semester collaboration between the AI Hub, graduate students, and Cornell's Treasury team turned a time-consuming manual investigation into an AI tool that **recovered $100,000** in unidentified payments on a first batch. A successful **AI4Business** use case (financial process) that illustrates the **Leader-Lab-Crowd** framework of **Ethan Mollick** almost point by point: the **AI Hub** plays the role of the **Lab** (a central, ambidextrous team of technologists plus students); **Treasury** (Cheryl Barnes, Marie Graves…) is the **Crowd** carrying business knowledge and the real pain point; and the **$100,000** constitutes the **visible reward** (vivid win) that anchors adoption — exactly the incentive lever Mollick considers decisive. Key method: **"context first, then plan, then build"** via **Claude Code Plan Mode**, a chain of **fuzzy matching → Gemini Enterprise Web Search → Claude synthesis**, all within the governed **Cornell AI Gateway**. *"The $100,000 is a start."*

## Titre Article

How Cornell Recovered $100,000 in Unidentified Payments With AI

## Date

2026-06-15

## URL

https://innovationhub.ai.cornell.edu/articles/how-cornell-recovered-100000-in-unidentified-payments-with-ai/

## Keywords

Cornell AI Innovation Hub, unidentified payments, payment reconciliation, treasury, finance, AI4Business, Leader-Lab-Crowd, Ethan Mollick, reward, incentive, vivid win, Lab, Crowd, business knowledge, Claude Code, Plan Mode, context first, fuzzy matching, Gemini Enterprise Web Search, Claude synthesis, Cornell AI Gateway, data governance, know your data, backtest, escheatment, Oracle GL, Kyriba, n8n, /treasury skill, Claude pilot, change management, automation, measurable ROI

## Authors

**Pete Stergion** — Desktop Engineer au Cornell AI Innovation Hub, co-tech lead du projet (avec Phil Williammee). Article institutionnel signé de l'AI Hub.

## Ton

**Profile**: an institutional case study / *progress report* written in the first-person plural (*"we"*), published on the Cornell AI Innovation Hub blog. Target audience: operational teams at universities and companies, IT leaders, AI change-management practitioners. Register: **sober, pedagogical, and honest**, medium technical level (assumes familiarity with fuzzy matching, pipelines, GL, backtesting) but readable by a non-engineer.

**Style**: step-by-step narration (*The Problem / The Team / How We Built It / The Pipeline / The Results / What's Next / The Takeaway*), alternating narrative and figures, with a **controlled honesty** that reinforces credibility: the tool's **limitation** is documented (vendors billing multiple departments), it is noted that *"resolution still depends on treasury staff follow-up,"* and governance is emphasized (*"know your data,"* Cornell AI Gateway, PII stripped). No overselling: the closing thesis is not "AI works miracles" but *"the project worked because the team put in the groundwork."* Shareable tagline: *"The $100,000 is a start."*

## Pense-betes

- **The headline number**: **$100,000** recovered on a **first batch** — 23 departments contacted, **7 responses**, **5 confirmed payments**. Active backlog ~**$1M** (historical peak **$4M**) across a few hundred transactions. Regulatory stake: New York State law mandates **escheatment** (turning over unclaimed funds to the state) if not resolved in time.
- **The insight that structures everything**: **vendor names appear on 99%** of unidentified payments, while **invoice/PO numbers appear on less than 4%**. → the bet is on the vendor name, not the missing identifier.
- **Mollick reading (core of the ask)**: the case illustrates the **Leader-Lab-Crowd** framework ([[mollick-making-ai-work-leadership-lab-crowd-2025-05-22]]).
- **Lab** = the **AI Hub** (Pete Stergion, Phil Williammee, cohort of students): the central team that *packages* the solution.
- **Crowd** = **Treasury** (Cheryl Barnes, Marie Graves, Kevin Mooney, Debra Federation): business knowledge + data (Kevin provides **3 years of Oracle GL, 10,000+ records**). It's the on-the-ground pain (up to **half a day of work** on this problem) that becomes the opportunity.
- **Reward / vivid win**: Mollick recommends **rewarding employees who discover transformational uses** and providing a **vivid picture** of the AI future. The **$100,000** is precisely that concrete, legitimizing proof — it makes the value tangible and desirable.
- **Anti-layoff reassurance**: AI **removes the drudgery** (email lookups, googling abbreviations, cross-referencing) rather than replacing staff; the stated goal is *"a sustainable workflow the team owns and operates directly"* (onboarding into the Claude pilot + the **`/treasury`** skill). This is the answer to Mollick's *"reassure your workers."*
- **"Context first, then plan, then build" method**: via **Claude Code Plan Mode**, *all* the context is loaded (background, manual process, semester notes, student prototypes, sanitized data); Claude Code **proposes a complete architecture to validate before writing a single line**. Result: from a **semester of notes to a working tool in a single session**.
- **Pipeline (3 steps), exposed as a *skill* in Claude Code**: (1) **historical matching** (fuzzy matching against the GL; filters noise words like *Inc/LLC/Corp*; a token-overlap *gate* to avoid false positives); (2) **vendor lookup** via **Gemini Enterprise Web Search** (within the Cornell AI Gateway); (3) **Claude synthesis** (weighs the evidence → likely department, **confidence level**, suggested contact). Output: **Excel sorted by confidence**, in a few minutes. Input: **Kyriba** (TMS) export.
- **Rigorous backtest**: 3 scenarios over **9,131** already-resolved payments, answers hidden. Recurring vendors: **97%** (fuzzy matching alone, 500 test payments) → **100%** with the full AI chain. Unknown vendors: **76% → 100%** once Gemini + Claude are activated. **Documented limitation**: vendors billing multiple departments (right vendor but not always the right departmental account).
- **Governance**: *moderate-risk* financial data, **Cornell AI Gateway**, **not used to train external models**, restricted access, **PII stripped**, a *"know your data"* approach. Models tested along the way: **Gemini, GPT, Claude**; initial prototyping in **n8n**.
- **Useful counter-narrative**: to invoke against the narrative that "AI doesn't create value in business processes." Here it does — **because** Lab + Crowd + groundwork. *"When it came time to build, the AI had everything it needed to design a real solution rather than a generic one."*

## RésuméDe400mots

The Cornell AI Innovation Hub recounts (June 15, 2026) how a two-semester collaboration made it possible to **recover $100,000** in unidentified payments using AI. The problem: every year, Cornell receives hundreds of wire transfers and ACH payments without enough information to route them (no invoice number, vague vendor name). The funds accumulate in a suspense account — active backlog ~**$1M**, historical peak **$4M** — and **New York State law mandates escheatment** if they are not resolved in time. Two treasury staff members were spending up to **half a day** a day on this.

The project's structure illustrates Ethan Mollick's **Leader-Lab-Crowd** framework. The **Lab** is the **AI Hub** (Pete Stergion and Phil Williammee, co-tech leads, plus a cohort of students). The **Crowd** is **Treasury** (Cheryl Barnes, Marie Graves, Kevin Mooney, Debra Federation), holder of the business knowledge and the data — Kevin provides **3 years of Oracle GL history (10,000+ records)**. The student analysis surfaces the key insight: **99%** of payments carry a vendor name, versus **less than 4%** an invoice number.

The build follows a **"context first, then plan, then build"** discipline: via **Claude Code Plan Mode**, the team loads all the context (notes, manual process, prototypes, sanitized data); Claude Code **proposes an architecture to validate before writing any code**. A semester of notes becomes a **working tool in a single session**. The **Python pipeline** (exposed as a *skill* `/treasury`) chains three steps: **fuzzy matching** against the GL (filtering out noise words like Inc/LLC/Corp), **vendor lookup** via **Gemini Enterprise Web Search**, then **Claude synthesis** producing, for each payment, a likely department, a **confidence level**, and a contact. Output: an Excel file sorted by confidence, in a few minutes — all within the governed **Cornell AI Gateway** (PII stripped, no external model training).

The **backtest** (9,131 resolved payments) shows **97% → 100%** accuracy for recurring vendors with the full AI chain, and **76% → 100%** for unknown vendors. Documented limitation: vendors billing multiple departments. Operational result: 23 departments contacted, 7 responses, **5 payments = $100,000** confirmed.

Beyond the figure, the case is a **counter-example** to the narrative that "AI doesn't create business value": it does here, because a **Lab**, an expert **Crowd**, and **real groundwork** came together. And the $100,000 plays the role of the **visible reward** Mollick prizes — the tangible proof that legitimizes and spreads adoption, by removing the drudgery rather than the jobs. *"The $100,000 is a start."*

## GrapheDeConnaissance

- Pete Stergion —travaille_chez→ Cornell AI Innovation Hub (ORGANISATION, 0.96)
- Cornell AI Innovation Hub —collabore_avec→ Cornell Treasury Operations (ORGANISATION, 0.97)
- Cornell AI Innovation Hub —a_créé→ pipeline de rapprochement des paiements non identifiés (TECHNOLOGIE, 0.95)
- pipeline de rapprochement des paiements non identifiés —utilise→ Claude Code (TECHNOLOGIE, 0.96)
- pipeline de rapprochement des paiements non identifiés —utilise→ Gemini Enterprise Web Search (TECHNOLOGIE, 0.94)
- pipeline de rapprochement des paiements non identifiés —utilise→ Cornell AI Gateway (TECHNOLOGIE, 0.93)
- Claude Code Plan Mode —permet→ passer d'un semestre de notes à un outil fonctionnel en une seule session (AFFIRMATION, 0.9)
- pipeline de rapprochement des paiements non identifiés —mesure→ 100 % de précision au backtest avec la chaîne IA complète (vs 97 % en fuzzy-matching seul) (MESURE, 0.93)
- Cornell AI Innovation Hub —mesure→ 100 000 $ récupérés sur un premier lot (5 paiements, 23 départements contactés, 7 réponses) (MESURE, 0.95)
- Cornell AI Innovation Hub —affirme_que→ les noms de fournisseurs figurent sur 99 % des paiements non identifiés, contre moins de 4 % pour les numéros de facture (AFFIRMATION, 0.92)
- Cornell AI Innovation Hub —est_instance_de→ framework Leader-Lab-Crowd (CONCEPT, 0.85)
- Cornell Treasury Operations —est_instance_de→ framework Leader-Lab-Crowd (CONCEPT, 0.85)
- Ethan Mollick —recommande→ récompenser les employés qui découvrent des usages transformationnels de l'IA (AFFIRMATION, 0.9)
- récupération de 100 000 $ —est_instance_de→ récompense visible de l'adoption IA (CONCEPT, 0.82)
- Cornell AI Innovation Hub —recommande→ une approche « contexte d'abord » (groundwork avant build) (AFFIRMATION, 0.88)

---
Canonical: https://www.thekb.eu/en/fiches/cornell-ai-hub-100k-unidentified-payments-2026-06-15/