OpenAI dots: Sponsorship Account Audit Reveals Discrepancies
Peter Yang used dots to review sponsorship partnerships, invoicing, and payment records to identify potential oversights. Task: Peter Yang tasked OpenAI's Dot agent with auditing his company's sponsorship revenue by cross-referencing scattered contract, invoice, and payment data. The agent identified five discrepancies, including an uninvoiced completed partnership with Riverside and a $10,000 mismatch in Google accounts. Human involvement was critical: Yang set the audit scope, and all flagged issues required his manual verification against bank statements before any business action was taken.; tools: OpenAI dots. Human role: Not stated in the source; limitations: This case relies on secondary media reports of a demo video; original contracts and bank records are not public. The $10,000 figure is self-reported and unverified. Prerequisites include access to structured or semi-structured financial data (PDFs, CSVs) and a human-in-the-loop for validation. The agent may fail to report errors clearly, as seen in other tasks, requiring careful monitoring..
Tools
Published on Agent Work Atlas:
Overview
Peter Yang used dots to review sponsorship partnerships, invoicing, and payment records to identify potential oversights.
Workflow
Peter Yang tasked OpenAI's Dot agent with auditing his company's sponsorship revenue by cross-referencing scattered contract, invoice, and payment data. The agent identified five discrepancies, including an uninvoiced completed partnership with Riverside and a $10,000 mismatch in Google accounts. Human involvement was critical: Yang set the audit scope, and all flagged issues required his manual verification against bank statements before any business action was taken.
Reported result
The agent successfully surfaced five potential revenue leaks, including a self-reported $10,000 discrepancy. However, these findings are unverified by independent financial records, and there is no evidence that the amounts were actually recovered or that the agent's output was error-free. The outcome demonstrates detection capability but not confirmed financial impact.
Human role
Not stated in the source
Limitations
This case relies on secondary media reports of a demo video; original contracts and bank records are not public. The $10,000 figure is self-reported and unverified. Prerequisites include access to structured or semi-structured financial data (PDFs, CSVs) and a human-in-the-loop for validation. The agent may fail to report errors clearly, as seen in other tasks, requiring careful monitoring.
Try it
Act as a financial operations assistant. I will provide you with [replaceable inputs: sponsorship contract PDFs, invoice CSVs, and payment logs]. Your task is to cross-reference these documents to identify discrepancies such as completed projects without invoices or mismatched payment amounts. First, perform a dry run on [sample file set] and output a summary table with columns: Project Name, Contract Status, Invoice Status, Payment Status, and Discrepancy Flag. Do not take any action on the flagged items. Wait for my explicit confirmation after I verify the flags against our bank statements. If you cannot access a file or encounter ambiguous data, list it under 'Unresolved Items' rather than guessing. Only proceed to draft follow-up emails for [target users: finance team] after I approve the verified list. Ensure you clearly state if any task failed or was incomplete.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.