What is this?

Since mid 2023, there has been an operational stand-still with funding from Nashville's Metro Arts Commission, the primary funding mechanism for arts organizations and artists in Nashville. Only starting in late April, early May of 2024 has there been movement with allocated funding pledged in FY23Q1 being released. Part of the stated reason for this delay were accusations of financial mismanagement within the Metro Arts Commission leadership, accusations that came from the leaders of Metro Finance and Metro Legal. However, publicly available documentation was not released to clarify or substantiate these statements, besides a cursory line-item budget showing a projected FY24 deficit. This is my attempt to present as much information about the spending patterns of Metro Arts in the past few years to the broader public for review and discussion.

Table of Contents

All Metro Arts Expenses from July 1, 2022 - Feb 14, 2024

All Recipients of Metro Arts Monies from July 1, 2022 - Feb 14, 2024

All Arts Contracts from Metro Arts July 1, 2022 - Feb 14, 2024

Historic (1988-2018) Metro Arts Grant Spending

Historic (1988-2018) Metro Arts Grant Spending Pivot Table

Why I Did This

As a former city government employee that worked in equitable grants funding, I have been concerned that a lot of discussion has been going around about the way Metro Arts funding has been used in the past few years, but the parties involved were not producing a coherent accounting of the budget in question. I found this strange since this is all obstensively public information. I submitted a FOIA request and spent a few days structing the data using AI/Machine Learning, and this is the result.

**THIS IS NOT AN AUDIT**

Something I want to make very clear is that this presentation of financial transaction is not no way, nor is attempting to be, a valid financial audit. This is a cursory, amateur attempt to organize the financial documents of a Metro Nashville department which I am in no formal way connected to, and I am not a CPA nor particularly good with my own finances. I am good at using machine learning (aka Artificial Intelligence) to process large amounts of data, and at organizing things in database formats to be able to "ask questions" of it. If you have any interest in the numbers that appear here, I strongly urge you to click through to the "Transaction PDF Link" on a given transaction to read the underlying document before taking my work for what is contained in that.

If you want to do an official report...

Begin with the FOIA invoices and purchase orders here.

FOIA Request

On March 13, 2024 I submitted a Freedom of Information Act (FOIA) request. The parameters of my request can be found in footnote 1. On May 2 Metro Finance responded to my FOIA request with a link to 1,276 PDF documents of Invoices and Purchase orders, 1,211 of which where in the relevant time period, and one file that was corrupted.² Metro Finance provided redacted versions of the documents, removing sensitive financial data before providing to me.

How was this information processed?

I was able to process the 1,276 PDFs in less than two days by used machine learning, namely OpenAI's computer vision and unstructured data processing via API, and then storing that data in an Airtable database for easy review and analysis. Prior to being fed into Open AI, the documents were converted to WEBP files via PDF.co's API for easier processing by Open AI's computer vision as the documents contained both scanned PDFs and searchable PDFs, which can cause problems for machine learning tools. I used OpenAI's GPT-4-0613 model to process the data as it appeared in WEBP format and output structured variables for each of the documents. The Open AI model uses "Computer Vision" combined with "Optical Character Recognition" to "see and read" the image files ( The prompt instructions for parsing the data are in the footnotes.³ I manually cleaned the data to remove duplicate organizations with alternative spellings, and to correct roughly 25 transaction dates.

The most manual part of this process was classifying recipients. This was not clearly indicated from the Metro Finance provided data. Identifying if a recipient of funds was, e.g. a supplier of business products, an independent arts org, a consultant, etc., required a pass of GPT-4 analyzing the description of the transaction, and was also aided by human review from members of Arts Equity Nashville. This was the only aspect of the data-prep process that involved people besides myself. The contribution of the Arts Equity team is greatly appreciated and their insight into the background of the department were greatly valuable.

I also took some data from an IRS database I previously created to help assign organization sizes to arts recipients as this was not in the data provided by Metro Finance. I made some assumptions on how Metro Arts categorizes this from how much groups were historically awarded, and this was my criteria:

By Last Reported Yearly Revenue

Micro: Below $50,000

Small: $50,000 to $100,000

Medium: $100,000 to $500,000

Mid-size: $500,000 to $3 million

Large: Over $3 million

Where errors could come from in this dataset

1.  The list of documents could be incomplete, having not arrived to me from Metro Finance.

2.  Expenses in the dataset provided to me were not actually processed by the Metro Arts department.

3.  Information could have been incorrectly written on the documents originally, rendering this database inaccurate.

4.  Information could have been put into the Metro Arts and Arts Department financial system incorrectly, rendering these calculations inaccurate.

5.  The Open AI computer vision, OCR, and unstructured data processing tool could make mistakes in any number of ways, including on invoice totals, organization attribution, transaction date, etc.⁴

6.  Mistakes in the formatting and calculations in the structure of the Airtable database. The way it is linking records, organizing it by fiscal year, could have errors in it that I have not found.

7.  Types of recipients could be categorized incorrectly, especially where single recipients could serve different functions.

8.  I had to estimate the total amount of salary that the department had out-going. I took the current staffing salary data from data.nashville.gov here.

9.  Many other things I have not accounted for.

¹ I would kindly like to request a copy of all invoices and purchase order, including both those under $25,000 and above $25,000, from the period July 1, 2022 to March 1, 2024 for the Metro Arts Department. On May 2 Metro Finance responded to my FOIA request with a link to 1,276 PDF documents of Invoices and Purchase orders, 1,211 of which where in the relevant time period, and one file that was corrupted.²

The details on the request transactions would include at least the following information:

1 Date of submission

2 Date of payment

3 Vendor Name

4 Vendor ID in R12/iSupplier

5 related contract if any

6 Additional details or notes attached to invoice/submission

7 All other fields you are able to provide that are associated with transaction record IDs.

It is understood that if this exceeds a certain threshold, payment may need to be remitted. That is acceptable.

² 17316495-file0001.pdf 

³ Prompt Instruction for parsing the PDF documents via Open AI's GPT 4-0613:

"""
1. **Transaction Date**

   - **Description:** This is the date on which the transaction occurred, such as when an invoice or purchase order was issued. The date should be formatted as YYYY-MM-DD.

2. **Description of Transaction**

   - **Description:** A general description of the transaction detailing the involved parties, amount, the nature of the payment, and whether it relates to an invoice or a purchase order. Information about payment breakdown by line items may also be included.

3. **Amount**

   - **Description:** The total amount of money involved in the transaction, represented in the format $XX.XX.

4. **Transaction ID**

   - **Description:** A unique identifier for the transaction as recorded in an accounting system. If multiple IDs are applicable, all should be used.

5. **Vendor Name**

   - **Description:** The name of the organization or individual submitting the invoice or purchase order. It is crucial to distinguish this from the recipient of the invoice.

6. **Transaction Type**

   - **Description:** The nature of the document, distinguishing between types such as 'Purchase Order' or 'Invoice'.

7. **Vendor Contact Person**

   - **Description:** The name and contact information of the representative at the vendor's organization directly involved with the transaction.

8. **Approval, Authorization Details**

   - **Description:** Details about who approved the transaction on behalf of the organization, including any relevant authorization information.

9. **Additional Details**

   - **Description:** Any other relevant information that may be of interest to an auditor that was not captured in the other fields.
"""

⁴ Transaction dates appeared to be the most common error, with no sub-totals being incorrect from my spotchecking. Most often the transaction date was off by one decade, with errors such as "9/15/2023" being recorded as "9/15/2013".