AI’s New Training Data: Your Old Work Slacks And Emails

Key Concepts

• Operational Exhaust: The "digital leftovers" of a company, including Slack messages, emails, Jira tickets, and internal documents.
• Agentic AI: AI models designed to perform complex tasks and "do work" rather than just generating text.
• Data Richness: A metric used to value data based on internal traceability and cross-platform linkages (e.g., a Jira ticket linked to a specific code commit).
• PII (Personally Identifiable Information): Sensitive data that must be scrubbed from datasets before they can be sold to AI labs.
• Asset Hub: A platform by Simple Closure designed to facilitate the sale of defunct companies' digital footprints.

---

1. The Emergence of "Operational Exhaust" as a Commodity

As AI labs have exhausted the supply of high-quality public internet data (Reddit, Wikipedia, books) by late 2024, they have shifted their focus to private, internal company data. This "operational exhaust"—the daily digital trail of a company’s operations—is now considered a "fossil fuel" for training agentic AI models. To achieve true workplace competence, AI models require examples of real-world workflows, including the "noise" and complexities inherent in professional environments.

2. Case Study: Cello 24

Shana Johnson, CEO of the transcription company Cello 24, utilized the startup Simple Closure to wind down her business. Beyond standard shutdown procedures (taxes, payroll, investor consents), she sold the company’s 13-year digital footprint. This sale provided "hundreds of thousands of dollars," transforming a difficult financial situation into a clean exit. Johnson noted that the data, which included internal communications and project management logs, would now serve a new purpose in training future AI.

3. The Market for Defunct Data

• Simple Closure: Has processed nearly 100 deals for defunct companies, recovering over $1 million for founders. Payouts typically range from $10,000 to $100,000 per company.
• Sunset: A competitor in the space that also acquires defunct company data.
• Valuation Factors: According to Brendan Mahoney (CEO of Sunset), the value of a dataset is determined by:
  • Company size and age.
  • Data Richness: The degree to which documents are linked (e.g., a Jira ticket connected to a code commit is more valuable than an isolated email).
  • Industry sector: Finance and healthcare data command a premium.

4. Methodologies and Technical Challenges

• Data Sanitization: A critical step in the process is the removal of PII. Simple Closure is currently refining its Asset Hub platform to ensure this scrubbing process is "rock solid" before a wider rollout.
• Simulation Environments: Companies like Micro One are creating "mock holding companies" (e.g., the product "Roots") where AI agents can practice skills in simulated environments, further driving the demand for real-world workplace data to populate these simulations.

5. Ethical and Privacy Concerns

The practice of selling internal communications has drawn criticism from privacy advocates. Mark Roenberg (Center for AI and Digital Policy) argues that:

• Even if employees signed away intellectual property rights, it does not necessarily grant employers the right to sell private internal communications to third parties.
• There is a significant expectation-of-privacy issue, as employees likely never anticipated their Slack messages or internal frustrations would be repurposed for AI training.

6. Notable Quotes

• Shana Johnson: "I’m still a bit emotional about shutting the company down but it’s cool to think that our data could be useful live on and help other people."
• Dory Yona (CEO, Simple Closure): "There’s a feeling of a gold rush from these companies trying to get their hands on real-world data."
• Mark Roenberg: "I think the privacy issues here are quite substantial."

---

Synthesis

The transition of "operational exhaust" into a valuable asset represents a new frontier in the AI arms race. As AI labs move toward agentic models, the demand for high-fidelity, real-world workplace data has created a secondary market for defunct companies. While this provides a lucrative exit strategy for founders and a vital resource for AI development, it simultaneously raises significant ethical questions regarding employee privacy and the repurposing of internal communications. The industry is currently balancing the technical necessity of "data richness" with the legal and moral complexities of data sanitization and ownership.