Katy Simmons
00Director, AI Operations · 2024 to 2026

Where AI Operations became a function.

I did not arrive at SoundCloud with a tidy definition of AI Operations.

I got there by solving enough real AI problems to realize the model was rarely the hardest part.

99.36%
MAIA detection accuracy
~700K
Tracks scanned per day
6
Functions on one roadmap
1
Enterprise AI access layer
  • 01 SCOPE
  • Customer Operations
  • Platform Integrity
  • AI-Generated Content
  • Fraud
  • Legal
  • Copyright
  • Distribution
  • Monetization
  • Marketing
  • Internal Knowledge
  • Enterprise AI
  • Governance

01 SCOPE labels scroll automatically. They pause on hover or keyboard focus, and there is a button to stop or start the motion.

01 SCOPE labels motion playing
02Flagship case study
01SoundCloud / 2024–2026

MAIA: AI-Generated Content Detection at Platform Scale

Once you can detect AI-generated music, what should actually happen to the track?

Detection accuracy
99.36%

Detection accuracy

Precision
99.75%

Precision

Tracks scanned per day
~700K

Tracks scanned per day

Latency per track
5–7s

Latency per track

Trust & Safety / AI Product / Evaluation / PlatformView the work
03Major work
02SoundCloud / 2024–2026

Building AI Operations from the Work Up

I started with the work: support, trust and safety, legal, marketing, internal knowledge. Each project taught me a condition AI needed to succeed. I turned those conditions into an operating model, shared infrastructure, and a community of practice, and that became the AI Ops function.

AI Strategy / Governance / Org Design / PlatformView the work
03SoundCloud / 2024–2026

Age Assurance Across Global Markets

Age assurance requirements arrived market by market, each with different definitions and enforcement. I turned a compliance scramble into a product decision framework.

Regulatory / Product / Privacy / RiskView the work
04Also underneath

  • The AI Collective

    An open forum where people showed what they were trying, including the failures. I did not want experimentation centralized. I wanted the learning centralized.

  • Customer Operations

    UltimateAI and Zendesk, AI agents, Help Center, knowledge quality, routing, containment, CSAT, deflection, Looker analytics, human escalation. Measure the journey, not the bot.

  • Platform integrity

    Fake streams, fraud patterns, high-risk accounts, and noise tracks.

  • Legal and copyright

    License summarization, copyright review, and contract indexing.

  • Artist operations

    Distribution and monetization review where the artist impact was the constraint.

  • Enterprise AI

    Centralized access, routing, visibility, cost control, and reusable capability.

A technically excellent model can still produce a terrible system.

All the work →

Contact

Let's get into it.

Ambiguous problem, AI adoption that stalled, an operating model that stopped scaling, a support function that should be a product. That is the conversation I want.

Book a call →Email me →Other ways to reach me →