AI Won’t Save Your Label. Your Operations Might.

Sammy Andrews published a piece in Music Business Worldwide last week that I’ve re-read three times. Her argument: AI advantage in music has nothing to do with automating creativity or predicting hits. It is about who wastes less money, allocates capital more accurately, and makes fewer bad decisions, faster. Read the full piece here.

She’s right. And I want to take it further — because we didn’t wait for AI to fix this problem. We started thirteen years ago.

The industry has the conversation backwards

Every panel at every conference talks about AI and creativity. Can AI write a hit? Should labels license training data? Will synthetic content cannibalize human artistry?

Meanwhile, most independent labels can’t answer a basic question: how much did a specific release earn last quarter, net of all splits and recoupment, by territory, within 24 hours?

That’s not an AI problem. That’s a spreadsheet problem. And you can’t build intelligence on top of a foundation that doesn’t exist.

Andrews calls this out directly:

Release forecasting at most organisations still relies on recent comparables, editorial signals and territorial intuition rather than machine-learning models trained on streaming behaviour across similar catalogues.

That’s the polite version. The blunt version: most labels forecast revenue by looking at what happened last time and hoping this time is similar. No modeling. No pattern recognition across catalogs. No system that learns from every release on the platform.

AI doesn’t fix that. Clean data, automated pipelines, and structured operations fix that. AI just makes the gap between labels who have it and labels who don’t even wider.

What we learned building the foundation first

We started iGroove in 2013 as a distribution company. We built the infrastructure because nothing on the market handled royalties, splits, and payouts the way a real label needs them handled — with the precision standards I’d seen in Swiss banking systems.

That decision — to build the foundation before building anything smart on top of it — is the reason we can do things today that labels on patchwork systems can’t:

• Revenue estimates with 95%+ accuracy from public data alone. Not because we have a better model. Because we have 10 years of proprietary distribution data to train against. The model is only as good as the data underneath it.
• Anomaly detection across the full revenue chain. Andrews mentions this as something finance teams should adopt from banking. We’ve been running it for years. Duplicated identifiers, abnormal payout patterns, territorial variance — flagged before anyone has to look for them.
• Natural language queries against catalog data. “Which of my releases are growing fastest in Germany this month?” answered in seconds with a chart. Not because of GPT. Because the data layer is clean enough for a language model to query without hallucinating.
• Social-to-streaming correlation. Andrews calls for “distinguishing correlation from causation” in marketing measurement. We do this daily — benchmarking every TikTok and YouTube post against the artist’s own baseline to separate signal from noise. But this only works because every data point is tracked, timestamped, and linked to the same identifier.

None of this is “AI.” All of it is impossible without the operational foundation that most labels haven’t built.

The snake oil problem

Andrews writes something that needs to be said louder:

Most music-specific AI products represent snake oil — generic models repackaged for music with minimal understanding of rights, accounting, or platform mechanics.

I see this every week. A startup wraps ChatGPT in a music-themed UI, trains it on publicly available data, and pitches it as “AI for labels.” The output looks smart until you check the numbers. The forecasts are off by 40%. The royalty calculations don’t account for advance recoupment. The territory splits ignore sub-publishing.

The problem isn’t the AI. The problem is that the AI doesn’t have access to the operational reality of how music money actually moves. And no amount of prompt engineering fixes that. You either have the data pipeline or you don’t.

What “AI readiness” actually means

Andrews defines it well: a single source of truth for catalog data, stable identifiers across systems, clear territorial and rights metadata, reconciled revenue streams, and governance enabling internal data movement.

I’d translate that into what it means for a 15-person independent label:

• Can you answer any financial question about your catalog without opening a spreadsheet?
• Do your artists have real-time access to their own data?
• Are your royalty calculations automated or does someone still do them manually each month?
• When you sign a new artist, can you model the deal economics in minutes or does it take a week?
• Is your release QC a human bottleneck or an automated pipeline?

If the answer to most of these is no, AI isn’t your next move. Infrastructure is.

The compound effect no one measures

Andrews’ closing line is the one that sticks:

The impact is visible not in grand narratives, but in day-to-day decisions that quietly compound.

This is what I’ve been trying to explain for thirteen years. The label that runs automated payouts doesn’t just save time — it earns trust from artists who see their money on schedule. The label that has real-time revenue data doesn’t just make better decisions — it moves faster than competitors who are still waiting for quarterly reports. The label that models deals in minutes doesn’t just close faster — it signs the artist that the other label was still building a spreadsheet for.

None of these advantages require AI. All of them are amplified by it. But only if the foundation is already there.

The labels chasing AI without fixing operations will spend money on tools that produce impressive demos and unreliable outputs. The labels that fix operations first will discover that most of what AI promises, they already have — and the parts they don’t have will actually work.

We’ll show you what the foundation looks like on your own catalog. One artist, 48 hours. Start here.