Why your royalty statement is wrong,
and how AI reconciliation finally finds the missing money.

Why the data is broken

Royalty data flows through dozens of intermediaries across 32 or more territories with inconsistent metadata, missing ISRCs, missing ISWCs, conflicting work-recording links, and rampant misattribution. The systems that report up to publishers and labels were extended onto the streaming economy without anyone rebuilding the foundations. ISWCs were standardised in 1995; the streaming economy is twenty-five years younger. Most of the leaks happen in the gap.

A catalogue with 6,000 works typically has 11% of activity silently mis-routed at any given time. Some of that is metadata that was wrong at registration and never corrected. Some of it is contract changes that PROs never picked up. Some of it is publisher mergers that left orphaned splits in the originating PRO's database. Some of it is derivatives that exist on streaming platforms but aren't linked back to the original work in any system anyone is reading. The aggregate number — that 11% figure — is consistent across the eight catalogues we've reconciled, plus or minus three points, regardless of genre or label size.

The structural problem is that no single party in the royalty chain has both the incentive and the data to fix it. PROs are paid on what they distribute, not on what they don't. Distributors pass through what they receive. Publishers see only what their administration tools show them, and those tools are usually a single view onto data the publisher cannot independently verify. Until very recently, the answer to "where is the missing 11%?" was "nowhere, because nobody is allowed to look."

Where money goes missing

Four specific categories absorb most of the leakage. Each one has a different mechanism, and each one responds to a different kind of intervention.

Unmatched neighbouring rights in continental Europe. Neighbouring rights (NR) — the recording-side rights distinct from publishing — generate substantial revenue across Germany, France, the Benelux, and Scandinavia, often more than a small label realises. NR money is collected by territory-specific agencies (GVL, SPRD, SENA, etc.) and distributed against registered performer and producer claims. Recordings that aren't registered with the right agency in the right territory accrue revenue that sits in suspense accounts. We see this category produce 1–3% of catalogue revenue in recovered claims, almost mechanically.

Black-box mechanicals in Latin America. Mechanical royalties in much of Latin America are administered with a delay measured in years, sometimes decades, and large pools of money sit unclaimed because no one with rights-holder authority knows to claim them. The categories of work most affected are the obvious ones — back-catalogue, deceased artists, indie labels that don't have local administration partners. Recovery in this category is slow and paperwork-heavy but the return is real.

Royalties paid on derivatives that didn't back-propagate. When a fan-made cover or short-form remix is monetised on a platform, the platform's content-ID system sometimes pays the original rights-holder and sometimes doesn't, depending on whether the derivative was flagged in the platform's matching system and whether the matching was confident enough. The cases where it doesn't, but should have, are exactly the cases where Nomad Listen surfaces a derivative that the platform missed. Closing that loop typically produces 0.5–2% of catalogue revenue.

Sync placements not registered with the right PRO. Sync money flows through a different pipeline from streaming money, and the registration touchpoints are different. A sync placement that wasn't reported to the right PRO at the right time — particularly across territories — generates revenue in the territory where the placement aired but doesn't flow back to the publisher who should be receiving it. We see this category produce 0.5–1% on average, higher for catalogues with active sync activity.

The reconciliation pipeline

Nomad Rights addresses these categories with a single continuous pipeline. The pipeline normalises usage data across PROs, MLC and neighbouring-rights agencies; matches it against an AI-resolved works graph; and surfaces the discrepancies as actionable claims with the evidence attached. The intent is that no money in any of the four categories sits in a suspense account for more than one reporting cycle.

The works graph is the centrepiece. It models works and recordings as nodes, splits and territory carve-outs as edges, and contract clauses as edge metadata. Where the source data contradicts itself — and it almost always does at scale — the AI resolver proposes the most likely split based on the contract corpus, PRO filings, and the publisher's own historical patterns. Every AI-resolved split is flagged, never silent; if you read a claim that came out of Nomad Rights, you can see whether the underlying graph was contract-evidenced or AI-resolved, and to what confidence.

The reconciliation step runs every time a statement lands — PRO statements, MLC statements, distributor statements, neighbouring-rights distributions. Each statement line is matched against an expected entry in the graph. If the expectation matches, the line is filed. If the line is short of the expected amount, a claim is generated. If the line is over, a claim-reversal is generated and we surface it to the team for inspection. The state machine that tracks each claim through dispute, resolution and payment is the operational core of the whole product.

What labels recover

On the eight catalogues we've reconciled in pilot during 2025, recovered annual royalty ranged from 3.4% to 9.7% of stated revenue. The median was 5.1%. The high-end cases (the 9.7% indie label and the 7.0% neighbouring-rights specialist) had significant unclaimed NR balances at the start of the engagement and substantial derivative-driven revenue that was being missed entirely. The low-end case (the 3.4% small indie) had already been administered carefully but still had specific sync placements and black-box mechanicals that hadn't been recovered.

Recovery is sensitive to engagement length. The numbers above reflect first-year recovery; the second year typically runs lower because the obvious leaks have been closed, but it doesn't drop to zero. The graph keeps getting better as new statements feed back into it, and the closer-to-zero error rate makes future claims faster to surface. By year three on the longest-running engagement, total cumulative recovery sits at about 11% of starting annual revenue, with the marginal year still adding 2–3%.

Operational discipline

Reconciliation is not a quarterly exercise. The whole product is designed around the discipline of continuous reconciliation — every monthly statement checked against the works graph, every discrepancy generating a tracked claim, every claim followed through to payment or to a written explanation of why no payment was due. The model is closer to financial audit than to data analytics. The people who run it on our side are accountants and rights specialists, not data scientists.

That operational discipline is, frankly, where the rest of the industry is going to have to follow. The structural breakage isn't getting fixed by metadata committees. It's getting fixed by graph-native reconciliation, evidenced claims, and the institutional patience to track each missing line through to a payment outcome. That's the work. We've systematised it; we haven't replaced it. The teams using Nomad Rights are still the ones doing the rights work — we've just given them a system that lets them do it across 32 territories simultaneously rather than one PRO at a time.