How AI Catches Royalty Underpayments a Quick Glance Misses

Why a quick statement review is not enough

Most royalty underpayments do not look dramatic at first. A statement may be broadly correct while still containing a missing track, an outdated ownership percentage, a duplicated deduction, or a payment that arrived under the wrong recording or writer name. When you are looking through hundreds of rows, those errors are easy to miss.

This is where AI royalty underpayment detection can be useful. The practical value is not that AI somehow knows what you should have been paid without evidence. It is that it can compare large amounts of structured information faster and more consistently than a quick manual check.

A producer may receive income from several places: distributor statements for the master, a performing rights organization for public performance, a mechanical royalty service, a publisher or publishing administrator, neighbouring-rights collection, and direct payments from artists or labels. Each source uses different reporting periods, territory names, identifiers, currency formats, and deduction labels. A track can appear as an ISRC in one report, a title variation in another, and an internal catalogue number somewhere else.

Looking at the total payout is useful, but it does not prove the detail is right. A $2,000 payment can still include a $150 shortfall that only becomes visible when you compare the underlying rows against your agreement and previous reporting.

What AI can actually flag

AI works best as a review layer over clean records. You provide the source material: statements, invoices, split sheets, contracts, track metadata, and prior reports. It then helps identify items that deserve a human look.

Unexpected changes in royalties

A system can compare the effective earnings rate for the same track across periods. For example, if a recording generated 500,000 streams in two similar periods and the reported net revenue per stream falls sharply, that is worth investigating. It may be normal: territory mix, subscription type, currency conversion, promotional activity, or recoupment can all change the result. But an unusual change should not be ignored simply because the total statement looks plausible.

The useful question is not, “Why is the per-stream rate different?” There is no single universal streaming rate. The better question is, “What changed in the data, deal terms, territory mix, or deductions that explains this difference?”

Split and credit mismatches

Suppose you produced a track under a signed agreement giving you 20% of net master income. If the label reports $10,000 in net receipts for that recording, your expected share before any separately agreed deductions is $2,000. If your statement shows $1,500, the missing $500 may reflect a valid recoupment, a reserve, an excluded income category, or an error. AI can flag the difference; the contract determines whether it is a real underpayment.

The same applies to songwriting. If a composition has four writers with equal shares, each writer owns 25% unless the split sheet says otherwise. A 20% writer share showing as 25%, or a 25% share showing as 12.5%, may be a registration issue rather than a payment calculation problem. Either way, catching it early matters.

Missing, duplicated, or mislabelled tracks

Track titles are messy. “Night Drive,” “Night Drive (Radio Edit),” and “Night Drive - Edit” may all refer to related assets, or they may be distinct versions with separate identifiers. AI can match likely variations using titles, artist names, ISRCs, release dates, and contributor credits. It can also flag a song that appeared consistently for six quarters and then disappears without an obvious reason.

It can also spot duplicates. Two rows with slightly different titles but the same ISRC, reporting period, territory, and units may indicate a duplicate line item. That is not automatically bad data, but it is worth checking before treating both rows as separate earnings.

Build a process that gives AI something reliable to compare

AI analysis is only as reliable as the records behind it. Before reviewing statements, create one reference file for every track you expect to earn from. Include the track title, artist, ISRC, UPC where relevant, release date, writers, producer points or percentage, master ownership, composition splits, administrator, and the agreement that governs payment.

Keep the original files too. A CSV export is easier to analyse than a PDF, but the PDF statement, contract, and royalty clause remain the evidence if you need to ask a distributor, publisher, collaborator, or label about a discrepancy.

1. Normalize names and IDs. Pick one canonical title for each asset and record common variations. ISRCs and IPI numbers are especially useful because names alone are unreliable.
2. Separate gross from net. A 15% distribution fee, a 20% manager commission, and a 50/50 net-profit split are not interchangeable. Record the order in which deductions apply.
3. Compare like with like. Do not compare a three-month statement against a six-month statement without adjusting for the reporting window and payment lag.
4. Set review thresholds. You might investigate a missing track, a changed ownership share, a duplicate identifier, or a variance above 10% where there is no clear explanation.
5. Keep an audit log. Note the statement date, line item, expected amount, reported amount, contract clause, and follow-up status.

A tool such as CheckMyRoyalty can make this kind of ongoing comparison less manual, but the underlying habit matters more than the software: keep your rights data organized and review each exception with the supporting documents.

How to investigate a flagged underpayment

A flag is a starting point, not an accusation. First, check whether the issue can be explained by timing. Streaming and publishing royalties often arrive months after use, and different territories or income types can report on different schedules. Then check recoupment, reserves, tax withholding, currency conversion, and any permitted deductions in your agreement.

If the difference still does not make sense, prepare a focused question. Include the track name, ISRC or other identifier, statement period, relevant line items, reported amount, expected calculation, and the contract language supporting your position. Avoid sending a vague message saying that “the numbers look wrong.” A specific query is easier for the other side to investigate and easier for you to track.

Finally, prioritize. A $12 discrepancy may still reveal a registration problem affecting future payments, while a larger one-off variance may be fully explained by a reserve or late reporting. The goal is not to challenge every unusual line. It is to catch the errors that a rushed glance would overlook, document them properly, and make sure your catalogue is paid according to the deals you actually made.