An independent music publisher, 12 years in operation, manages a roster of 340 active songwriters and artists across indie rock, hip-hop, and Latin genres. The team signs between 18 and 28 new artists per year at advances ranging from $25K to $600K — each one a financial commitment made under meaningful uncertainty.
Their publishing administration software held the complete history of every deal ever signed: advance paid, royalty rate, recoupment timeline, and actual revenue per quarter across 12 years of operation. That deal history had never been connected to anything analytical. Every signing decision was made by the Head of A&R reviewing a scouting platform screenshot alongside a gut read built from 20 years of experience.
A&R discovery tooling has become sophisticated. The platforms the team relied on track more than 12 million artist profiles and answer the discovery question well: who is growing, how fast, and across which platforms. What none of them answer is the deal question:
“Given this artist’s current trajectory — what advance is appropriate, what marketing spend will be required to reach a commercial milestone, and what is the realistic return over a 24-month window?”
Existing A&R tools do not answer that question. They are built for discovery — identifying who is growing and how fast. None of them touch deal structuring, advance calibration, or return prediction. And none of them have seen this publisher’s 12 years of deal outcomes: which advances recouped, which trajectories translated into commercial success for their specific genre mix, which signals at signing time predicted each outcome. That knowledge sat entirely in the Counterpoint database — never queried analytically — and in the Head of A&R’s experience built over two decades.
One was untapped. The other was not scalable.
Coral Tree built a deal outcome model trained on the publisher’s own historical portfolio, with streaming and social signals ingested as real-time input features via API.
Building the Training Dataset.
The engagement started with a structured data extraction from Counterpoint and the publisher’s deal files. Every historical signing was reconstructed as a labeled record: the artist’s streaming metrics and social signals at the time of signing, the deal terms, and the actual outcome at 12, 24, and 36 months. Recoupment timeline. Whether the deal was renewed.
Raw streaming numbers are weak predictors of deal outcomes. What the model learns from are the shapes of those numbers and their relationships to each other — streaming velocity slope, TikTok sound usage acceleration, save-to-stream ratios, platform concentration, geographic listener spread, and co-writer network centrality, among others. These signals capture momentum, listener intent, and commercial connectedness in ways that absolute counts do not.
Model Outputs.
For any prospective artist fed through the system, the model produces three outputs:
- Recoupment probability.
The likelihood of recouping the advance within 24 months at a given advance level — expressed as a probability, not a yes/no.
- Recommended advance range.
Calibrated to the publisher's own historical recoupment rate targets, not industry averages.
- Projected revenue distribution.
Shown at 18 and 36 months as a range with confidence intervals, not a falsely precise point estimate. Artist outcome variance is genuinely high. The tool communicates that honestly.
A separate component models marketing spend sensitivity— what incremental streaming lift has historically been associated with each tier of marketing investment in this publisher’s own past campaigns, not generic industry benchmarks.
The Interface.
The A&R team accesses the system through a lightweight internal tool: search an artist name, pull their current signals automatically, enter a proposed advance and deal structure, and receive all three outputs alongside the key factors that most influenced the prediction.
When a prospective artist differs significantly from the publisher’s historical portfolio, the model flags lower confidence explicitly rather than producing a prediction that looks equally authoritative. The system knows what it has seen. It communicates the limits of that clearly.
The Head of A&R still makes every decision. The model provides a structured second opinion trained on 12 years of the publisher’s own outcomes — challenging assumptions and anchoring the conversation before it begins.
Time from artist identification to deal decision: 6 weeks → 4 days.
(down from 6 weeks)
The compression in deal cycle time is the headline — but what it represents competitively deserves equal weight. Independent publishers operate in a market where the major labels move fast, have more capital, and now control most of the intelligence tools purpose-built for this problem. An independent that can move from discovery to a financially grounded term sheet in four days operates in a meaningfully different competitive position than one that takes six weeks.
In a market where the fastest credible offer frequently wins the signing, that speed advantage is not an operational improvement. It is a structural edge — built on data the major labels do not have access to, because it belongs entirely to this publisher.