AI as Talent Scout: PaRa Music's Gavthi Bet and the Limits of Algorithmic A&R in India

Updated:

In July 2026, a YouTube channel called PaRa Marathi Gavthi uploaded a song titled "Pikel Karand," sung by Kiran Vartha with Sanjana Ravte and produced by the composer RK King. By mid-November it had passed half a million views. That is a respectable number for a Marathi-language village song, but nothing extraordinary. What made it newsworthy was the story told about it: the label behind the channel, PaRa Music, says its in-house AI system pointed it to the genre, the songwriting pair and the release in the first place.

Most coverage of AI in music right now is about machines that generate songs — the lawsuits against Suno and Udio, streaming services labelling synthetic tracks, the argument over training data. PaRa Music is making a different pitch. Its AI does not make music. It tells the company where to look. If that works, it is less a threat to musicians than a threat to the traditional A&R scout. If it doesn't work, it is a familiar analytics dashboard with a new label on it. The gavthi case is a useful test of which one it is.

What PaRa Music Says It Built

PaRa Music announced itself on June 2, 2026 in a press release distributed through ANI, calling itself "India's first AI-powered music company." The release carries an advertorial disclaimer, so it is the company's own account. The founder is Rashna Pochkhanawala, who, according to Music Ally's launch report, was previously an executive vice-president at Saregama, one of India's oldest record labels. Funding came from a consortium of angel and institutional investors led by Apollo Growth Capital. The company has not disclosed the amount.

PaRa Music logo

The company describes its system, PaRaMeter, as an "AI Chief of Music Intelligence." The release says it "does not create music." Instead, it analyses audience signals across platforms and regions to spot emerging demand and to guide catalogue, investment and release decisions. PaRa has promised that its catalogue will be "entirely artist-made."

The business targets are large. PaRa wants a catalogue of 40,000 songs within four years, covering film and non-film music in Hindi and regional languages. Music Ally later reported that the catalogue would span more than 20 languages. Music Ally also noted that a significant part of that target is likely to come from buying existing catalogues rather than recording new songs. Apollo's statement in the release states the investment thesis directly: "Music IP will be one of the defining asset classes of the next decade." So the AI is one part of the plan. Underneath it is a bet on owning rights.

The Gavthi Claim, Step by Step

On November 12, Music Ally South Asia published what the company presented as its first success story. Here is the sequence as PaRa described it:

  1. While reviewing the Marathi-language market, the team found that a large share of the artists and songs gaining traction worked in gavthi, a style from rural Maharashtra. In Marathi the word roughly means "of the village."
  2. PaRaMeter's "artist-level return-on-investment analysis" compared gavthi with Marathi pop. The published sentence is missing its comparison word, but the context indicates gavthi came out ahead.
  3. The system then searched for the most successful creative pairings within the style. It flagged RK King and Kiran Vartha as having "an especially strong record."
  4. PaRa released a series of their tracks, beginning with "Pikel Karand" in July. It later signed both artists to exclusive deals.

Co-founder Pavan Sood's summary: PaRaMeter "put gavthi firmly on our radar, and it won't be the last."

All of this is the company's account, passed on by a trade publication. No ROI figures, sample sizes, comparison groups or contract terms have been published.

What the Public Record Adds

YouTube shows something the press account leaves out. On November 13, 2026, a search for "Pikel Karand" brought up a much older song with nearly the same title. "Karandil Pikel Karand," credited to Deoram More and Sanjana Ravate with RK King, was uploaded about four years ago on another channel and had about 5.6 million views. Around it were dozens of smaller uploads from 2022 onward: covers, DJ edits and tarpa remixes. The tarpa is a wind instrument closely tied to Adivasi (Indigenous) communities in Maharashtra's Palghar region, and several video titles and hashtags refer to that region and those communities. PaRa's July release, which credits Kiran Vartha and Sanjana Ravte, had about 534,000 views at the time of the search. That fits Music Ally's "over 500k." (These view counts come from the author's own check of YouTube search results and change daily.)

Read alongside the press account, this changes the story. PaRa did not discover gavthi, and it did not discover RK King. It found a regional scene that already had a multi-million-view hit with RK King's name on it, recorded a new version with a related lineup, and signed the artists. That may well be smart business. It is also exactly the kind of signal a human scout reading YouTube comments in Palghar could have found. Whether PaRaMeter found it faster, more cheaply or more reliably than people would have is the question the company has not answered.

Why India's Regional Market Invites This Bet

The commercial logic is clearer than the AI claims. The FICCI-EY Media & Entertainment Report 2026, published in March, says India's music sector grew revenue by 10% in 2025. Paid music subscriptions rose 37% to 14.4 million. According to Bollywood Hungama's summary of the report's music chapter, label revenue reached about Rs 59 billion. Digital licensing made up 58% of that. Streaming reached about 178 million people, who generated 5.8 trillion streams, up 15% year on year.

Two details in that data matter most for a company like PaRa.

The Hindi-film share is falling. Film music was 57% of consumption in 2025, down from 65% in 2023. Hindi's share of listening fell from 64% in 2024 to 59%, followed by Punjabi at 9%, Tamil at 7% and Telugu at 6%. Marathi does not appear among the top languages listed. A small language with a fast-moving sub-scene is the kind of long-tail market where a central dashboard could, in theory, beat a scouting team spread across 20 languages.

Growth is coming from outside the streaming apps. EY notes that digital licensing grew by only 2%, while revenue from "other OTT platforms and social media channels" grew faster. For regional music, that mostly means YouTube and short-video platforms. Those platforms took over when India banned TikTok in June 2020. The ban cut off about 200 million users, and Instagram launched Reels in India within a week, followed by YouTube Shorts. That move made views, remixes and dance clips the main public measure of what is catching on. Those are exactly the signals an analytics system can collect.

The press release puts the long-term figure at INR 7,500 crore by 2028, at a 9% compound annual growth rate, citing FICCI-EY. This article has not checked that projection against the full report.

Data-Driven A&R Has a History

Using numbers to choose artists is far older than PaRa. In March 2018, Warner Music Group acquired Sodatone, a Toronto start-up founded in 2016. Its technology combined streaming, social and touring data to track early signs of success, such as fan loyalty and engagement and how likely a track or demo was to go viral. At the time, Music Ally called it "A&R's data moment." The idea had moved from a niche that A&R veterans dismissed at conferences to something a major label would pay for.

The TikTok era then pushed the approach to its limit, and the results were mixed. In a 2023 Billboard investigation, label sources said data can predict a hit single but is far less useful for finding artists who last. Several data-led signings could not produce even two songs that streamed well, and a number were quietly dropped. Music attorney Don Passman described the structural problem: "The companies get the same data, they're all chasing the same artists." When every label watches the same charts, the data stops giving anyone an edge.

PaRa's version differs in one way that matters. It applies the method to regional Indian markets, where the major labels' research teams are thinner and the signals are scattered across languages. Being first in a less crowded market is a plausible advantage. It is a very different claim from AI that can spot talent in general.

The Limits

Data sees what has already happened. The gavthi case shows this clearly. A return-on-investment model trained on past performance can only rank styles that already have a track record. It can find a wave once it is visible. It cannot create one, and it is weakest with music that has no data trail yet. The most valuable thing an A&R scout traditionally does is back something before the numbers justify it, and that is outside what this kind of system is built to do.

Ranking by ROI pushes toward repetition. Spotify's own researchers, in a 2020 study of its users (Anderson et al.), found that listening driven by recommendations was linked to lower diversity than listening users chose themselves. That study looked at listeners, not labels, so it does not prove anything about catalogue strategy. Still, the same pressure is easy to imagine on the production side. If investment goes to whichever "creative pairing" has the best past returns, the result is more of what already works. That is the author's inference, not a measured result.

Rights, credit and splits. PaRa signed RK King and Kiran Vartha to exclusive deals whose terms are not public. The scene it is mining is rural, partly Adivasi, and built on shared melodies that circulate as covers and remixes. In that setting the questions are concrete: who is credited as songwriter when a well-known regional song is recorded again, how royalties from a re-recording are split, and how much bargaining power village-based musicians have against a funded company whose stated thesis is accumulating IP. None of this suggests PaRa has done anything improper. It means the publicity release left out the parts that matter most to the artists.

The claims have not been independently checked. "AI-powered" here is the company's own description. Music Ally passed on the story without seeing the model, the ROI comparison or the data. Until PaRa publishes even rough numbers — how many sub-genres it screened, how many it acted on, and how its picks performed against a baseline — the gavthi story reads as an anecdote picked to promote the company, not evidence that the method works.

What to Watch

The useful test is not whether PaRa can sign a style that already has five-million-view songs. It is whether its next finds arrive before the scene is obvious on YouTube, and whether those artists are still releasing successful records in two years. Also worth watching: how much of the 40,000-song target comes from algorithmic discovery and how much from simply buying catalogues. Using AI to scout talent is a reasonable and much less controversial use of the technology than generating songs. But the AI label on its own proves nothing. Disclosed numbers and artist outcomes over time are what would.

Sources