Last week, we intended to predict Grammy winners using our AI tech. We used our Hit Potential algorithm to see whether there were any trends in the Best Record of the Year category. Then we used the last three years of awards data to try and pick the Best Engineered, Non-Classical winner, based on our AI-generated recording quality tag. Ultimately, we stopped short of making any concrete predictions, but we could have been a little more bullish in our forecasts, at least in one category.

Best Engineered Album anti-correlates to 'Quality'

We previously noted recent trends in the Best Engineered Album, Non-Classical category. The award is presented to recording and mastering engineers for quality engineering. However, what we found was counterintuitive: nominated albums that scored lower for recording quality by our AI tended to be more successful at the Grammys. Therefore, we predicted that the winner would likely have a lower 'Quality' score.

Our theory held up. At the 2022 awards, Lady Gaga and Tony Bennett’s swing album, Love for Sale, continued the trend. Our ‘Quality’ tag ranked it with the second-lowest score in the field, yet it still won.

But why is that? We asked the Head of Music at Musiio, Olivia Coléon.

“What we could be seeing here is the AI picking up on a stylistic choice by the artists. When we look at the winners in the Best Engineered, Non-Classical category, we often hear the use of retro, lo-fi or 'old school' effects that give tracks a different kind of texture; and these textures are what make the music stand out. These might be subjectively great artistic decisions that resonate with voting academy members and listeners, but they can throw off our Quality tag.”

Senior AI Developer Vignesh Badrinath Krishna explains: “The Quality classifier isn’t only trained to know what a so-called “good” recordings sound like, but also understands what “bad” recordings sound like, so lower scores could indicate that the AI has recognised something out of the ordinary.”

Commercial Director Mack Hampson surmises: “If your album has already been nominated in the Best Engineered category, it’s already passed an exceptionally high bar for quality. I believe those nominated albums with lower Quality scores are more likely to win because they have elements of the unusual, so they stand out against more “perfect” recordings. They may also deliver greater realness – something we’re all looking for in music.”

The unpredictable Best Record of the Year

Record of the Year was the other category we analysed last week. According to the National Academy of Recording Arts and Sciences (the people behind the Grammys), this award honours “artistic achievement, technical proficiency and overall excellence in the recording industry” regardless of sales or chart position.

We analysed nominated tracks from the last three years with our Hit Potential algorithm to see whether scores correlated to Grammy wins. We could see that nominated tracks with lower Hit Potential scores tended to win the category. In fact, in the last three years, the third-lowest scoring track (of eight) had always won.

This seemed like a statistical anomaly because we couldn’t think of an underlying cause for the result. And in data, we must always be aware that correlation does not equal causation. If we had decided to make a prediction based on the flawed logic that the track with the third-lowest Hit Potential score would win, we would have bet on ABBA or Brandi Carlisle this year.

Instead, Silk Sonic took home the award for “Leave The Door Open”. As big fans of the 70s-inspired super duo, we’re delighted, but we’re also glad we didn’t make any predictions. In our analysis, Silk Sonic’s winning track scored third-highest – not lowest – for Hit Potential, which is a complete departure from the previous three years.

This is a lesson in not reading too much into the data, and the limitations of predictions. There were also mitigating factors in 2022 that made predicting Record of the Year virtually impossible using only our Hit Potential algorithm. The judging body changed dramatically, and the number of nominations in the category increased from eight to ten.

Hit Potential isn’t trained to recognise artistic achievement, technical proficiency, or “overall excellence in the music industry”. It’s trained to find future hits – which is a different challenge altogether.

Looking back at recent Grammy wins, one has to assume that voting members of the Recording Academy ask themselves whether nominated tracks are culturally relevant, change the sound of modern music, or whether that artist has gained popularity since a record’s release. The answer is a resounding yes for artists such as Billie Eilish (winner 2020, 2021) and Childish Gambino (winner 2019).

This could shed a little more light on Silk Sonic’s win. As a record that only came about because of Covid touring restrictions, that leant heavily on nostalgia and virtuosic performances, it was a perfect fit for a public looking for reassurance and positivity as the world slowly returns to normal.

AI-based music tagging and nominations

We’ll never be able to perfectly predict who’s going to win the Grammys. Still, perhaps AI tagging could be more beneficial for accurately placing tracks in genre categories – an area that can be a little contentious. In 2022, for example, there was controversy in the classical music world. Jazz/electronic violinist Curtis Stewart was nominated in the Classical Instrumental Solo category alongside more traditional performances, and 2022’s big winner Jon Batiste was nominated for Best Contemporary Classical Composition.

Peers in the classical world claimed these works were miscategorised as classical. But do they have a point? And could AI help?

For Curtis Stewart, the genre data is pretty compelling.

Is 12% Classical enough to warrant a nomination in this category? What might be a more reasonable figure? Only the Recording Academy has the power to decide, but this data could help avoid these sorts of discrepancies in future.

Take aways

Hit Potential is a great way for record labels to identify competitive musical and production talent. But if labels are looking to sign future talents that have award-winning potential, music is only a piece of the puzzle. As we can see in the Best Record of the Year category, non-musical factors are equally important. For example, how much fame an artist has gained since they released the work, whether their music ties in with a current cultural movement (BLM in the case of Jon Batiste), and whether the work has any scope to change the way the public views music.

And, as far as the Best Engineered Album award goes, the data suggests that embracing imperfection is the best strategy that artists and producers have for landing a gong.

To find out more about how you can use Musiio technology, or if you see anything in the data that you think we’ve missed, let us know! Just drop us a message on Twitter, LinkedIn or email

Share this story