Fan intelligence
at scale.

The brief: use AI to do what traditional merch research can't.

A friend who runs a concert merchandising company was onboarding Jelly Roll's 2026 tour. The standard approach — gut instinct, what sold last cycle, what looks cool — had always been the industry default. He wanted to test something different: could AI-assisted research produce better-informed merch concepts than traditional methods?

He brought me in to design and run a research pipeline from scratch. No existing framework. No precedent. The deliverable was a full product concept deck for Manhead, Jelly Roll's merch partner — due before the tour launched.

The underlying research question was deceptively simple: what does this fanbase actually respond to, and how do we translate that into things people will buy at a merch booth?

Four parallel research threads, one integrated pipeline.

I built the research in four parallel tracks, each designed to answer a different dimension of the product question. All four fed into a single synthesis layer before anything went into the deck.

The first track was raw behavioral data. Using Apify, I scraped 341 TikTok videos across 243 creators spanning five years — 267 million plays, 21 million likes, and 1 million shares. This gave me a longitudinal picture of which content types, themes, and moments actually moved audiences, not just which ones got posted.

The second track was comment intelligence. I pulled and coded 25,985 fan comments with 9.7 million total likes — the unfiltered voice of the fanbase in their own words. I built a theme taxonomy, scored comments by engagement weight, and isolated the highest-resonance lyric references, product requests, and emotional language patterns.

The third track was geographic and timing intelligence. I mapped 76 geo-tagged posts to identify where content was being created and consumed, which markets showed the deepest signal, and what day and hour combinations drove the highest average plays. Friday at 2pm CST turned out to be 3x the dataset average — a scheduling finding with direct operational value.

The fourth track was influencer efficiency analysis. Rather than ranking creators by follower count, I built a plays-per-follower score to identify accounts that punched above their weight. A creator with 1,785 followers who generated 2.8 million plays at a 1,569× efficiency ratio is far more valuable for seeding than a verified account with millions of followers and moderate engagement.

Once all four tracks were synthesized, I used the findings as a creative brief — mapping emotional themes to design territories, lyric data to specific print candidates, and geographic signals to inventory recommendations by market.

I then generated Midjourney prompts for each concept, curated the visual output, and assembled everything into a full lookbook deck with data rationale for each product.

The fans were already telling us what to print.

The most important finding wasn't a number — it was a pattern. The highest-resonating content across the entire dataset wasn't about music. It was about character: generosity, redemption, faith, and the recovery narrative. Fans follow Jelly Roll because of who he is, not just what he makes. Merch designed around his philosophy would outperform merch designed around his image.

The comment data surfaced specific lyrics fans were already treating as tattoo-worthy. Three lines had explicit tattoo intent confirmed in the comments, with combined like counts in the tens of thousands. This isn't preference data extrapolated from a survey — it's direct behavioral evidence of emotional attachment strong enough to put on your body.

The theme breakdown also revealed two underserved markets. Faith-themed content accounted for 12.3% of all comments — "Amen" was the fourth most common meaningful word across nearly 26,000 comments — yet faith was largely absent from the existing merch footprint. Similarly, recovery-community comments averaged 968 likes each, second only to Grammy-related content, signaling a deeply engaged segment with purchasing intent and no dedicated product.

A secondary finding with brand protection implications: 261 comments referenced a "fan card" costing $900–$6,000 with patterns consistent with a scam impersonating the Jelly Roll brand. Fans were preparing to pay. This was flagged to the team as an urgent brand integrity issue alongside the merch recommendations.

19 product concepts, each with a data rationale.

The final deliverable was a full merch lookbook presented to Manhead — 19 product concepts across tees, hoodies, hats, a hockey jersey, and a metal keychain. Every concept included the specific data signal that justified its existence: the comment volume, the like count, the hashtag trajectory, or the geographic pattern that made it a priority rather than a guess.

The project established a repeatable methodology: scrape social data at scale, synthesize behavioral signals into design territories, generate visual concepts from the territories, and deliver everything with the evidence trail intact. The approach is now transferable to any artist, any tour, any product category where fan behavior leaves a data footprint.

More broadly, this project demonstrated that consumer research methods — when applied rigorously to entertainment and fan culture — can surface insights that traditional industry instinct misses entirely. The fans were already telling us what to print. We just needed to listen at scale.