A 38-column Shopify export and no idea where to start
The setup
Lumen & Oak sells candles and diffusers that people get weirdly loyal to. Priya runs growth for the brand: one marketer, one founder, and a Shopify store doing about 2,000 orders a month. Her quarterly goal was simple to say and hard to do. Stop blasting the whole list and start talking to customers like they’re different people. Because they are.
The problem
She pulled the customer export and froze. Thirty-eight columns. Lifetime spend, order count, days since last order, average order value, discount usage, first product category, email engagement score, region, refund count, on and on. Some were obviously useful. Most were noise. A few looked useful but were basically duplicates of each other.
“I don’t have a data team,” Priya said. “I had a spreadsheet and a gut feeling, and the gut feeling kept changing.” Every time she tried to pick variables to segment on, she second-guessed herself. Cluster on spend and frequency? Add recency? What about category? Pick wrong and the segments come out mushy and useless, and she wouldn’t even know they were wrong.
The turning point
She dropped the CSV into Divisio and opened AI variable suggestions. Instead of guessing, she let it read the shape of her data first.
The AI scanned each column’s profile, how it’s distributed, how much it varies, how much it overlaps with the others, then came back with a shortlist. It recommended segmenting on average order value, purchase frequency, days since last order, and discount dependence, and it said why: spend and frequency carried the most independent signal, recency split the list cleanly, and discount usage separated full-price loyalists from sale-chasers. It also told her what to leave out. Region barely varied, and refund count was too sparse to mean anything.
That last part mattered as much as the picks. It wasn’t a black box handing her an answer. It was a reason she could actually agree or disagree with.
How she did it
- Dropped the 38-column Shopify export straight into Divisio, no cleanup.
- Opened AI variable suggestions and read the recommended columns and the reasoning.
- Accepted three of the four picks, dropped discount dependence for v1 to keep it simple.
- Ran pattern detection on the suggested variables and let the elbow chart point her to five segments.
- Read the heatmap summary to see how each segment differed from the average customer.
The payoff
The five segments were sharp the first time, with no re-running it ten times on different inputs. One jumped out immediately: a small group with high AOV, low frequency, and long gaps since their last order. Premium buyers who’d drifted. Priya built a single “we saved your scent” win-back email for exactly that group.
That one email pulled a 31% open rate against her usual 19%, and reactivated enough lapsed premium buyers in three weeks to cover the year of Divisio twice over. But the real win was quieter: she stopped guessing which columns to trust.
“I went from ‘which of these 38 things matters’ to a defensible answer in about ten minutes. It didn’t just pick for me. It told me why, so I could actually own the decision.”
Priya, growth marketer at Lumen & Oak
Feature spotlight: AI variable suggestions (Pro)
Not sure which columns to segment on? Divisio’s AI reads your column profiles and recommends the ones carrying the most signal, and tells you why, including which columns to skip. It turns the scariest part of segmentation, the blank-page “where do I even start,” into a shortlist you can sanity-check in minutes.