How Juniper Coffee grew repeat orders by 31% with cohorts

Juniper Coffee roasts beans in a converted garage in Leeds and sells them online to about four thousand regular customers. Eighteen months ago, the founder, Amira Okafor, knew her sales were growing. She did not know why people stopped buying.

Today, repeat orders are up 31%. Here is how she and her team of five used cohorts in Tally to find out what was going wrong, and what they changed.

The question

Juniper's customers usually buy a bag every three or four weeks. Most of them come back. Some do not, and Amira could not see a pattern.

"I had a spreadsheet of orders," she says. "I could see the total going up. I could not see that the people who bought in January were disappearing in March, because new people were filling the gap."

What a cohort shows

A cohort is a group of customers who started at the same time, followed over weeks or months. Instead of asking "how many orders did we get in March?", a cohort chart asks "of the people who first bought in January, how many were still buying in February, March and April?"

For Juniper, it looked like this:

Something happened between February and March. Customers who started in March were much less likely to come back.

Finding the cause

Amira's first guess was the price: Juniper had raised prices in March. But the drop affected new customers only, and new customers had never seen the old price.

The answer came from splitting the cohorts by the first product people bought. In March, Juniper had started selling a "discovery box" of four small bags. It sold well. And the people who started with it almost never came back.

"They tried four coffees and liked none of them enough to buy a full bag," Amira says. "Or they liked all of them and could not choose. Either way, the box was a dead end."

The box was our best-selling new product and our worst product for the business. Without cohorts, I would never have known.

Amira Okafor

What they changed

Juniper made three changes, one at a time, and watched the cohorts after each.

1. A follow-up after the box. Two weeks after a discovery box, customers now get an email asking which coffee they liked best, with that coffee pre-selected in a one-click order.
2. A smaller box. Four coffees became two, chosen from a short quiz. Fewer choices made it easier to pick a favourite.
3. A subscription offer at the right moment. The offer now appears on the second order, not the first. People who have already chosen a favourite are far more likely to say yes.

The results

Six months later, the March-style drop is gone. Customers who start with a discovery box now return at the same rate as everyone else. Across all customers, repeat orders are up 31% on the year before, and subscriptions have doubled.

Amira's advice

"Look at your customers in groups, not in totals. Totals hide everything. And change one thing at a time, or you will never know which change worked."

She pauses. "Also, taste your own discovery box. We never had."

How they set it up

Juniper's shop sends three events to Tally: a visit to a product page, an order, and a subscription started. Each order carries two properties, the first product in the basket and whether it was the customer's first order. That was all they needed. Setting it up took Amira's developer, a friend who works on the shop two days a month, about an afternoon.

"I thought analytics would mean dozens of events and a data person," Amira says. "In the end it was three events and one question."

What they watch now

The team has one dashboard with four charts. The Monday one is second-order rate by first product. The others are monthly cohorts, subscriptions by month, and the time between first and second order. When a new product launches, they add it as a filter and watch its cohort for eight weeks before deciding whether to keep it.

The discovery box is still on sale. It is now one of their best products, for the business as well as for customers.