Loiale logo
Call
BACK TO CASES
OrigenDTC FOOD & BEVERAGE

Built the data layer that made a wine club forecastable.

Origen ran a club on spreadsheets and instinct. We built the customer data layer, then used it to make allocation, reorder timing and club upgrades predictable.

Wine bottles and gourmet tins in a wooden crate on a stone table in a vineyard
MODELClub + one-off DTC
CLUB MEMBERS~6k
SEASONALITYHigh
TEAMFounder-led marketing
THE QUESTION

Can we forecast club demand and reorder behaviour accurately enough to plan allocation, instead of reacting to it every season?

+21%CLUB UPGRADE RATE
47%REORDER WITHIN 60D+13pt
-38%FORECAST ERROR
SERVICES
  • Data audit
  • Stack buildout
  • Lifecycle architecture
STACK
  • Shopify
  • Klaviyo
  • BigQuery
  • Looker Studio

The problem

The club worked commercially but nobody could predict it. Allocation decisions were made on last season's totals, which meant good vintages sold out early and weaker ones tied up capital.

What we built

  • A household-level customer model with varietal affinity and consumption cadence.
  • Reorder prompts timed to when the wine is likely finished, not to a fixed calendar.
  • Upgrade targeting based on observed consumption instead of membership tenure.
  • A demand forecast the founder reviews monthly, built on the same tables as marketing.

The result

Reorder within 60 days rose to 47% and forecast error dropped by 38% across two allocation cycles. The same data layer now serves both marketing and operations.

"The point was never the dashboard. It was being able to commit to an allocation without guessing."– Founder, Origen
HOW IT RAN, 4 MONTHS
  1. PHASE 1Diagnose

    Reconciled club records, Shopify orders and the spreadsheet allocation model. Found the join key that actually worked.

  2. PHASE 2Build the layer

    BigQuery customer model with club status, varietal affinity and consumption cadence per household.

  3. PHASE 3Activate

    Reorder timing driven by cadence rather than calendar. Upgrade offers targeted on affinity and consumption, not tenure.

  4. PHASE 4Forecast

    Allocation forecast built on the same model, reviewed regularly with the founder and operations.

HOW WE MEASURED THIS

Upgrade and reorder rates measured in BigQuery on matched member cohorts across two comparable seasons. Forecast error compares the model's projection against actual demand for two consecutive allocation cycles, which is a small sample and is labelled as preliminary. Client-approved.

Read our evidence standard

Want the same system inside your stack?

See more cases