Order predictions and churn risk

Breww can automatically analyse your customers' ordering patterns and predict when their next order is likely to arrive. This helps your sales team focus their time on the customers who need attention, plan production around expected demand, and spot customers who may be at risk of churning.

Order predictions are built from each customer's confirmed order history. Once a customer has at least 4 confirmed orders, Breww fits a statistical model to the gaps between their orders, weighted towards their most recent behaviour. This means that if a customer has recently started ordering more (or less) frequently, the predictions will adapt accordingly.

Predictions update automatically whenever an order's status changes (e.g. when an order is confirmed), and a nightly process keeps overdue probabilities and churn risk scores up to date across all customers.

Customer detail page

When viewing any customer with enough order history, you will see additional stat cards on the Overview tab:

Next expected order -- The predicted date of the next order based on the customer's recent ordering frequency, adjusted for seasonal patterns. The card is colour-coded: red if overdue, amber if due within 7 days, and blue-green if further out.
Recent order frequency -- The typical number of days between orders, weighted towards recent ordering behaviour. More recent orders have a greater influence than older ones, so this adapts as a customer's ordering cadence changes.
Average order interval -- The simple average number of days between all orders over the lifetime of this customer. All orders are weighted equally. This is useful for comparison against the recency-weighted figure.
Churn risk -- The probability that this customer has stopped ordering entirely, calculated nightly using a statistical model. Colour-coded: green for low risk, amber for medium, and red for high.

Below the stat cards, Breww will show a contextual guidance callout when relevant:
Significantly overdue (red) -- Shown when there is a high probability (over 70%) that the customer should have ordered by now. Includes how many days since the last order and their typical ordering frequency.
Slightly overdue (amber) -- Shown when the customer is moderately overdue (40-70% probability). Suggests it may be worth a check-in.
High churn risk (grey) -- Shown when the customer has a high probability of having churned, even if they are not heavily overdue. Their ordering frequency has declined significantly.
Not enough order history -- Shown for customers who do not yet have 4 confirmed orders. Predictions will begin once enough orders have been placed.

Order prediction reports

Four reports are available under the Order prediction section of the reporting dashboard (Reporting -> Pre-built sales reports). Two of these reports are also accessible directly from the customers dashboard sidebar menu for quick access.

Reactivation priority list

This report shows customers who are overdue for an order, ranked by how overdue they are relative to their personal ordering cadence. It is designed to help your sales team prioritise outreach to customers who are most likely to place an order if contacted.

Each row shows the customer name, account manager, number of orders, last order date, expected order date, days overdue, overdue probability (%), and churn risk (%).

The report includes several filters:

Account manager -- Filter by a specific sales person.
Customer type -- Filter by customer type/segment.
Churn risk -- By default, customers with high churn risk (over 60%) are excluded, since they may already be lost. You can change this to show all customers, or focus on a specific churn band (low, medium, or high).
Minimum/maximum days overdue -- Narrow the list to customers within a specific overdue window.
Minimum/maximum orders -- Focus on customers with a certain order history depth.

Expected orders calendar

A forward-looking view of customers expected to place an order in the next 8 weeks. This is useful for production planning and demand forecasting.

Each row shows the customer name, expected order date (colour-coded), recent order frequency, and number of orders.

Ordering frequency trends

Shows all customers who have a fitted prediction profile, along with their ordering cadence, consistency, and next expected order date.

Each row includes:

Recent order frequency -- Typical days between orders, weighted towards recent behaviour.
Order gap range -- A visual bar showing the range within which 90% of the customer's orders fall. A narrow bar means the customer orders very consistently; a wide bar means their ordering is more variable.
Consistency -- A label derived from the spread of the customer's ordering gaps: Very consistent, Consistent, Variable, or Irregular.
Next expected -- The predicted next order date, colour-coded by urgency.

Filters include customer type, minimum orders, consistency level (checkboxes), and minimum/maximum days overdue.

New customer onboarding

Lists customers who have placed at least one confirmed order but do not yet have enough order history (4 orders) for predictions to be generated.

Each row shows the customer name, orders placed, orders remaining until predictions begin, days since first order, and last order date. This report helps you track new customers and understand how close they are to having a full prediction profile.

For quick access, the customers dashboard sidebar menu includes direct links to:

Reactivation priority list -- Jump straight to the list of overdue customers.
Expected orders -- Jump to the upcoming expected orders calendar.

API and BrewwQL fields

Four new fields are available on the customer record for use in the API and BrewwQL filtering:

next_expected_order -- The predicted date of the customer's next order.
contact_after_date -- The date after which the customer should be flagged for follow-up.
prob_overdue -- The probability (0 to 1) that the customer is overdue for an order.
prob_churned -- The probability (0 to 1) that the customer has churned.

These fields are read-only in the API and are updated automatically by Breww's background processing. They can be used in BrewwQL filters to build dynamic customer lists, for example: prob_overdue > 0.5 to find all customers who are likely overdue, or prob_churned < 0.3 and next_expected_order < "2026-05-01" to find active customers expected to order soon.

How predictions work

Breww uses a statistical approach to model each customer's ordering behaviour:

Order gap analysis -- The gaps (in days) between each confirmed order are collected. The most recent gaps are given more weight than older ones, so the model adapts to changes in ordering frequency.
Log-normal fitting -- A log-normal distribution is fitted to the weighted gaps. This captures both the typical gap length and how variable the customer's ordering is.
Seasonal adjustment -- Monthly seasonal factors are applied to adjust predictions for times of year when ordering tends to be faster or slower.
Churn probability -- A BG/NBD (Buy 'Til You Die) model runs nightly across all customers to estimate the probability that each customer has stopped ordering entirely.

All predictions update automatically:

Per order -- When an order status changes involving a confirmed or completed order, the customer's profile is refitted immediately.
Nightly -- Overdue probabilities and churn risk scores are refreshed for all customers.
Monthly -- Seasonal factors are recalculated on the 1st of each month.