AI Integration

Shopware AI Reorder Prediction: Proactive B2B Repeat Sales

Shopware 6 plugin that predicts B2B reorder timing from order history and emails customers before they run out, with a pluggable AI layer and built-in licensing.

Industry:: E-commerce / SaaS Tooling
Project Type:: Plugin Product
Duration:: Ongoing
Location:: Germany
Published:: August 6, 2026

Outcome

Pluggable AI providers

Click to reorder

Scheduled tasks

Manual outreach

The Challenge

B2B customers who buy the same products on a cycle don't always reorder on time, not because they've churned, but because nobody reminded them. Shopware has no concept of "this customer is due to reorder" anywhere in core.

Critical Pain Points

Reorder timing varies per customer and per product; a blanket "reorder in 30 days" email is wrong for most accounts
Manually tracking who was due for what meant either spamming everyone or emailing no one
Any prediction system risked recommending the wrong thing if it ran on too little order history
The plugin needed to work as a sellable product, not a one-off customization, which meant licensing and usage limits from the start
Whatever powered the predictions had to be swappable: not every shop wants to send order data to the same AI vendor

Without a system watching purchase cadence per customer, repeat revenue depended entirely on the customer remembering to come back.

The Solution

I built a prediction engine that reads a customer's real order history and reaches out before they'd normally run out, with the AI layer swappable and licensing built in.

Pattern extraction from real order history

A dedicated analyzer works out each customer's actual reorder interval and quantity per product from their own past orders, not a generic assumption.

Pluggable AI, provider-agnostic

A factory pattern swaps between Anthropic, OpenAI, or a local model behind one interface, with predictions cached so repeat runs don't re-hit the API.

One click from email to cart

Reorder emails link straight to a prediction-filled cart. A dedicated account page also shows predicted reorders any time, not just when the email lands.

Licensing and usage metering from day one

A feature-gate service and usage tracker sit underneath the prediction logic, built so the plugin ships as a licensed, tiered product rather than a single-shop customization.

Results & Business Impact

Reorder emails triggered by real behavior

Predictions come from each customer's own order history, not a fixed reminder schedule that's wrong for most of them.

No vendor lock-in on the AI layer

Anthropic, OpenAI, or a local model all sit behind the same interface, swappable without touching the prediction logic.

Cart pre-filled, not just suggested

The one-click reorder path removes the re-shopping step entirely for repeat purchases.

Built to sell, not just to run

Licensing and usage tracking were part of the architecture from the first commit, not retrofitted later.

Technologies Used

Shopware 6.7
PHP 8.2
Symfony Scheduled Tasks
DAL
Anthropic API
OpenAI API
Licensing/Usage Metering

Leaving repeat revenue to memory?

If your B2B customers reorder on a cycle and nobody's reminding them, that's revenue sitting on the table. Let's talk through what a prediction-driven reorder flow could look like for your shop.