Written by the WM Tech product and engineering team — practical guides, case studies, and research on automating ecommerce support at scale.
After analyzing 142,000+ flagged orders across our platform, we found that the vast majority share a small, consistent set of behavioural signals detectable at checkout. This post breaks down exactly what those signals are and how to configure your fraud threshold to catch them without blocking legitimate customers.
Vague policies like "items must be in original condition" are impossible for AI — or humans — to apply consistently. Here's how to structure rules your automation can act on without ambiguity.
Live chat and SMS have overtaken email as the primary support channels for US D2C brands. Here's a complete playbook for automating them without losing the personal feel your customers expect.
Off-the-shelf NLP models misclassify ecommerce support intent roughly 18% of the time on our benchmarks. Here's why we built a domain-specific fine-tuned classifier and what it took to get to 98.9% accuracy.
A Texas-based electronics retailer was losing a third of its margin to card fraud and chargebacks every month. This is the full story of how they deployed WM Tech and what the data looked like week by week.
A look at the infrastructure design choices — event queue architecture, stateless microservices, and model serving optimization — that let us maintain <2s response times at our current scale.
Our public launch release. Enterprise customers can now fine-tune the AI model on their own product catalog and return history. Plus: a new Make module, expanded language support, and improved fraud signal detection.
One email per week. No spam. Unsubscribe anytime.