Every product I've built in the last year has the same backbone: n8n running on a VPS, a few Supabase tables, and a set of webhook-triggered workflows that quietly handle the operational work.
Before Xeref.ai went into beta, we had to prove this foundation held.
Lead Capture Without a CRM
The first automation we wired up was inbound routing. Every lead source — YouTube, Skool, direct form submissions — hits a single webhook. From there, intent is classified and the contact is either queued for a personal reply or entered into a sequence. No CRM required in the early stages.
Email That Doesn't Feel Automated
The hardest part of email automation isn't the logic — it's the tone. We use condition branching based on lead source and behavior, so no two sequences look identical. The content is still manually written, but the routing and timing are hands-off.
Task Routing
As team size grew (even as a solo operator), the mental overhead of routing incoming tasks became real. We built a simple classifier that reads incoming requests and assigns them to the right queue — agent, human review, or async processing.
Why This Matters for Xeref
None of this was throwaway infrastructure. The patterns we validated here — webhook ingestion, intent classification, async queues — became the architecture that v1.0 of Xeref.ai shipped with.
Build the foundation before you build the product.