← Builds
Case study · AI-native ops · self-directed project

AI Lead Triage

An AI tool that reads an inbound lead, works out who they are and how serious, routes them, and drafts the right reply. I built it as a self-directed project, so instead of describing how I use AI, it does a real piece of the job. You can try it below.

AI-native reads · classifies · scores · routes · drafts 3 buyer types live interactive demo

The project

I built this as a self-directed project around a real B2B SaaS scenario: a security and AI observability product. Instead of describing how I use AI in marketing ops, I built a working tool that does an actual piece of the job, the first-pass sorting of inbound leads.

The problem

The product sells to security teams, and the inbound is a mix. A CISO with a live incident who wants to pilot this quarter. A curious engineer who just wants the technical writeup and is not a buyer. A vendor pitching their own product. Those three should not get the same response, and a small marketing team cannot hand-sort every message that comes in. That first-pass sorting is exactly the repetitive judgment AI can take off your plate.

What I built

A tool that reads an inbound message and does the first pass a marketer would do by hand. It classifies the lead by buyer type, scores intent and fit, routes it to the right next action, and drafts a reply in a credible, no-hype voice. The buyer types, security lead, engineer, and AI-adoption lead, come from how this market actually segments, which I mapped out before building anything.

Try it

Load one of the three sample leads, or paste your own message, and run the triage.

The detail that matters

One of the samples is a vendor spam pitch. The tool flags it as not a fit and declines to engage, keeping it out of the queue. Handling the non-fit is the real test. Anything can answer an ideal lead; knowing what to ignore is the harder call, and it is where a triage system earns its place.

An honest note

This public demo uses representative, baked-in outputs so it runs reliably on a shared link. A version that calls the model live in real time also exists. The point of the project is the workflow and the judgment behind it, read, classify, score, route, draft, not the model wiring underneath.

What I'd do next

Connect it to a real inbound source so it triages live form fills and emails, feed in real lead data to measure routing accuracy against what a human decides, and tune the buyer types and reply voice to each company it runs for. The framework stays the same; the segments are what change per business.

Built with AI / LLM workflow lead classification intent scoring HTML, CSS, JS