I tested a well-known AI chatbot with the same exact lead scenario twice, same message, same context, back to back. It called the lead "warm" the first time. It called the exact same lead "hot" the second time. Nothing about the situation had changed. Only the model's guess had.
That's when I knew I couldn't build Speak2MyCV's scoring on an LLM alone. Years as a cloud and solutions architect taught me one thing above all else: if a system produces a different answer to the identical input twice, it isn't a system you can trust with something that actually matters, and a lead you might follow up on, or ignore, absolutely matters.
So I split the two jobs apart. The AI conversation, answering a visitor's question, still runs on a language model, grounded strictly in the professional's own CV so it can't invent things. But deciding whether that visitor is a genuine Hot, Warm, or Cold lead runs on a separate, deterministic, rule-based engine that never touches model inference at all. Same input, same output, every single time. Auditable. Reproducible. Not a guess dressed up as intelligence.
Heading into 2026, as more products slap "AI-powered lead scoring" onto their marketing pages, this distinction matters more, not less. Ask what's actually happening underneath. If the answer changes when nothing else does, it isn't scoring, it's a coin flip with better branding. Speak2MyCV was built specifically so you never have to wonder which one you're getting.