How to route a lead to an agent (without breaking it)
Teams rush to automate lead routing, then wonder why good leads stall. The problem is never the automation — it's handing an agent a process no one wrote down. Here's how to give routing to an agent so it amplifies your best rep's judgment instead of your worst process.
Write down the three checks a human does in five seconds
Sit with your best SDR and capture the snap judgments they make but can't quite articulate: the fit read, the intent read, and the timing read. They never get these wrong — you just have to get them out of their head and onto paper.
Turn each check into an explicit rule
Fit becomes ICP fields (industry, size, region). Intent becomes signal thresholds (pricing-page visits, a demo request, repeat sessions). Timing becomes recency plus a working-hours window. Each check has to be something an agent can actually evaluate — no vibes.
Define the routing outcomes
Decide exactly where a lead goes for each combination: instant human follow-up, agent-nurtured, or held for more signal. Name every destination. If you have a 'misc' bucket, you haven't finished this step.
Set the escalation and fallback path
Decide what happens when signals conflict or data is missing. The default should be to route to a human, not to guess. A good agent asks for help at the edges instead of quietly making a bad call.
Instrument it before you trust it
Log every routing decision with the reason attached. For the first month, review a sample every week — you're checking the agent's judgment against your best rep's. This is the step teams skip, and it's the one that earns trust.
Hand over the loop, keep the human above it
Once the agent's decisions match your best rep's about 90% of the time, let it run routing autonomously — with a human reviewing the exceptions, not every lead. Humans above the loop, agents inside it.
Automation amplifies whatever logic you feed it. Make sure it's the logic your best human already trusts — then get out of its way.