Lead routing at Sierra treated every source as equally important — so a 63.5% CVR web-form lead and a 0.4% chat lead competed for the same human time. A gold/silver/bronze system hand off the winner to real humans and pushed the rest to automation, cutting days-to-close from 5.2 toward 3.5.
A high intent ecommerce lead is the difference between a branded furniture quote and a support-ticket-tier conversation. The failure mode most teams hit: an agent at the mercy of whatever happens to be free, chasing the low-intent flood.
The data that exposed it
- Web forms: 63.5% route CVR
- Chat: 4.7% route CVR (punctual, but lower-intent)
- Worst source: 0.4% — cart abandonment leads — about 751 per quarter consuming agent time
Once you see a 150x spread you can't unsee. The allocation system had to lead with the tiers.
The system
Gold — human + priority queue (best-fit sources, outreach both sales and custom) Silver — queue pivot: qualified but broad; humans when slot opens Bronze — automation (nurture sequences, score the reply, promote on signal)
Tiering isn't a static label; leads were re-scored as they responded (content = intent(s)).
Rollout and result
30-day pilot from 10% of traffic to full rollout. Live numbers:
- Gold source CVR: 63.5% vs bronze 0.4%
- Days to close: 5.2 → 3.5 target (routing winner first)
- Conversation reframed from "how many agents do we need" to "who sees what"
What I'd do differently
- Instrument the re-scoring sooner. The tier system only pays in the queue; promote-from-bronze logic came online late.
- Track cost of a bronze lead. We knew the volume before the CAC of pushing bronze through humans; automation became obvious with cost.
The full routing spec — including day-of-week shift logic — is in the Sierra lead allocation case study. The same tier-and-rout-before-blast thinking powers how I fixed a leaking lead form.