The lead-routing workflow is where most automation engagements start, because it is the workflow that breaks weekly when it is manual. Inbound lead arrives. Someone needs to enrich it, score it, route it, and assign it. Done manually, the median time-from-submission to in-rep-inbox is 4-12 hours. Done well in 2026, it should be under 60 seconds. This post documents the pattern.
The seven steps
A production lead-routing workflow has seven stages. Each is independently failable and independently retryable.
- Capture: the webhook hit, the form submission, the imported CSV row. The capture step extracts the canonical record and writes a stub to the database with status=pending_enrichment.
- Deduplicate: check against the existing CRM. If a record matches on email, domain, or phone, link to the existing record and skip ahead to scoring. Roughly 25-40% of inbound leads in B2B SaaS are duplicates worth catching.
- Enrich: hit a data provider (Clearbit, Apollo, Hunter, or a custom enrichment agent) for firmographic data. Company size, industry, tech stack, funding stage. The enrich step is the most failure-prone because of vendor latency and outage. Build retries + fallback providers.
- Score: an AI agent or a deterministic rule set assigns a score in [0, 100]. The agent prompt includes the enriched record, the ideal customer profile, and a few canonical labeled examples. Output is the score plus a one-line rationale.
- Route: based on score + tags (industry, geography, source), assign to a sales rep or a sales pod. Round-robin within the assigned pod, with weighting for current pipeline load.
- Notify: post into Slack, send an internal email, write to the CRM activity log. The rep gets a single notification with the score, rationale, and a one-click open-in-CRM link.
- Audit log: write the full decision trail to the agent's observability layer. Operator can reconstruct what happened for any lead in under 30 seconds.
Where it breaks
Three failure modes account for 90% of lead-routing incidents.
Vendor enrichment outage: Clearbit or Apollo goes down. Without a fallback, the agent stalls. The fix: configure two providers in priority order, fail fast on the primary, fall through to the secondary. Cache enrichment results aggressively (most companies do not change firmographics on a 24-hour timeline; cache for 7 days).
Score disagreement: a sales rep argues the score is wrong on a specific lead. Without a feedback loop, the disagreement becomes a Slack thread that updates nothing. The fix: a one-click "score override" in the notification that writes back to the agent's training data. The next prompt eval run incorporates the override as a labeled example.
Pod load imbalance: round-robin assignment overloads whichever rep happened to be up at the moment of the spike. The fix: weight assignment by current pipeline depth, not pure round-robin. The rep with 30 open opportunities gets fewer new leads than the rep with 5.
What it costs to run
For a B2B SaaS doing 200-800 inbound leads per month, the cost breakdown looks like this.
Tools: $50-$200/month for enrichment provider API. $20/month for n8n self-hosted. $30-$80/month for token spend on the scoring agent. Total tools: $100-$300/month.
Operator audit time: 30-90 minutes per week on edge cases. At a fully-loaded operator rate of $50/hour, $100-$300/month.
DPL retainer covering the entire workflow: $2,500/month. Includes operator hours, agent operation, monthly written report, replacements and updates.
Compare to the in-house alternative. A sales operations associate to run the same workflow manually is $50K-$75K base ($70K-$105K fully loaded). The math heavily favors the agent stack at any volume above 100 inbound leads per month.
The week-one shipping checklist
Every DPL lead-routing pilot ships against the same checklist on day 7 of the 30-day pilot.
- Webhook capturing all inbound sources (form, CSV, Apollo sequences, demo requests) into a single canonical record.
- Dedupe against existing CRM with 95%+ precision on email match.
- Two-tier enrichment with provider failover documented.
- Scoring agent prompt frozen, eval suite of 20 inputs running on weekly cron.
- Routing rules in code, not in someone's head.
- Slack notification with score + rationale + open-in-CRM link working end-to-end.
- Audit log searchable by lead-id, rep, time range, score band.
- Operator escalation queue wired with documented SLA (30 min business hours, 4 hour off-hours).
- Cost dashboard showing per-lead cost, weekly trend, vs in-house alternative.
If your current lead-routing breaks on Monday morning traffic spikes or surprises you with rep complaints, a recovery diagnosis audits where the gap is and the pilot ships the fix.