Lead routing on autopilot: ICP scoring without an SDR rotation

Most lean B2B teams hit the same problem at the same headcount: the founder cannot keep handling every inbound lead, but it is too early to hire a four-person SDR team. The half-measure — a single SDR who routes — is fragile. They take a week off, leads pile up. They quit, you lose the routing logic with them.

The fix is to put routing on autopilot, with rules so explicit anyone in the team can read them. AI can run the rules in real time. The rules themselves should be written down by humans, in human language, and updated quarterly.

What does ICP scoring even mean for a small team?

ICP scoring is the practice of attaching a numeric score to each inbound lead based on how well they match your ideal customer profile. The score then drives routing: who gets alerted, what playbook runs, how fast.

For a small team, the score does not need to be sophisticated. It needs to be honest. Three pieces of input cover most of the value:

• Firmographics. Industry, company size, geography. Easy to enrich, easy to score.
• Stated pain. What did the lead say in their own words on the form, the demo or the chat?
• Behavioural signal. What did they do on your site? Pricing page visit, multiple sessions, KB search activity?

Three buckets. Each scored 0–10. Sum is the lead score. That is enough to drive 90% of routing decisions in a small team.

A simple ICP-scoring scheme that actually ships

Here is a scheme that works, in roughly 50 lines of business logic:

Firmographics (max 10):

• In your top-3 target industries: +5
• Company size in your sweet spot range: +3
• Geography you can serve (regulatory + language): +2

Stated pain (max 10):

• Pain text contains a phrase that matches a high-value category: +5
• Pain text contains a phrase that matches a medium-value category: +3
• Pain text is unclear or missing: 0

Behavioural signal (max 10):

• Visited pricing page in the last 7 days: +4
• More than 3 sessions in the last 14 days: +3
• KB search query about a high-intent topic: +3

The point is not the exact numbers. The point is to write them down. A scheme on a wiki page that anyone can change in a pull request beats an undocumented gut-feel rotation.

Where AI fits in

AI is genuinely useful at the parts of routing that humans are bad at: parsing free text fast, classifying signal at volume, never forgetting.

Specifically:

• Pain classification. When the prospect says "we are drowning in cold replies", AI maps that to your "reply triage" pain bucket without you having to define the synonyms.
• Spam triage. Spam, automation testers, students, and competitor research all look like real leads on the form. AI catches most of them with high precision and low recall — i.e., it errs on the side of letting a human decide.
• Tone reading. A short, terse message from a senior buyer reads differently from a polite but vague message from a researcher. AI can capture that as a signal alongside the literal score.

What AI should not do:

• Decide who closes the deal. Routing assigns ownership; closing is human.
• Override your geography or compliance rules. Hard rules stay hard.
• "Re-score" leads forever. Lock the score at the moment of capture and let humans re-score on review. A score that drifts undermines the whole loop.

Routing patterns that work

A handful of patterns cover almost every small-team situation.

Round-robin within tier. Lead score above your threshold goes to the next available founder or rep, round-robin. Lead score below threshold goes to the nurture sequence.

Specialty routing. Within tier-1, route by industry (or product surface) to the human who knows that domain best. Two-person team: one routes by industry, one by region. Done.

Time-based escalation. If a tier-1 lead is not contacted within 60 minutes during business hours, escalate to a second person. Most teams get burned not by missing the lead but by missing the response window.

Quiet hours nurture. Tier-2 leads outside business hours go straight into a nurture sequence with a polite acknowledgment, never to a sleeping rep's notification feed.

Notice that none of these patterns require sophisticated tooling. They require the rules to be written and the rules to be enforced.

Common failure modes

Three failure modes account for most broken routing:

• Score lag. A lead is scored when they fill the form, then re-scored when they visit pricing, then again when they reply. The scores diverge across systems and the rep sees the wrong one. Fix: lock score at capture, log subsequent signals as separate events.
• Tier creep. Everyone on the team adjusts the threshold to claim more leads. Fix: write the threshold into the scoring config and require a quarterly review with revenue impact attached.
• Silent humans. The AI pings, the human ignores, no one notices. Fix: every routed lead is acknowledged within an SLA, and every unacknowledged lead becomes a ticket.

How does this fit with my existing CRM?

The CRM is the source of truth for ownership. The routing layer reads from your scoring config and writes the assignment back. AI runs as a service on top, classifying and enriching but never overruling. If your CRM does not support the rule (most do not, natively), put the logic in a small middleware that reads webhooks in and pushes assignments out.

Frequently asked questions

Should the score show up on the lead record?

Yes — and the breakdown that produced it. A naked number is opaque. "Score 18: industry +5, pain match +5, pricing page visit +4, 4 sessions +4" is actionable. Reps trust the system more when they can see the math.

How often should we update the scoring rules?

Quarterly is right for most teams. More often than that, you are reacting to noise. Less often, you are leaving money on the table. Tie the review to a fixed calendar slot — the same day each quarter — so it actually happens.

What if AI mis-classifies a tier-2 lead as tier-3?

Two safeguards: log the classification confidence, and let humans re-tier any lead with a single click. The audit trail of "AI said X, human said Y" is your training data for the next iteration.

Do I still need an SDR if routing is automated?

You probably do at scale, but not for routing. The SDR's job becomes outbound, account research and meeting prep — the things AI cannot do. Routing should not be a person's job in a team smaller than fifteen.

What does this look like in SoloRiff?

The widget captures the lead with the conversation transcript and behavioural signal attached. Scoring runs in your dashboard's rules engine, and the assignment is written into your CRM via the standard webhook. Setup is one click, and the rules are visible to your whole team.

If you want to put routing on autopilot today, set up SoloRiff free in under a minute.