Build an AI lead scoring workflow with Make, Claude, and your CRM in a day

AI lead scoring workflow tutorial: you can build a practical lead scoring and routing system in Make with Claude and HubSpot or Pipedrive in under a day by combining CRM rules, AI fit/intent analysis, and clear thresholds for sales alerts and owner assignment.³⁴⁹

How does this AI lead scoring workflow tutorial fit into modern sales ops?

Modern AI lead scoring workflows pair your CRM’s native scoring and automation with external AI for richer fit and intent analysis, then route leads using clear thresholds and tasks.³⁴ This keeps sales teams focused on high-probability leads without abandoning proven rules-based models.

Since 2025, most practical tutorials follow the same pattern: trigger on an inbound event, enrich the lead, call an AI model for a 1–100 score plus reasoning, and update CRM properties and routing.⁹ HubSpot’s scoring editor and workflows make this especially straightforward, but the same structure works with Pipedrive or any other CRM.¹¹

In this tutorial, we’ll build:

• A composite score (Fit + Intent + Total) mapped to CRM properties.¹³
• A Make scenario orchestrating Claude (or Perplexity) calls for AI scoring.¹⁰
• Routing logic that moves leads between tiers (Hot, Warm, Nurture) and alerts sales within minutes.

The goal is pragmatic: measurable pipeline lift this quarter, not a speculative AI science project.

What lead scoring model should you use before adding AI?

You should define a simple, rules-based lead scoring model in your CRM first, with clear fit and intent criteria tested against known good and bad contacts.³⁴

HubSpot’s own guidance is explicit: you start by defining criteria in the “HubSpot Score” property and test against a sample of qualified and unqualified leads before you automate anything.³ Recent 2026 templates push this further, recommending separate properties such as Fit Score, Intent Score, Total Lead Score, Lead Tier, Last Score Update, Score Decay Pause, Routing Owner, and Last High-Intent Action, plus four supporting workflows.¹³

A practical baseline scoring model:

• Fit Score (0–50) – company size, industry, region, tech stack, role.
• Intent Score (0–50) – pages viewed, demo requests, pricing visits, email engagement.
• Total Lead Score (0–100) – Fit + Intent.
• Lead Tier – A (80+), B (50–79), C (0–49).

Use CRM rules to populate Fit and Intent from structured fields first. AI should layer on nuance (subtle fit signals, job titles, website copy), not replace well-defined ICP logic.

Common mistakes to avoid when defining your model

Most teams get three things wrong:

• Relying on AI without an ICP. Tutorials stress defining “what good looks like” and validating scoring against existing contacts; skipping this makes AI outputs inconsistent.³⁴
• Using a single score only. Modern templates emphasise separate Fit and Intent scores and a composite Total to avoid over-prioritising noisy but poor-fit activity.¹³
• Routing all “high” scores identically. Best-practice workflows branch by tier: high-fit to sales, medium to nurture, low to disqualification or long-term tracking.⁷⁸

Get the rules right first. Then AI will amplify a solid scoring system rather than compensate for a vague one.

What data should you send to Claude or Perplexity for lead scoring?

You should pass rich, structured and unstructured lead and account data to your AI model, including firmographics, behaviour, and free-text answers, then ask for structured JSON back.⁹⁷

Recent recipes for HubSpot + AI recommend returning a lead score (1–100), qualification reason, key talking points, and recommended next action as JSON so you can map fields cleanly into CRM properties and workflows.⁹ That means the prompt matters: it has to describe your ICP, your scoring scale, and your routing thresholds.

For each new lead, aim to send:

• Contact data: name, role, seniority, location, custom fields.
• Company data: industry, size, domain, any enrichment you have.
• Behavioural data: source, last page viewed, number of visits, assets downloaded.
• Form/Chat text: “What problem are you trying to solve?”, free-text notes.

Claude (or Perplexity via its API) can then analyse fit and intent against your ICP description, returning a numeric score, tier, and notes that sales can actually use.¹⁰¹²

A practical scoring prompt you can reuse today

Your Make module for Claude might send a prompt along these lines:

{
  "instruction": "You are a B2B SaaS lead scoring assistant.",
  "input": {
    "lead": { /* contact fields */ },
    "company": { /* firmographics */ },
    "activity": { /* behaviour */ }
  },
  "output_format": {
    "fit_score": "0-50",
    "intent_score": "0-50",
    "total_score": "0-100",
    "tier": "A/B/C",
    "summary": "string",
    "recommendation": "string"
  }
}

This mirrors the 1–100 scoring recommendations seen in 2026 lead qualification recipes and makes mapping into HubSpot or Pipedrive straightforward.⁹⁷

How do you wire Make, Claude, and your CRM together in under a day?

You wire Make, Claude, and your CRM together by building a single scenario: trigger on a new lead, enrich and validate data, call Claude for scoring, update CRM properties, and route high-scoring leads via tasks and alerts.¹⁰⁷

A 2026 guide on Make + Claude for client lead generation breaks this down into five simple steps: trigger on new lead, validate data, enrich company info, have Claude analyse business and lead fit, and return an AI score used to update CRM and route high-scoring leads.¹⁰ That’s exactly the pattern we’ll reuse here.

Step 1: Choose your trigger

In Make, create a new scenario and select your trigger:

• HubSpot: “New contact” or “Form submission” webhook.¹¹
• Pipedrive: “New deal” or “New lead” event.⁶

Best-practice tutorials emphasise triggers like form submissions, chat completions, or key property changes, so AI sees the richest possible data at the moment of interest.⁷⁸

Step 2: Validate and enrich the lead

Add Make modules to:

• Check required fields (email, company, role). If missing, create a task to complete later.
• Enrich firmographics via your existing data sources or simple domain lookups.

Clean data is critical. AI scoring looks impressive, but if the underlying data is thin or wrong, your routing will be noisy.

Step 3: Call Claude (or Perplexity) for scoring

Add an HTTP or native AI module in Make to call Claude with the JSON schema above.¹⁰ Include your ICP description and scoring logic in the prompt to keep outputs consistent.

You can swap in Perplexity’s enterprise API here for analysis and suggestions while keeping Make as the orchestrator.¹² The important part is structured output: numeric scores, tiers, narrative summary, and recommended next action.

Step 4: Map AI outputs into CRM properties

Use Make’s CRM modules to update properties:

• Fit Score → numeric property.
• Intent Score → numeric property.
• Total Lead Score → numeric property.
• Lead Tier → A/B/C string.
• AI Summary / Recommendation → text fields.

HubSpot’s scoring UX and workflows are designed for exactly this kind of structure, with properties for fit, intent, total score, and tier assignment.¹¹¹³ Pipedrive’s Pulse AI feature, launched in July 2025, already highlights promising leads and urgent activities, and your Make + Claude flow can complement this with deeper scoring and notes.⁶¹⁵

Step 5: Route and alert based on thresholds

Finally, create routing logic in Make or your CRM workflows:

• Total Score ≥ 80 (Tier A): rotate owner to sales, create “Follow up within 24 hours” task, send Slack/Teams alert, set lifecycle stage to Sales Qualified Lead.⁴⁵
• Total Score 50–79 (Tier B): assign to SDR, add to nurture sequence, create “Review within 3 days” task.
• Total Score < 50 (Tier C): mark as nurture-only or disqualified, add to long-term campaigns.

A 2026 HubSpot + Opper AI guide calls out 75+ as a common starting threshold and uses Slack for instant notifications when leads cross it, while inbound lead recipes use a 1–10 scale where 7+ triggers a #hot-leads channel message.⁷¹⁴ Your thresholds should reflect your volume and team capacity, but the pattern is the same: clear tiers, explicit actions.

How does this AI workflow compare to native CRM lead scoring tools?

Native CRM scoring tools (HubSpot, Pipedrive Pulse AI) handle rules and basic AI scoring well, but a Make + Claude workflow gives you more control over prompts, thresholds, and multi-system routing.¹¹⁶

Here’s a practical comparison.

Capability	HubSpot native scoring	Pipedrive Pulse AI	Make + Claude workflow
Scoring model	Rules-based with optional predictive scoring.11	AI highlights promising leads and urgent activities.615	Custom Fit/Intent/Total logic in prompts and mapping.
Scale	0–100 score with composite properties and tiers.13	Internal AI qualification, no external prompt control.6	Any scale (0–100, 1–10) with JSON outputs.9
Routing	Workflows for owner rotation, tasks, lifecycle stages.45	Basic assignment and activity surfacing.6	Cross-tool routing via Make (Slack, email, data warehouses).
Setup time	30–60 minutes for initial rules-based scoring.3	Enabled in Pipedrive settings; AI launched July 2025.15	Under a day for end-to-end scoring and routing.10
Flexibility	Strong inside HubSpot, limited beyond.11	Tightly coupled to Pipedrive UI.6	Fully custom, orchestrating multiple CRMs and channels.

For many teams, the right answer is hybrid: use HubSpot’s scoring and workflows as the backbone, run Make + Claude alongside them for nuanced fit/intent scoring, and plug into Pipedrive Pulse AI where your reps live.

How do you measure pipeline lift from this workflow this quarter?

You measure pipeline lift by tracking changes in lead response times, conversion rates between lifecycle stages, and rep focus on high-scoring tiers before and after deploying the workflow.²⁵

HubSpot Academy and step-by-step workflow guides emphasise measuring conversion across defined stages (e.g., MQL → SQL → Opp) and tying routing rules to score thresholds so you can see whether higher scores correlate with faster progression and wins.¹⁵ To keep this grounded:

• Baseline: Record current MQL → SQL conversion rates and average time to first touch.
• After launch (4–8 weeks): Compare those metrics for A-tier leads vs others.
• Qualitative: Ask reps whether AI summaries and recommendations help them prioritise and personalise outreach.

Because you’re using explicit properties like Fit Score, Intent Score, and Lead Tier, you can slice pipeline by AI tier and see whether the workflow is genuinely producing more qualified opportunities, not just more noise.¹³

You can iterate quickly: adjust thresholds (e.g., raise the hot-lead cutoff from 75 to 80), refine your Claude prompt with better ICP examples, or expand the data you send (adding product-interest tags or chat transcripts). The infrastructure you’ve built in a day – Make scenario, CRM properties, and routing rules – is flexible enough to support these revisions without starting over.