Sales & CRM

How to qualify inbound leads with AI

Sales & CRM3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

Qualifying inbound leads means deciding, quickly and consistently, which people who've expressed interest are worth your time. It's the front door of your sales process — and most operators handle it manually: reading form submissions, skimming LinkedIn profiles, cross-checking company size, and making a gut call on whether to book a call. At volume, this work is exhausting. At low volume, it still eats focus you'd rather put toward closing.

The workflow feels like an AI problem because the inputs are text and the output is a judgment call. You're reading signals — job title, company stage, stated use case, form language — and mapping them against criteria you already know. That's pattern-matching on structured information, which is exactly what large language models are good at. You can describe your ICP in plain English, feed in a lead's details, and get a reasoned score back in seconds.

ChatGPT, Claude, and Gemini can genuinely help here today. You can paste in a lead's form response, LinkedIn summary, and company description, give the model your qualification criteria, and get a scored output with a recommendation. For one lead at a time, this works well. The models reason clearly, ask clarifying questions if you prompt them to, and produce outputs that are often better than a quick human skim.

Skip the manual work → See the Starch version

Sales & CRM3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ChatGPTClaudeGemini

Step-by-step

1 Write a qualification rubric in plain English — define your ICP, the signals that matter (company size, role, use case, urgency), and what a good/bad/borderline lead looks like. Save this as a reusable system prompt.

2 When a new inbound lead comes in, copy their form submission, LinkedIn URL summary, and any enrichment data you have (company headcount, funding stage, industry) into a single text block.

3 Paste your rubric system prompt first, then the lead data, into Claude or ChatGPT. Ask it to score the lead on each criterion, give an overall tier (hot/warm/cold), and write one sentence justifying the call.

4 Review the model's output and decide whether to act on it. For borderline leads, prompt the model to list the specific questions it would want answered before upgrading the tier — use those as your discovery call agenda.

5 If you're triaging a batch, paste multiple leads at once and ask the model to output a ranked table with columns for score, tier, and reasoning. Copy this into a spreadsheet or Notion doc for your records.

6 For leads where you want deeper company context, ask Perplexity to research the company and summarize recent news, funding, and headcount, then feed that summary back into ChatGPT or Claude alongside the lead record.

7 Manually move qualified leads into your CRM or sales tool. Update your rubric document when you notice the model consistently misfiring on a category.

Prompts you can copy

Here is my ICP: B2B SaaS companies, 10-200 employees, ops or finance buyer, annual contract value potential above $10k. Score this lead on a 1-5 scale for each criterion and give me an overall tier: [paste lead data]

This lead submitted our contact form. Based on their message, LinkedIn title, and company description below, tell me: are they a decision-maker, do they match our ICP, and what's the one thing I'd want to clarify before booking a call? [paste data]

I have 8 inbound leads from this week. For each one, give me a hot/warm/cold rating and a one-sentence reason. Format your response as a table with columns: Name, Company, Tier, Reason. [paste all 8 records]

Here is a lead who seems borderline — they match on company size but I'm unsure about their role and urgency. What three questions would you ask in a discovery call to determine if they're worth pursuing? [paste lead context]

Summarize this LinkedIn profile and company page in 3 bullet points focused on: buying authority, company stage, and fit with a product that [describe your product]. Then give a fit score from 1-10. [paste profile text]

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No live connection to your form submissions, CRM, or inbox — every lead requires a manual copy-paste before you can run the qualification prompt.

Qualification criteria drift silently. The rubric you carefully wrote last month lives in a doc or prompt file you have to remember to load, and it's easy for the version in the model to diverge from your actual ICP as you update your thinking.

Batch processing is brittle. Paste ten leads and the model's output format shifts unpredictably, making it hard to pipe results anywhere useful without reformatting by hand.

Nothing writes back to your CRM automatically. A qualified lead still requires you to manually update a record, create a task, or send a follow-up email — the AI's judgment stays trapped in the chat window.

No memory across sessions. The context of why you marked a similar lead cold last week, or what follow-up you sent, isn't available to the model unless you paste it in every time.

At any real volume — more than a dozen leads a week — the copy-paste loop becomes the bottleneck. The AI is fast; the manual handoffs around it are not.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system. For qualifying inbound leads, that means an agent builds a persistent app connected to your actual lead sources, CRM, and inbox — so qualification runs continuously against live data, not on demand when you remember to open a chat window.

The CRM starter app gives you a working lead pipeline out of the box. Describe your qualification stages and ICP in plain English — 'hot, warm, cold based on company size, role, and stated use case' — and the agent builds the schema around how you actually think, not a generic template.

Starch syncs your Gmail or Outlook on a schedule, so inbound lead emails appear in your CRM automatically with thread history attached. Ask 'who submitted a form in the last 7 days that I haven't responded to?' and get a real answer from live data.

Connect your lead forms, Apollo.io, or HubSpot from Starch's integration catalog. The agent queries them live when your qualification app runs — no manual export, no CSV upload, no copy-paste loop.

LinkedIn enrichment runs through browser automation — no LinkedIn API needed. Starch pulls current job titles, company headcount, and recent activity to fill gaps in your lead records automatically.

Describe the automation you want in plain English: 'When a new lead comes in, score them against my ICP, tag them hot/warm/cold in the CRM, and draft a personalized first-touch email for my review.' The agent builds that automation and runs it on every new submission.

The Email Triage app handles the inbox side — surfacing inbound leads by priority, summarizing long threads, and drafting replies you can send in one click, so qualified leads get a fast response without you triaging manually.

Get closed-beta access →

Toolkit