The AI Lead Research Pipeline We Run Daily — And How to Build Your Own

It’s 7am. Nobody on the team is awake yet. But the pipeline already ran. By the time the first person logs in, there’s a fresh batch of enriched, scored, pre-qualified leads sitting in ClickUp — with a message already in the team’s WhatsApp group telling everyone they’re ready to review. This is the AI lead research pipeline Varnan runs every single day — a fully automated system that takes prospect research from 4 hours per week to under 20 minutes of human review time. According to Salesforce’s 2025 State of Sales Report, top-performing sales teams are 2.8x more likely to use AI for lead qualification than average performers — and the gap is widening as these tools become more accessible. Here’s how it works, and how to build your own version.

“The goal isn’t to automate outreach. The goal is to automate research so humans can focus on the judgment that actually determines whether a message gets a reply.”

— Amal Jandheer, Founder, Varnan

What Does the AI Lead Research Pipeline Actually Do?

At 6am every day, a scheduled automation kicks off a five-step sequence. No one triggers it. No one watches it run. It just happens, the same way every day, and by the time the team starts work the output is sitting in a queue ready for human judgment.

That’s the whole point. The repetitive, mechanical part — pulling data, enriching it, scoring it, drafting an opener — happens while everyone’s still asleep. The part that needs a human happens during work hours, on pre-processed material. We measured this at Varnan: the pipeline runs in approximately 45 minutes each morning and processes between 40–80 companies daily, producing 8–15 qualified, reviewed-ready leads for the team.

The Five-Step Daily Lead Research Cycle

1. Pull — A saved search in Apollo.io, filtered by industry, company size, job title, and tech stack, exports a fresh batch of contacts matching our ICP.
2. Enrich — Each contact’s company gets pulled into Clay, which layers on LinkedIn activity, recent hiring, funding stage, and any public signals worth knowing before reaching out.
3. Score — The enriched data is passed to the Claude API with our ICP criteria. It returns a fit score and a one-line rationale — not just “yes” or “no,” but why.
4. Draft — For leads above the fit threshold, the same prompt generates a personalised opening line, grounded in whatever signal made the lead stand out.
5. Push & Notify — Make takes the qualified, scored, drafted leads and creates a ClickUp task for each — then sends one WhatsApp message to the team: “X new leads ready for review.”

How to Build Your Own Version: Tool-by-Tool Setup

1. Apollo.io — Saved Search

Start narrow. A saved search with 4-5 filters (industry, employee count, seniority, location, technology used) beats a broad search with none. Set it to refresh and export on a schedule — daily or every few days, depending on your volume needs.

2. Clay — Enrichment Table

Connect Clay to your Apollo export. Add enrichment columns for LinkedIn company page data, recent posts, and hiring activity. Clay’s built-in waterfall enrichment means you’re not paying for data you can’t find — it tries multiple sources before giving up.

3. Claude API — Scoring Prompt

Write a structured prompt that takes the enriched row as input and returns a JSON object: fit score, rationale, and a suggested opening line. Keep the ICP criteria explicit and specific — vague criteria produce vague scores. We rewrote this prompt three times before getting the false positive rate below 10%.

4. Make — The Daily Scenario

One scenario, scheduled to run early morning. It polls Clay for new enriched rows, calls the Claude API for scoring, filters by score threshold, and routes the result to ClickUp and WhatsApp. This is the glue — everything else just feeds it.

5. ClickUp — The Inbox Board

A simple “New Leads” list with a custom field for fit score and rationale. Each task is one lead, pre-filled with enrichment data and the suggested opener — ready for someone to review, tweak, and send.

What Changes on Your Team’s Side

Nobody on the team opens Apollo or LinkedIn to search for leads anymore. Their day starts with a ClickUp list that’s already populated, already scored, already drafted.

The job becomes review and judgment: does this rationale actually hold up? Does this opening line sound right for this specific person? Send, tweak, or skip. That’s a 2-3 minute decision per lead, not a 20-30 minute research task.

The other shift is psychological. When research is manual, there’s a natural ceiling on volume — people only research as many leads as they have energy for. When research is automated, the ceiling moves to “how many qualified leads can the team actually follow up with well” — which is usually a much higher number.

Three Mistakes That Break This Setup

• ICP criteria that are too broad. If your Apollo filters or scoring prompt are loose, you’ll get volume without quality — and the team stops trusting the queue, which defeats the purpose.
• No score threshold. Pushing every enriched lead to ClickUp, scored or not, just moves the noise problem downstream. Filter before it reaches a human.
• Treating the opener as final copy. AI-drafted openers are a strong starting point, not a send-as-is. Build the review step in — it’s where most of the remaining value gets added.

None of this requires an engineering team. Apollo, Clay, and ClickUp are no-code. Make is low-code. The only real skill is writing a good scoring prompt — and that’s learnable in an afternoon with a few iterations.

Frequently Asked Questions

What does an AI lead research pipeline actually do?

It automates the repetitive parts of prospecting — pulling fresh contacts from a tool like Apollo.io, enriching each company with signals from Clay, scoring fit with an AI model like Claude, drafting a personalised opening line, and pushing the qualified leads to a task board with a team notification. The output is a ready-to-review queue, not a raw export your team has to dig through.

What tools do you need to build a daily lead research pipeline?

Five tools cover the full cycle: Apollo.io for the saved search and export, Clay for enrichment, the Claude API for scoring and drafting, Make for the scheduled scenario that ties everything together, and ClickUp as the inbox board the team works from. Apollo, Clay, and ClickUp are no-code, and Make is low-code — the only real skill required is writing a good scoring prompt.

How does AI scoring work in a lead qualification pipeline?

Enriched lead data — company info, LinkedIn activity, hiring signals, funding stage — is passed to the Claude API along with explicit ICP criteria. It returns a fit score and a one-line rationale explaining why, not just a yes or no. For leads above the score threshold, the same prompt drafts a personalised opening line grounded in whatever signal made that lead stand out.

What’s the most common mistake when setting up a lead pipeline like this?

The biggest one is having ICP criteria that are too broad. Loose Apollo filters or a vague scoring prompt produce volume without quality, and once the team stops trusting the queue, the whole point of automating it falls apart. The fix is to start narrow — four or five specific filters beat a wide net with none — and always filter by a score threshold before anything reaches a human.

How much manual work does this remove from a sales team’s day?

It removes the research step entirely. Instead of spending 20-30 minutes per lead searching Apollo or LinkedIn, the team opens a ClickUp list that’s already populated, scored, and drafted, and makes a 2-3 minute send-tweak-or-skip decision per lead. Varnan builds and hands over this exact system — Apollo, Clay, Claude, Make, ClickUp, and WhatsApp wired together — so a team’s queue fills itself every morning.