← Back to Research

15 February 2026

LinkedIn Network Mining: Using AI to Turn 7,000 Connections into a New Business Pipeline

I ran an experiment in February 2026: could a general-purpose AI model turn a raw LinkedIn data export into a segmented, prioritised new business pipeline in under an hour? The answer was unambiguously yes.

I ran an experiment in February 2026. I wanted to know whether a general-purpose AI model, Claude or ChatGPT, nothing proprietary, could turn a raw LinkedIn data export into a segmented, prioritised new business pipeline in under an hour. No specialist software. No sales intelligence subscription. No developer involvement. Just a CSV file and a well-constructed prompt.

The answer was unambiguously yes. And the implications for how professional services firms, consultancies, and advisory businesses approach new business development are significant enough that I wanted to document the method, the results, and the strategic thinking behind it.

This is not a product review. There is no tool to sell. This is a workflow I built, tested, and now use, and I think anyone with a meaningful LinkedIn network and access to an AI assistant is leaving value on the table if they are not doing something similar.

The Problem This Solves

Most senior professionals have LinkedIn networks that are simultaneously their most valuable business development asset and their least utilised one.

LinkedIn now has over 1.2 billion members globally, with approximately 310 million monthly active users (Brenton Way, March 2026). The platform generates an estimated 80% of all B2B leads originating from social media (Snov.io, December 2025). For anyone in professional services, consulting, or advisory work, your first-degree connections represent a curated universe of people who already know who you are.

And yet most of us interact with this network in the worst possible way: scrolling the feed, occasionally liking a post, and periodically searching for a name we already know. We almost never interrogate the network as a dataset, asking structured questions like "who in my network is a founder of a B2B SaaS company?" or "which of my connections hold C-suite roles at companies backed by private equity?"

LinkedIn's own search cannot answer these questions. You can filter by current company or keyword in a job title, but you cannot run the kind of natural language, inference-heavy queries that would actually surface the people who matter for a given business development initiative.

AI can.

The Method

The workflow is straightforward enough to describe in five steps. The nuance, and the value, is in how you frame the prompts and what you do with the output.

Step 1: Export Your LinkedIn Connections

LinkedIn allows you to download your entire first-degree network as a CSV file. You request it through Settings & Privacy > Data Privacy > Get a copy of your data > Connections. LinkedIn emails you a ZIP file, usually within ten minutes. Inside is a file called Connections.csv.

The CSV contains six columns per contact: First Name, Last Name, LinkedIn URL, Email Address (only where the contact has opted in), Company, Position, and Connected On date.

It is sparse data. No industry field. No company size. No seniority tag. Just a name, a title, a company, a URL, and a date. For a human scanning this manually across thousands of rows, it is nearly useless for strategic segmentation. For an AI model, it is more than enough.

Step 2: Upload to Claude or ChatGPT

Both Claude and ChatGPT support CSV file uploads in conversation. You start a new chat, attach the file, and ask your question. For large files, my export was over 7,000 connections, Claude handles the full dataset reliably in a single pass. ChatGPT sometimes needs prompting to process all rows rather than sampling.

Step 3: Ask Structured, Specific Questions

This is where the method becomes genuinely powerful. The AI model can interpret job titles, infer seniority, recognise company types from names alone, handle title variations (CEO, Founder, Managing Director, General Partner all mean different things in different contexts), and categorise results by confidence level.

Here are examples of the kinds of queries I ran:

Finding SaaS founders:

"From this CSV of my LinkedIn connections, identify everyone who appears to be a CEO, founder, or co-founder of a B2B SaaS company. Categorise them as High Confidence or Medium Confidence. Include name, company, title, and LinkedIn URL. Output as a table."

Finding senior agency executives:

"Find all contacts who hold senior positions, VP and above, Managing Director, C-suite, Partner, Group Head, at major advertising agency holding companies including WPP, Publicis, Omnicom, Dentsu, Havas, IPG, Stagwell, and their subsidiary agencies."

Finding PE and VC contacts:

"Find all connections who work in private equity, growth equity, or venture capital. Include Partners, Principals, Directors, and Operating Partners. Note the fund name and their role."

Cross-referencing against a target list:

"Here is a list of 50 target companies. Which of my connections work at these companies, and what are their roles?"

Network composition analysis:

"Analyse my full LinkedIn network and give me a breakdown by category: agency, brand-side, SaaS/tech, PE/VC, media, consulting, and other. For each category, show the count and list the ten most senior contacts."

The AI returns structured tables within seconds. Not approximate guesses, properly categorised, confidence-scored, URL-linked results that would have taken hours to compile manually.

Step 4: Refine and Export

The first pass rarely captures everything. Follow-up prompts are essential:

• "Check for anyone you may have missed whose title contains 'Partner' or 'Principal' at companies that could be investment firms."
• "Remove anyone whose title contains Consultant, Freelance, or Advisor."
• "Show me only contacts I connected with in the last 12 months, these are warmer relationships."
• "Export this as a spreadsheet with columns for Name, Company, Title, LinkedIn URL, and Confidence Level."

You can also ask the AI to suggest personalised outreach angles for each contact based on their role and company. This is useful for drafting the first line of a message, though I would always rewrite it manually for senior contacts.

Step 5: Review, Clean, Outreach

The AI output is a draft pipeline, not a finished one. I spend ten to fifteen minutes manually reviewing, removing obvious false positives, and checking a sample of profiles to validate the categorisation. In my experience, accuracy runs at roughly 90–95%, the occasional miscategorisation comes from ambiguous company names or outdated titles, which is inherent in self-reported LinkedIn data.

Then the outreach begins. Highly personalised. One-to-one. No automation for senior contacts. The entire point of this method is that it produces a small, high-quality list of people you already have a relationship with, not a mass blast list.

Why This Works Better Than You Would Expect

Three things make this approach surprisingly effective.

First, the inference capability of current AI models is precisely suited to this task. Interpreting job titles, recognising company types from names alone, assessing seniority across different naming conventions, and handling the ambiguity inherent in professional identity, these are natural language interpretation problems, and they are exactly the kind of thing large language models excel at. A rule-based filter cannot tell you that "Head of Growth" at a 20-person company is functionally equivalent to a CMO. An AI model can.

Second, your first-degree LinkedIn network is an absurdly underleveraged asset. The average active LinkedIn user has approximately 1,300 connections (Thunderbit, June 2025). Senior professionals in services industries often have 5,000 to 30,000. These are not cold prospects. These are people who have accepted a connection request from you, or sent you one. They have, at minimum, a passive awareness of who you are. For new business purposes, that is a meaningful head start on any cold outreach programme.

Third, the method converts a static network into an active pipeline in under an hour. The traditional alternative, Sales Navigator searches, manual list-building, cross-referencing in CRM systems, is not only slower but produces worse results for this specific use case. Sales Navigator is designed for searching LinkedIn's full database, not for mining the relationship value that already exists within your own network.

What It Cannot Do

It is worth being precise about the limitations so that expectations are correctly calibrated.

This method only covers your first-degree connections. It does not surface second or third-degree contacts. For that, you still need Sales Navigator or a referral strategy.

Job titles and company names in the export are self-reported and may be outdated. Someone who changed roles last month may still show their old title. The AI is working from a snapshot, not a live feed.

Email addresses are sparse. LinkedIn only includes them in the export where the contact has opted in to sharing, which in practice means roughly 30% of connections (Evaboot, 2023). For most senior outreach, you will be going via LinkedIn direct message or InMail, not email.

The AI is inferring industry and seniority from limited data, a title and a company name. It gets most things right, but expect 5–10% noise in the results. This is why the manual review step matters.

And critically, this method works because LinkedIn allows you to export your own connection data. It is using a legitimate platform feature, not scraping or violating terms of service. The ethical dimension matters: you are analysing data about relationships you already have, not harvesting data about strangers.

The Strategic Implications

What struck me most about running this experiment was not the efficiency gain, though going from "I should probably look through my LinkedIn network" to "here is a prioritised, segmented pipeline of 200 contacts" in forty-five minutes is genuinely transformative for a time-poor professional.

What struck me was the strategic implication.

Your network is a dataset. Treat it like one.

Most professionals think of their LinkedIn network as a social graph, a collection of relationships they navigate by memory, serendipity, and the feed algorithm. Exporting it as a CSV and querying it with AI transforms it into a structured business intelligence asset. You can run the same query against it every quarter as your network grows. You can cross-reference it against new target account lists as your strategy evolves. You can segment it by connection date to identify warmer, more recent relationships worth prioritising.

The compound value of a curated network increases with AI.

Every connection you have ever accepted is now queryable, segmentable, and actionable in ways that were not practically possible twelve months ago. This changes the calculus of how you build your network going forward. Every connection request you accept is a row in a future query. The deliberateness with which you grow your network now has a direct, measurable impact on the quality of the pipeline you can generate from it later.

This is a preview of how AI reshapes professional services business development.

The LinkedIn export method is manual, lightweight, and deliberately low-tech. But it demonstrates a principle that will define business development over the next several years: AI does not replace relationships. It makes the intelligence layer around relationships dramatically more accessible. The firms and individuals who recognise this early will build pipelines that their competitors cannot see, from assets they already own.

The Recommended Workflow

For anyone who wants to replicate this, here is the complete sequence:

1. Export your connections from LinkedIn (Settings & Privacy > Data Privacy > Get a copy of your data > Connections).
2. Upload the CSV to Claude or ChatGPT.
3. Run a query tailored to your target segment, using the prompt structures above as starting points.
4. Ask the AI to export results as a spreadsheet.
5. Review and clean the list, spend ten to fifteen minutes removing obvious false positives.
6. Cross-reference against any target account lists or campaign briefs you are working with.
7. Check profiles for each high-priority contact, note recent activity, shared connections, or conversation starters.
8. Begin outreach, highly personalised, one-to-one, no automation for senior contacts.

The entire process from export to prioritised outreach list takes thirty to sixty minutes, depending on the size of your network and the specificity of your query.

Closing Thought

I have spent twenty years in performance marketing, and the single most reliable driver of advisory and consulting revenue has always been the same thing: knowing the right people and staying in useful contact with them. What has changed is that AI now gives you a way to interrogate the network you have already built with a precision and speed that was previously impossible.

Your LinkedIn connections are not a vanity metric. They are a queryable database of professional relationships. The only question is whether you are asking the right questions.