This is Part 1 of a 3-part series on turning your LinkedIn network into an AI-powered opportunity engine.

You can use this article completely on its own.

And if I asked you right now:

Who are the 20 most strategically useful people in your network?

Could you answer?

I couldn’t.

Because LinkedIn is very good at helping us collect connections.

It’s terrible at helping us understand them.

You connect with:

• former colleagues

• customers

• recruiters

• people you met at events

• partners

• friends of friends

• people from three jobs ago

• people you genuinely cannot remember meeting 😂

And over time they become one enormous list.

Somewhere inside that list may be someone who:

• works at your dream company

• could hire you

• could buy from you

• could introduce you to someone important

• could become a partner

• could invest

• could speak at your event

• could help you make your next move

The problem is not necessarily that your network is too small.

The problem is:

you can’t see it.

So I tried something simple.

I downloaded my LinkedIn connections.

Uploaded the file to AI.

And instead of asking:

“Who do I know?”

I asked:

“Who matters for what I’m trying to do next?”

That is a much better question.

What we’re building

By the end of this tutorial, you’ll have:

Your Network 20

The 20 existing connections most relevant to one specific goal.

You can use it to find:

• your next employer

• potential customers

• investors

• partners

• sponsors

• speakers

• advisors

• board opportunities

Same method.

Different goal.

What you’ll need

• LinkedIn

• ChatGPT or Claude

• 10–15 minutes

• one clear goal

This is the result we’re after: thousands of connections reduced to a small number of people worth actually thinking about.

Subscribe now

Why I wouldn’t just ask AI to “find useful people”

Because useful is meaningless without context.

Useful for what?

Getting hired?

Selling a service?

Finding an investor?

Getting into a target company?

The quality of the result changes dramatically when you give AI one specific outcome.

And that is where the actual tutorial starts.

The rest of this article is the exact 10-minute workflow I use:

1. Where to get the LinkedIn file

2. How to check AI has read it properly

3. The prompt that ranks the network

4. How to reduce the result to the 20 people that matter

The rest of this article is the exact 10-minute workflow I use.

STEP 1 — Download your LinkedIn connections

Go to LinkedIn:

Settings & Privacy

↓

Data Privacy

↓

Get a copy of your data

↓

Select:

Connections

Request the archive.

LinkedIn will prepare a CSV for you.

The export normally contains fields such as:

First Name | Last Name | Company | Position | Connected On

The exact columns may vary slightly.

That’s fine.

We mainly need:

name + company + role.

Privacy note: Your LinkedIn export may contain personal data. Before uploading it to any AI tool, check the columns and remove anything you don't need — especially email addresses. For this exercise, Name + Company + Position is enough. Also check the privacy/data settings of the AI service you're using, particularly if you're using a work account or handling company data.

STEP 2 — Upload the file to AI

Before uploading → delete Email Address and any unnecessary columns.

Open a fresh conversation in ChatGPT or Claude.

Upload the Connections CSV.

Before asking it to analyse anything, type:

Tell me how many connection rows are in this file and what columns are available. Do not analyse the connections yet.

Why do this?

Because I want to know the model has actually read the file before I ask it to make recommendations.

STEP 3 — Pick ONE goal

Do not write:

Find the most useful people.

Instead write something like:

I want to find people who could help me land a manager/director/VP-level role.

or:

I want to identify people who could plausibly buy my consulting services.

or:

I want to find potential sponsors for an event.

or:

I want to identify people who could help me access investors in B2B technology.

Be specific.

STEP 4 — Run the Network Scan

Copy this:

You are my Network Intelligence Analyst.

I have uploaded my LinkedIn connections export.

My goal is:

[INSERT GOAL]

Analyse the full dataset and identify the connections most relevant to helping me achieve this goal.

Do not simply select the most senior people.

Evaluate connections based on:

• relevance to my goal

• current role

• company

• seniority

• likely influence

• potential ability to create access or opportunity

Give each relevant person an Opportunity Score from 1–10.

Only show people scoring 7/10 or above.

Create this table:

Name | Company | Role | Opportunity Score | Why They Matter | Best Next Action

Sort highest score first.

Important:

Do not invent personal relationships, past conversations or facts that are not present in the file.

If more information is needed, write:

RESEARCH FIRST

Hit enter.

STEP 5 — Sanity-check the answer

Do not blindly trust the ranking.

Look for:

“I completely forgot I knew this person.”

Great.

“Their role is out of date.”

Research them.

“Relevant, but I barely know them.”

That matters later.

“No idea why AI picked this person.”

Remove them.

The point is not perfection.

The point is:

5,000 → maybe 30 plausible people.

STEP 6 — Reduce it to 20

Now ask:

From this shortlist, select the 20 connections with the strongest combination of:

relevance × potential impact × realistic ability to help

Do not rank purely by seniority.

For each person explain in one sentence why they belong in the top 20.

That’s your:

NETWORK 20

And stop there.

Don’t message them all.

Because in Part 2 we’ll look at something more interesting:

What if the person you need isn’t actually in your network — but someone who can get you to them is?

Your network is probably more valuable than you think.

This is one of the practical AI workflows I share here — things you can actually use, not AI news you’ll forget tomorrow.

Subscribe free and I’ll send you the next one.

Subscribe now

There’s a question I keep coming back to in conversations about enterprise AI:

Where’s the money?

Not where could we use AI. That’s increasingly easy to answer.

Give almost any process to an LLM and it will happily suggest 10 ways AI could improve it.

The harder question is:

Where is AI valuable enough to actually be worth doing? Or more simply, why bother? Why invest here and not there?

Because an interesting use case isn’t necessarily a valuable one. A technically impressive agent isn’t necessarily solving an important problem. And a successful pilot isn’t necessarily something worth scaling.

So, rather than write another post about it, I built something.

Where’s the Money?

It’s a simple AI value tool

You give it a little context about an organisation:

1. What are you trying to achieve?

2. Where does it hurt?

3. Where are people spending too much time?

4. What data do you actually have?

5. What constraints are real?

And instead of asking AI to generate a list of use cases, I ask it to make choices.

It returns:

• START HERE — strongest value + feasibility

• QUICK WIN — something worth proving now

• BIG BET — potentially significant, but harder

• DON’T START HERE — because sometimes the best AI recommendation is not to use AI.

I’m now adding a second layer: Can you actually capture the value?

Because identifying an opportunity and being ready to execute it are very different things.

The end result is deliberately simple:

Find the value. Test the readiness. Build the path.

How I built it

This was also part of my own AI learning.

I built the interface in Lovable and connected Claude through a secure server-side Edge Function, so the API key never sits in the browser.

The user answers a handful of questions. Those answers are sent as structured context to Claude, with a very specific job: Don’t sell me AI. Challenge where the value really is

.

Claude returns structured output rather than a long chat response, and the interface turns that into the Value Map.

No complicated multi-agent architecture. No RAG. No enormous technology stack.

For this problem, I don’t think it needs it.

And perhaps that’s part of the experiment too: learning not only what we can build with AI, but how little technology we actually need to solve a useful problem.

What I’m learning

The quality of an AI tool has surprisingly little to do with how many models you put behind it.

The interesting work is often in framing the question, constraining the reasoning and designing the output around a decision someone actually needs to make.

I’m still testing this one.

But I already know what I’m building next is focusing on better judgement.

One AI truth at a time. 🕶️
Alina

Hi, I’m Alina.

I’ve spent 15+ years working somewhere between technology, business, transformation and people.

These days, a lot of that world is AI.

I talk about it. I sell it. I build with it. I run a community around it. And I spend probably too much time wondering which parts are genuinely transformational and which parts are… very good PowerPoint. 😂

So I thought I’d start writing some of those thoughts down.

There will be serious topics here — AI strategy, agents, data, ROI, careers, leadership, the future of work.

But I’m going to try something slightly different.

I’m going to explain a lot of them through memes.

I’m calling them AI truths 🕶️ — little observations from spending far too much time around AI.

Some will be funny. Some slightly uncomfortable. Some I suspect people will disagree with.

Then, when one deserves more than a meme, I’ll write about what sits underneath it.

AI is changing incredibly quickly.

I’d quite like to explore it without pretending any of us has all the answers.

So, hi Substack. 👋

I’m new here. Where are the interesting AI people hiding?

One AI truth at a time. 🕶️
Alina