Storytelling at scale

How to tell a thousand amazingly human stories with AI

On January 7th, 2015 I hopped on a plane and moved to London. A local startup accelerator decided to invest in my machine learning startup. Me and Zoltan (my then cofounder) built an algorithm that would allow us to analyze how visitors are related to each other on a website with near real-time accuracy, identifying outliers who were responsible for traffic.

You know like when you see a funny video and then show it to your entire family on a Messenger group? Without you, Youtube would never have gotten those extra 8 views. Chances are you do this often.

We called them micro-influencers, even though the term didn’t exist yet.

Nobody was able to do that before because of two things: no reliable way to get the data and real-time analysis posed an NP-complete problem. We solved it with a 20-line JS snippet and smart architecture.

I was young, inexperienced, and — in hindsight — in way over my head.

On Kickoff Day all the partners (there were quite a few of them sat down with us asking a bunch of questions. A rugged-looking man in his 50s was sitting in the middle listening. He was wearing Fivefingers toe shoes, cargo pants, and a T-shirt that was at least 10 years old.

After about 30 minutes of conversation, he looked up and asked me a simple question:

“What you’re trying to do has been attempted many times since I was running strategy at Dell in the early 90s. If you can convince me that you solved this, I’ll personally mentor you because it means you’re one crazy motherfucker.”

Then I went on to explain to him how we were caching the network to bypass network analysis. All we needed was to update node states and we were good. It was an ugly workaround that evaded the NP-complete problem, but it worked. We identified the right people so quickly, that when they started bringing in a few people we could use that as a reliable predictor for future uptick in traffic.

In other words, our algorithm predicted what would go viral next.

The guy looked at me for a minute then said in the most nonchalant way possible:

Fuck me you’re crazy. We’re friends now.

Years passed and he not only mentored me but became somewhat of a fatherly figure to me. Most of what I think about business I learned from him.

His name is David Prais, Chairman of the Greater London Investment Fund.

Good stories open doors

David is probably the best storyteller I’ve ever met. Wherever we would go, he would instantly build rapport with strangers. He is not charismatic in any way and you would probably miss him if you walked past him. He lives a very private life. But when he opens his mouth, everyone listens — and has a great time.

I’m nothing like him.

I was always socially awkward.

When I moved to London I started going to networking events only to anxiously stand by the canapés, shoving them into my mouth to have a legitimate reason for being silent. I still made connections but I never initiated them. I was just a blue-collar kid from rural Hungary trying to blend in at the rooftop area of Unilever HQ at a private investor soirée and all I could think of was how embarrassing the whole thing was.

I’m a lot more comfortable with networking now, but I’m still learning how to tell good stories that open doors.

But why is it that prolific networkers are always amazing storytellers? What’s the secret?

Because we’re human. As Yuval Noah Harari said:

“Homo sapiens is a storytelling animal that thinks in stories rather than in numbers or graphs, and believes that the universe itself works like a story, replete with heroes and villains, conflicts and resolutions, climaxes and happy endings. When we look for the meaning of life, we want a story that will explain what reality is all about and what my particular role is in the cosmic drama. This role makes me a part of something bigger than myself, and gives meaning to all my experiences and choices.”

A good story is the most valuable asset you can have.

Good stories open doors.

From a social perspective, your worth is derived from how good your stories are.

Because good stories command attention and we all know that today we live in an attention economy. So the next time you open Substack or Youtube, think about this:

It’s a marketplace where people trade their stories in for attention. Better stories mean more attention received. More attention means more opportunities.

Stories are the ultimate asset in the human world.

…the only problem is that we’re pretty shit at storytelling. To tell stories is counterintuitive. To tell stories every day, online, creating content over and over again, it’s…tiresome. You need to force your nervous system to put yourself out there again and again on autopilot.

Then the fruit of this hard work is worthless, better used as cannon fodder for training AI models (or at least according to Zuck).

Yet this is just how things are.

If you have any economic incentive in human society you need to be great at telling amazing stories, at scale.

Banking confidence

Pia Silva founded no BS Agency. In her TEDx talk, she talks about “banking confidence” by taking leaps into the unknown, because overcoming challenges builds your confidence.

I can relate to that. I remember when I got my first job. I had to make a decision: do I take a well-paid internship position at IBM as a marketing intern or do I take an unpaid internship at a small business as a product manager? I choose the latter, taking a leap of faith.

Choosing IBM would’ve given me a job.

Choosing the unpaid internship gave me a career.

Watch Pia’s video and you’ll see how banking confidence gives you an endless source of amazing stories too. But it’s not enough to have those stories, you need to be able to tell them too.

Storytelling at scale

Pia helps branding agencies get off the ground by embracing her radical philosophy and it works well.

But right now, Pia’s business is facing a different challenge. They’re at a point where unlocking more growth means more investment into content generation. They’ve outgrown themselves. They can tell great stories, and convert them into attention but there are too many stories to tell now.

They’re a small, lean team and frankly, none of them wants to work all the time so complicating the business with scaling is not really attractive. I mean let’s face it. Working with an agency is at best frustrating and, at worst damaging, while hiring and building your own team will turn you into a slave of your own business.

Naturally, Pia doesn’t want to give up the freedom she worked so hard to build.

Same Pia, same.

So, I had a chat with Pia and her COO Amanda because they thought that AI should be able to do this, right?

Um, right. It should.

But have you seen all the social media accounts that are managed by AI?

They’re worse than the Soviet Lord of the Rings.

We needed to spend some time on this problem.

Amanda asked me “How much damage could you do in a single day if you had access to both of us?“

I called it a “Power Day”. There’s probably a better name for it, idk but I figured we’d power through the problem until we got a solution.

We sat down and spent 4 hours together digging into their business and figuring out if AI can be plugged into it. Then I spent another 4 hours of research and prototyping to build an action plan to deliver a solution ~3 months.

Off of this session, we understood that we needed storytelling at scale, but we can only do it at a high enough quality (trust me, brand experts do not take it lightly when we talk about the risks of damaging their brand with shit content).

So the solution was born.

Personal Timelines and Narratives

They already have a Circle community with tens of thousands of posts, and comments. All the activities, client introductions, wins, challenges, everything is logged in their community. Even better, all their sales calls, onboarding calls, intake forms, everything is recorded.

(Amanda has been doing an amazing job recording everything in Airtable which makes this whole project a lot easier. Most companies do not have someone like Amanda, but that doesn’t mean this project isn’t doable, it just takes a bit more work.)

So I said hey, let’s collect EVERY story we have.

Every single customer came to Pia with their own story. They told her that story. Pia changed the course of that story with her coaching and mentorship.

Let’s collect those stories first.

I then went and transcribed all recorded calls with a simple Make scenario, and saved the transcripts into Airtable so we can access them later.

Then I used Gemini 1.5 Pro to take all the raw data (posts, comments, transcripts, everything) and generate a Personal Timeline and a Personal Narrative of every single student she has.

Once I got them done, I created an interface in Airtable so they could scroll through the stories. Now we can feed this story to another agent that will enrich the stories with different types of media so we can publish them on social media.

With this simple approach, we created probably the most personal, most valuable, and most relatable stories she could’ve had: her customers.

I recorded the process and uploaded a timelapse video of me building it with a deeper explanation of how this works. You can watch it here:

If you liked this video, don’t forget to subscribe to my Youtube channel. I’m still pretty bad at this video thing but enjoying it, so feel free to give me advice on the kind of content you’d like to see more of!

Content Generation Machine

Note how we’re not using AI to generate stories. We feed it on our own. We’re only using AI to repurpose what we already have. It takes more work to build it but it gets rid of hallucination issues and quality issues.

The next step is that we take all these stories — now in a format that’s worthy of generating content from — and start enriching the stories with media.

Once that’s done, Pia’s VA will need to spend 1 hour every week scrolling through all the generated media selecting the really good ones, setting a schedule within the Airtable interface, and letting the automation do its job.

Very human stories are being told at scale — by AI.