Ask Mark

On a Monday afternoon, one of my direct reports told me she was quitting to go freelance. I was genuinely happy for her, it felt like the right move, but I also felt a wave of panic. Our Growth team is small, and she was the one running most of our social ads experiments. She wasn’t just a channel expert. She was core to how one of our product squads moved. One of the people most affected was Elyes, the Product Manager on that squad. There was no easy replacement. No one to brief. No clean handoff. So I gave myself a couple hours to find a temporary solution; something that would buy us time and keep the squad from freezing.

By Tuesday afternoon, I had built a custom GPT named Mark. The idea was simple: give the squad someone to talk to while I reorganise.

The flow is simple:

• Ask Mark first.
• Come to us once you have a better idea for your next iteration.

That's it. Mark wouldn't replace anyone but he could help unlock, clarify and challenge thinking.

How I built Ask Mark

Instructions

A lot of how we run projects in the Operations department comes from Discovery Discipline, a book that’s helped us structure our thinking around project dynamics, ownership and decision-making.

So before writing a single instruction, I started by sharing the squad's frame, just like we would in a real kickoff:

• Who’s involved.
• What’s already been tried.
• Budget guardrails.
• Success metrics.
• Known constraints on execution.

All of this went into Mark’s Instructions. I wanted him to speak with clarity, not assumptions.

Then I added behavioural rules:

• Always ask clarifying questions before responding.
• Say "I don’t know" if the answer isn’t in the knowledge base.
• And most importantly: be a little sassy. When someone leaves a team, that levity becomes a kind of glue.

Knowledge

The most important decision wasn’t what to upload but who to trust.

To ensure the quality of Mark’s answers, I didn’t just dump documentation or random articles into his knowledge base. I handpicked operators I’ve followed for years; people whose thinking has consistently raised the bar for me and my team.

Instead of centralizing everything into one file, I created a PDF for each author. This wasn’t just for convenience; it’s a way to flag, debug, and reflect when something feels off. If Mark says something that doesn’t land, I know exactly where it came from.

"It really feels like I have the right resource in front of me" Elyes told me after a few uses. "For someone who’s not very marketing-oriented, it helps me build strategies with a level of autonomy I didn’t have before."

Mark’s usefulness depends entirely on the judgment behind his sources. Building him meant being just as careful about input as we are with output.

Mark, the unexpected teammate

A few days later, I started hearing things like:

"I asked Mark, and he pushed back."
"Mark doesn’t like that idea."
"Mark says we’re missing data on last week’s performance."

That’s when I realized it was working.

Mark wasn’t replacing anyone, but he became a decent proxy. He helped Elyes move faster on early iterations. He answered questions on performance snapshot and unblocked decisions that otherwise would’ve lingered until our next catchup.

"He’s not there to say yes to everything" Elyes said. "When the idea I suggest isn’t great or needs reworking, he says so and always offers better alternatives."
One recent example? "I wanted to run three ad sets with a different performance metric in each. Mark said that didn’t make sense and suggested testing different audiences instead."

What Mark can and can’t do

Mark doesn’t give green lights. He doesn’t ship experiments. He doesn’t think like a full teammate.

But he does something surprisingly valuable:

• He does give structure.
• He orients ideas back to the right brand (since we have several brands).
• He makes you stop and ask better questions.
• He sharpens ideas that are still fuzzy.
• He helps you move forward when you’d otherwise wait.
• He provides a layer of autonomy for people who aren’t marketers but need to make marketing decisions.

Sometimes, that’s all you need.

"He saves me a lot of time, especially when it comes to interpreting campaign results."

He doesn’t give you the final word. But he makes your first draft better and faster.

What's next for Mark

Right now, Mark answers questions based on static knowledge. But I'm already thinking about the next version.

With Actions, Mark could connect to our Meta Business Manager and:

• Pull live performance data
• Flag underperforming creatives
• Even launch or pause campaigns based on criteria we define

He wouldn’t just advise. He’d act. That’s not today. But it’s not far off either!

We all talk about LLMs as tools, sidekicks, copilots. But in small teams, when people leave and priorities stay the same, you sometimes just need someone to talk to.

Mark was my scrappy, no-budget answer to that.

It didn’t solve everything. But it kept the team moving.

And in moments like this, that’s more than enough!

Special thanks to Elyes for believing and testing Mark in the early days. And thanks to Maeva for laughing out loud while using Mark!