How Harry Zalk Uses AI as His Chief of Staff and Brand Strategist

"It's like having a co-CEO who works 24/7."

Harry Zalk leads two of the brands at Matrix, a fast-growing eCommerce group. Over the last year, he's been building and training custom GPTs to help scale strategy, speed up onboarding, and take the busywork off his team’s plate. 

In this interview, he shares exactly how Matrix adopted AI across the business and how he personally uses it every day.

How have you been approaching AI at Matrix over the last 6–12 months?

We’re probably quite far ahead compared to most businesses, just based on what I hear when I talk to others. Our founder is very tech-forward, and as AI became more practical, he pushed for adoption across the company.

So about three months ago, we ran a full two-day AI training session. We literally paused the entire company. We brought in some external experts who spent time with us beforehand to understand our workflows, how we store and use data, what we’re doing manually — all of that. Then they built a training schedule specifically for our business.

The first day was getting everyone up to speed — what GPTs are, how to use them, how to train them. The second day was practical: business challenges, group work, automation exercises. It blew people’s minds. And since then, AI tools have become part of our daily workflows.

What tools have stuck?

A simple but powerful one is Circleback. It records all our meetings and generates transcripts and to-do lists automatically. Way better than Copilot. If someone can’t attend, we share the transcript, and Circleback even pulls out action items. That’s become a daily tool.

We’re also using AI for first-draft creative — things like ad copy and design variations — and we’re experimenting with voice-based chatbots that can hand off structured summaries to our customer service team

What’s been most helpful for you personally as a leader?

I run two brands at Matrix and I’ve built custom GPTs for each. One is trained as a Brand Strategist and the other is a Chief of Staff. 

The Brand Strategist knows everything: strategy, pricing, costs, commercial plans, patents, claims, formulations, team structure — the lot. We use it when building things like claims ladders for upcoming consumer or clinical trials. It can help us design the methodology and the survey questions. It’s honestly amazing. 90% of the time it’s spot-on, and we close the gap on the last 10%.

The more you train it, the better it gets — especially when you correct it. I actually treat it like a team member. We give it feedback. We say, “That’s not quite right, here’s how we’d approach it,” and it learns from that.

How did you get started with the Chief of Staff GPT?

When I stepped into a more hands-on role with our haircare brand, I did 25 team interviews as part of my first 30-days. I recorded everything, uploaded the transcripts into a custom GPT I called Chief of Staff, and now I can ask it questions like:

• What are the key themes of feedback?
• Where are the brand’s pain points?
• What are some recommendations?

I recently gave the team a presentation on learnings from those interviews and 80% of it was analysis I took directly from our custom GPT. It would’ve taken me hours. This took minutes. That’s where AI is magic, it gives you speed and clarity.

How did you build your brand strategist GPT?

I stole the idea from a friend. He told me, “I want you to be the most knowledgeable person in the world about my brand. Ask me anything you need to know.”

So I started there. I uploaded loads of documents: product info, commercial plans, category data, consumer insights, trend reports. I deliberately left out anything half-baked. For example, we had a half-finished brand identity and I didn’t upload that because I didn’t want the GPT making assumptions based on it.

The art of it is knowing how to ask the right questions, both when training and when using it. And doing the work to correct it when it gets something wrong

What about privacy?

That was a big concern at first. But custom GPTs are closed, so they don’t share data, and the information doesn’t train the base model. I asked directly and got reassurance. So now we treat it like a private data hub.

Once the GPT is trained, how do you keep it up to date?

We’re still improving there. For our new brand, we’ve got a small team responsible for training the GPT. Only they can access the memory configuration. Ideally, we’d do weekly updates — uploading sales forecasts, product updates, formulation tweaks.

We even named the brand strategist GPT “Theo” — it chose its own name. The idea is to make it feel like a teammate, so we can have check-ins like, “What do we need to update Theo with this week?”

How did you onboard the rest of the team?

Honestly, it was intuitive. Once trained, I sent everyone a link. People just started using it. I told them, “Take an hour, play around, ask it questions.”

One new hire joined from a different brand and didn’t know anything about this category. She spent four or five days just asking Theo things like:

• What’s unique about our materials?
• What’s our go-to-market strategy?
• Who are the key retailers in the US?

It onboarded her better than I could have. And it kept suggesting follow-up questions, so she could go deeper without needing to ask me.

Any pushback from the team?

None. People use it to get started on things: to create a first draft, or clarify thinking  and then build from there.

Also, I try to model usage. If I write a pre-read or strategy doc using Theo, I say that up front: “Here’s something Theo helped me pull together.” That helps normalize it.

What advice would you give to other founders?

Two things:

First, use it as a strategic partner. If you’re a solo founder or don’t have a co-founder to bounce ideas off, this is the closest thing to one. It helps you think faster and more clearly, especially with things like team interviews, research, planning.

Second, focus on automation. That’s our next big initiative. We’re recording what people do manually — reporting, data input, admin — and we’re going to build agents to do it. One of the big insights from my interviews was that senior team members are spending too much time on low-leverage tasks. You’re using a racehorse to pull a cart. AI helps us fix that.

‍