Data Matas

“The best data teams aren’t built around superheroes. They’re built around trust.”

In Series 4, Episode 3, we sit down with Toby Henley Smith, Senior Analytics Engineering Manager at Moneybox, to talk about what makes data teams genuinely valuable to a business, and why some accepted “best practices” might actually be getting in the way.

We get into:

→ Why great analysts bring trust and rationality to decisions, not just reports
→ Why startups need psychological safety to experiment, get things wrong and learn quickly
→ How over-engineering and “best practice” stacks can become expensive distractions
→ Whether AI agents writing SQL could make heavyweight semantic layers less relevant
→ How AI is shifting engineering from writing code to defining problems and reviewing solutions
→ Why AI could accelerate the development of junior engineers rather than replace them
→ What tighter budgets and deliberate constraints can teach data teams about building better systems
→ Why cutting cloud costs starts with deciding what the business genuinely cannot live without

At the heart of the conversation is a simple question: are we building better data systems, or just more complicated ones?

"AI won't replace data engineers. But engineers using AI will."

In Season 4, Episode 2, we sit down with Julian [add last name + role] to unpack what's actually changing in data engineering — and what's about to disappear.

We get into:

→ Why dashboards as we know them are dying (and what replaces them) → Does AI really need a semantic layer? Julian's answer might surprise you → The low-code trap quietly racking up tech debt across data teams → Where AI is genuinely useful today: tech debt, testing, and data governance → The one skill that still matters most when AI can write the code → A spicy closing question for the next guest about cloud cost

⏱ Chapters 00:00 — Intro 01:30 — Julian's path into data 04:00 — The career pivot that changed everything 07:30 — How his team is adopting AI (and who resists) 10:00 — Does AI need a semantic layer? 13:00 — Local models are closer than you think 15:30 — Where AI is actually working: tech debt, tests, governance 19:00 — What the data industry is getting completely wrong 23:00 — The belief Julian held 3 years ago that's now wrong 26:00 — A question for the next guest

Season four, episode one of the Data Matas Podcast.

Aaron Phethean sits down with Kevin Sampson, the first data hire at Vertex Service Partners, who joins us after four and a half years at Amazon. The conversation is about what it actually takes to build a data and analytics platform from zero.

We cover dashboard sprawl at Amazon, the gap between insights and answers, why "confidently wrong" is the worst failure mode an LLM can have inside a real business, and a hot take on whether AI even needs a semantic layer anymore.

This is the first episode of our new format. Every guest faces a hot take. One question they have never heard before, designed to challenge how they think on the spot.

Data Engineering as a Team Sport: Coaching, Accountability, and Pragmatism

In this episode, Sam Wrench, Lead at Reality Mine, joins us to explore what actually makes data teams perform at a high level.

We cover:

Why data engineering breaks down when teams optimise for tools instead of people
 How coaching principles translate directly from elite sport to data leadership
 Why dogfooding your own data creates faster feedback loops and better quality
 How to think about AI as a managed colleague, not an autopilot
 Why pragmatic, security-first tooling often beats chasing industry hype

This is a grounded conversation for anyone building data platforms that real teams and real decisions depend on.

Analytics Engineering: Internal Risk vs. External Rigor | Ft. Jack Doherty (Fresha)
The stakes are higher than ever for Analytics Engineers. When your data becomes a core, customer-facing product, the game changes.

Jack Doherty (Head of AE, Fresha) discusses the massive difference between internal and external analytics:

Risk vs. Rigor: Why internal projects can move fast (risk), but product-facing data demands DevOps-level rigor, testing, and governance.

Real-Time Data: The technical shift from scheduled batches to CDC for meeting customer demands for speed and consistency.

The Missing Link: Why the Semantic Layer is the future of AE, crucial for codifying business logic and powering accurate AI/Chat interfaces.

A must-watch for any AE treating data as a product.

Stop Coding, Start Diagramming: How to Build Data Platforms That Deliver

If you're rushing to hire a data engineer before you have a clear business question, you’re doing it backwards.

I'm joined by Teddy Bernays (Freelance Data Engineer) to unpack his "business first" approach. Teddy shares his journey and explains why simplicity and a solid plan always beat the latest tech stack.

His top advice: "Find the problem you want to solve first. Is data the answer? Only then should you start building."

In this episode, we cover: 
▶️ Why you should hire a Data Analyst before a Data Engineer 
▶️ The "Diagram First" rule for technical projects 
▶️ How to escape the painful world of legacy spreadsheets 
▶️ Finding freelance clients in the real world (get off LinkedIn!) 
▶️ Using AI to finally solve your documentation problems

Build Strategy First. Choose Technology That Deserves It.

If your company’s first move is buying another tool, this episode’s for you.

I’m joined by Dylan Anderson (Head of Data Strategy, Perfusion) to unpack why so many data strategies fail before they begin — and how to build one that actually delivers.

Dylan’s worked with some of the world’s biggest organisations, helping them turn data chaos into clarity. His advice is refreshingly simple:

“Buying more tools doesn’t give you a strategy.”

In this episode, we talk about:
 ▶️ Why data strategy starts with people, not platforms
 ▶️ How honest technology builds trust
 ▶️ Why simplifying your stack beats automating chaos
 ▶️ The real future of AI — and it’s not chatbots

It’s a straight-talking conversation about cutting noise, making smarter technology choices, and building data foundations that actually deliver.

Most data teams think their next dashboard will finally deliver clarity, but every new report just adds to the noise.
In this episode of Data Matas, Aaron Phethean speaks with Phil Thirlwell, former Director of Analytics & Data Strategy at FIS, about the three things every data leader should kill before building another dashboard: dashboard sprawl, service-desk habits, and KPI overload.
Phil shares how he tackled 600-plus Power BI dashboards, why “can you just…” requests destroy team focus, and how the best data leaders rebuild trust through simplification and shared ownership.

You’ll learn:
 ▶️ Why dashboard sprawl erodes clarity and confidence
 ▶️ How to escape the “service desk” trap and focus on real business outcomes
 ▶️ Why trimming KPIs forces better decisions and stronger alignment
 ▶️ How co-developing metrics builds trust across teams
 ▶️ Why AI and automation can’t fix messy data foundations

This is a practical, candid conversation for data leaders who want less noise, more trust, and teams that deliver measurable impact.

📺 Subscribe for more real conversations with data leaders: https://www.youtube.com/@matatika
🎙️ Listen on all major podcast platforms: https://www.matatika.com/podcasts

Most of the modern data stack has transformed, pipelines, compute, governance. But BI? It’s still the same dashboards and reports we were using 20 years ago. Expensive, read-only, and delivering the worst ROI in the stack.

In this episode of Data Matas, Ollie Hughes, CEO of Count, joins Aaron Phethean to share why BI tools eroded trust, why AI won’t fix reporting chaos, and how data teams can escape the “service trap” to become real decision-making partners.

You’ll learn:
 ▶️ Why BI tools are the worst ROI in the modern stack
 ▶️ How the “service trap” caps your team’s value
 ▶️ Why AI makes reporting faster but not better
 ▶️ How to build trust in data beyond accuracy alone
 ▶️ Why ruthless prioritisation is the ultimate lever for data leaders

This is a practical, candid conversation about the real challenges data teams face — and how to refocus BI on clarity, trust, and decisions that matter.

📺 Subscribe for more real conversations with data leaders: https://www.youtube.com/@matatika
🎙️ Listen on all major podcast platforms: https://www.matatika.com/podcasts/

Want 97% AI Adoption? Start By Saying “Not Yet” 

Hypebeast withheld access for 10 weeks, teased value, and turned demand into near-universal adoption. 

In this episode of Data Matas, Aaron speaks with Sami Rahman, Director of Data & AI at Hypebeast, about what it really takes to embed AI inside a modern business.

Sami shares how his psychology background shapes his approach to adoption, why fear of AI is more about broken safety nets than the technology itself, and why Hypebeast uses AI as a force multiplier — not a replacement for creative teams.

He explains how he deliberately teased AI’s potential for 10 weeks before giving access, using curiosity and scarcity to spark demand. The result? 97% adoption across the company.

Listeners will also hear how Hypebeast prioritises boring but valuable use cases — automating system updates, consolidating research, scanning trends — and why Sami treats AI agents as disposable tools with clear lifecycles, not permanent fixtures.

It’s a grounded, practical conversation about the human side of AI adoption and the discipline it takes to keep hype from overrunning reality.

Watch the full episode on YouTube, or listen on Spotify and Apple Podcasts.

👤 Sami Rahman on LinkedIn
👤 Aaron Phethean on LinkedIn
🎧 Data Matas Podcast
📺 YouTube
🌐 Matatika Website