“We just spent $$$$ on our new predictive model. What worked? An intern and a spreadsheet.”

Why This Matters

Some data teams are building solutions that are smarter than they need to be - and dumber than they think.

In our chase for sophistication, we often ignore the Pareto Principle. We overfit for elegance when brute-force practicality would deliver 80% of the value in 20% of the time.

This newsletter is a wake-up call for every data leader trying to be clever - when being useful is far more powerful.

The Problem: We Overthink, Overbuild, and Overcomplicate

Let me take you back to when I was leading the data team at Rocket Internet - the venture builder behind Zalando, Delivery Hero, and HelloFresh.

Like everyone else, we wanted to build a recommendation engine for our e-commerce portfolio companies. Not just any engine, we wanted our own version of Amazon’s “Customers who bought this also bought that.”

So we did the usual:

• Explored all the big-name vendors.
• Built custom machine learning models.
• Tuned algorithms for weeks.
• Spent big money

And the results?

Meh.

We didn't see any significant uplift in Profit Contribution per Visitor.

And then, frustrated and borderline embarrassed, we had an epiphany: The top 1,000 bestsellers contributed ca. 50% of total revenue. So we had an intern manually assign relevant recommendations for those products. We then just ranked those items based on some simple rules:

• Stock availability
• Profit margin
• Click-through rates
• Conversion Rates
• Seasonality

That’s it.

The outcome? 12% uplift in Profit Contribution per Visitor. This simple heuristic beat every ML model we tried. And more importantly - it shipped in a week.

Heuristics > AI (In Most Real-World Cases)

This isn’t just a fluke or a funny war story.

It’s a pattern.

And it’s backed by science.

My friend, Dr. Markus Wuebben, even wrote his PhD Thesis about it. Together with Florian Wangenheim he published the paper "Instant Customer Base Analysis: Managerial Heuristics Often "Get it Right" in the Journal of Marketing (2008). The paper compared sophisticated stochastic customer base analysis models (like Pareto/NBD and BG/NBD) with the simple heuristics managers use every day - things like recency and frequency rules.

And the kicker?

Simple heuristics performed just as well - and sometimes even better - in predicting customer behavior, segmenting users, and identifying future best customers.

Why?

• Heuristics align with how decisions are made in the real world: fast, under uncertainty, and with imperfect data.
• Managers optimize for accuracy and cognitive effort, not just precision.
• In real environments, feedback loops (especially in direct marketing or ecommerce) allow people to learn fast and iterate: without waiting for months on the "perfect data stack".

So yes, you could build that predictive CLV model using BG/NBD: but in practice, your stakeholder will probably go with “customers who bought 3+ times in the last 6 months.”

And they'll often be right.

Why We Ignore the Simple Stuff (Even When It Works)

Let’s be honest.

There’s status in building complex things.

ML models are sexy. Heuristics feel… pedestrian.

But this mindset costs us:

• It delays impact. While you're tuning hyperparameters, the business is bleeding margin.
• It erodes trust. Stakeholders don’t care how smart your model is if it doesn’t drive decisions.
• It creates maintenance debt. That model you built last quarter? Nobody’s monitoring it now.

Actionable Takeaways 🛠️

Here’s some food for thought for those quiet days around Christmas:

✅ Audit your most complex solutions

→ Ask: “Did a simple heuristic exist here?” If yes: what would it have looked like?

✅ Institute a “Heuristic First” policy

→ For every data product, build the simplest version first. Validate. Iterate only if needed.

✅ Educate your stakeholders

→ Show them why simplicity often wins - and when it doesn’t.

Final Thought: Don’t Build a Ferrari When a Bike Will Do

There’s a time and place for deep models, AI, and advanced analytics.

But most of the time?

A sharp intern with a spreadsheet and the right business context will run laps around your overengineered solution.

Build less. Deliver more.

See you next week.

Cheers,

Sebastian

P.S.: In my masterclass 👉 "From Dashboard Factory to Strategic Partner" I'm sharing the whole Rocket Internet story and how I developed a growth-oriented mindset to build data teams that consistently deliver high business impact.