Martinus: From Freefall to a Profit Engine

Tomáš is constantly testing new ideas that accelerate our profit. The big benefit is his overlapping knowledge from search ads to SEO, websites analytics and PPC automatization especially. Within the season he managed to double the revenues keeping the same ROAS.

Jan Keselak

Marketing manager at Martinus

Six Years Inside a System That Had to Be Understood Before It Could Be Fixed

The Situation

It started quietly.

Revenue slipped by roughly 20% YoY. Then 40%. Then 50. Then almost 60%. Month by month, the decline accelerated — right before peak season.

Martinus’ Czech business relied heavily on PPC. The agency reacted in the usual way: loosen ROAS targets, spend more, chase volume.

But revenue didn’t recover.

Cash started burning faster.

That’s when I joined — end of October 2017 —with clear goal. Stop the freefall and save the peak season.

Seeing What Others Missed

At first glance, performance had looked fine. The revenue quietly eroded. Nothing dramatic. Just enough to mislead. To start looking elsewhere:

• Products are too expensive
• Landing pages aren’t good enough
• The market is saturated

But the pattern didn’t fit.

Once I separated retention from new customer acquisition, the real problem surfaced.

Smart bidding had optimized itself into a corner — pushing more and more budget into retargeting and bottom-funnel queries. It was harvesting existing demand, not creating new one.

ROAS masked the damage.

Revenue paid the price.

Regaining Control

The first real decision was uncomfortable. Turning smart bidding off.

Not because automation is bad — but because you can’t debug what you don’t control.

Campaigns were rebuilt around query intent, not categories. Reporting was restructured so I could see performance across many layers.

With control restored, decisions became obvious.

• Waste was cut fast.
• Growth drivers were scaled aggressively.

Within weeks, the decline stopped.

By the end of the year:

• The freefall was reversed
• +16% YoY growth was recovered
• The strongest month delivered +65% YoY growth
• ROAS improved instead of collapsing

The season was saved.

Curiosity Beats Comfort

The account had more than 500 conversions per month.

Smart bidding should have worked well.

So why didn’t it?

Instead of moving on, I tested.

In 2018, I began running controlled experiments.

First, the rule-based bidding logic vs smart bidding

The result: My bidding logic beat the smart bidding

The work became part of my academic thesis and was later presented at PPC Date conference – see slides here. (Thanks Matouš Ledvina, Google analyst to validate the smart bidding setup and professor Stříteský for validating the results with academic accuracy.)

Second, on structure alone.

What are the variables/inputs I could use to improve my bidding logic.

The best predictor for conversion rate? – Competitiveness

Competitiveness – Our price compared to the market average. Hard to get, but I used a few app scripts to get the scraped price comparison data into the feed. It predicted conversions far more reliably than discounts. So products with small discount but when competitors did not discount at all often outperformed discounted items that were still overpriced compared to competitors.

Yet it pushed the discounted products more.

Which means less margin. And Revenue and ROAS are not the real goal.

Profit was.

I split the products by margin level, not category. Exact margins weren’t available, so approximations were built using category data, discounts, and historical performance.

Accuracy was ~70%.

Good enough.

Low-margin products had strict bidding rules while high-margin products were allowed to scale.

The business results by the end of 2018:

• +22% YoY revenue
• +34% YoY ROAS
• Significantly higher profit

Making Profit the Signal

Margins were calculated weeks after the click. Perfect data wasn’t coming.

So we were building and adjusting something imperfect – a system that would replace revenue with gross margin as the primary conversion.

With the Martinus team — especially Martin Kružík — margin data was sent from CRM to Google Analytics via Measurement Protocol, then imported into Google Ads.

Manual bidding rules evolved as accuracy improved.

And it was a breakthrough year. Revenue dropped by 11%.

But Profit exploded.

• +600% in the Czech Republic
• +70% in Slovakia

Imperfect margin data beat perfect revenue data.

Profit Generator 2.0

By 2020–2021, margins flowed directly into Google Ads via offline conversions. Click IDs were reused across platforms. Pricing and bidding became profitability-driven.

The system was finally complex enough that smart bidding worked — and outperformed even my best rule-based setups.

To be honest, the margins sent to Google Ads were were still messy:

Why This Worked

This story isn’t about Google Ads.

It’s about how you think when systems fail.

Break them apart.

Understand the constraints.

Feed them better signals.

Move faster. Don’t wait for perfection solutions.

Accept that most experiments fail.

The ones that work pay for everything.

That’s why I love this work.

Not because of the wins —

but because every day there’s another system waiting to be understood, and pushed just a little closer to the goal.