MagicHolz: over 4,000 conversions from a single placement

Up to that point MagicHolz had grown through Google, paid as well as organic, plus social ads. Demand for gifts there was huge but largely skimmed off. Further growth on the scale being aimed at was no longer available that way.

At that point influencer marketing was not structured at all: occasional requests came in from smaller creators wanting products in exchange for a mention — no process, no tracking, no analysis of whether anything came of it. For the founder the channel was personally unfamiliar territory too; by his own account he follows nobody.

Added to that was the worry that being a marketplace without an own brand might be an obstacle to partnerships. And as a bootstrapped company, MagicHolz could not afford to burn budget.

MagicHolz is a brand of MagicGroup GmbH, founded in 2020 in a four-square-metre office near Stuttgart. The company explicitly does not see itself as a classic own-label brand but as a marketplace for mechanical wooden puzzles and model kits from various manufacturers — comparable to a fashion retailer carrying other people's brands.

The range covers around 700 products, from marble runs through 3D models to puzzle boxes. An own label has only existed for about a year. The business is strongly seasonal, peaking in December: conversion rate there is, by the company's own account, more than twice the off-season level.

We started with a roughly three-month test phase, across several channels and with small monthly budgets in the region of EUR 10,000. The aim was not to scale immediately but to establish audience, vertical and buying behaviour from data rather than assume them. The obvious first hypothesis — wooden product equals DIY-minded audience — we dropped after a short time: too niche, not scalable.

Instead we moved towards broader audiences from gaming, outdoor and sport. From the start there was also a target system for cost per new customer, with a deliberately higher test value for the off-season, from which we could derive which creators could be rebooked profitably in the fourth quarter.

Being able to drop your own hypothesis

Moving away from the DIY-minded audience meant throwing away work already done. That switch is precisely what became the growth lever.

Calculate prices backwards

The target cost per new customer and the expected conversion rate together give you the most a placement may cost. That lets you negotiate on facts rather than on asking prices.

Every channel needs its own window

YouTube needs up to three months, Instagram stories are spent after a week, Twitch keeps delivering long afterwards. Judge every channel over the same period and you will cut the wrong ones.

Margin steers the product selection

Regardless of which product was shown, the same model was always the most bought. From then on every creator was given that product specifically, tied to the client's margin data.

In 2024, 88 placements ran across three channels. Lars Leenen presented the breakdown below himself at the OMR masterclass in May 2025 — which makes it the best-evidenced dataset we are able to show publicly.

The three channels behave completely differently. YouTube delivers 60 percent of the orders at a conversion rate of 0.13 percent — the reach is enormous, the share that buys is small. Twitch inverts that: 21,700 people reached, but a conversion rate of 7.74 percent and, at EUR 13.86, by far the best cost per new customer. Instagram produced 94 orders from 301,000 people reached and a ROAS of 0.34. The channel was subsequently cut from the mix.

Anyone looking only at reach would never have booked Twitch. Anyone looking only at orders would have kept Instagram far too long.

The best evidence that the channel carried is not a single metric but where the money went. Within one year MagicHolz more than doubled its push marketing budget — and raised the influencer share from just under a quarter to more than half.

Conversions over time

On YouTube, 18.2 percent of orders land on the first day after publication and 17.9 percent on the second; after that the curve flattens — but it does not stop. A meaningful share of orders arrives only after the third month. Twitch behaves entirely differently: there the peak sits in weeks three and four, not at the start. Instagram is essentially finished after two days.

That is precisely why Twitch was once cut too early and brought back later. Judge all three channels over the same window and you will cut the wrong ones.

Seasonality

Cost per new customer swings by more than fivefold across the year. On YouTube it ranges from EUR 108 in July to EUR 20 in December. A placement that looks unprofitable in summer can be the strongest single item of the year in the fourth quarter — with the same creator.

Products

One product gets promoted, the whole range gets bought. In a December 2024 partnership the Luminous Globe was promoted; of 179 items sold, 42 were the promoted product and the rest spread across twelve other models. Steering therefore runs on margin, not revenue: the Luminous Globe carries 62 percent, other models under 40 percent.

Verticals

The obvious niche was not the best one. Entertainment produced the highest net revenue in 2024 at EUR 111,461; Movies the lowest cost per new customer at EUR 8. DIY — the vertical where a wooden model kit ostensibly belongs — sits at EUR 24 and EUR 39,431. Autos and Knowledge cost EUR 60 and EUR 58, more than double DIY.

Channel

Average order value is highest on Twitch at EUR 58.51, ahead of YouTube at EUR 50.31 and Instagram at EUR 44.17. On ROAS the gap is starker still: Twitch 4.22, YouTube 1.65, Instagram 0.34.

Age structure

Here too the analysis shows the opposite of what you would expect. The cheapest buyers are 45 to 54 year olds at EUR 11 per new customer, followed by 25 to 34 year olds at EUR 14. The youngest group, 18 to 24, is the most expensive at EUR 26.

The most striking single finding of the year: the same six creators, booked once in March and once in December, deliver completely different results. For Felix von der Laden, cost per new customer fell from EUR 218.57 to EUR 26.48; for BeHaind from EUR 46.68 to EUR 6.35.

Not one of these creators got better between March and December. What changed is the timing. Cancel in March because a creator is not delivering and you lose precisely the one who carries the fourth quarter.

The 2024 Easter campaign deliberately ran against the company's own numbers: 32 placements, EUR 114,178 of budget, EUR 91,703 in revenue. Cost per new customer came in at EUR 64.51, far above target. Viewed as a campaign, it was a loss.

Viewed as data collection, it was the foundation for the fourth quarter. Those 32 placements produced the information about which creators, which verticals and which age groups carry. The December figures in the table above come from the very channels that failed in spring.