1. The bubble meter
So we looked at the numbers, and this week the bubble meter looks like this:
I would say the market is still high but more or less stable. With a probably healthy correction in the US.
But even if the market as a whole looks expensive and we are mostly risk-on, that doesn’t prevent us from finding interesting opportunities.
Well, when you go small, that is.
Many stocks sit in our funnel for processing and maybe eventually a deep dive.
Here’s an interesting one ⬇️
2. Stu(o)ck in the funnel
It’s time for another company that is stuck in my funnel. Just outside the typical market cap I go for (I usually stay under 500M), but it’s an interesting setup.
That company is Mensch und Maschine (MUM)
What is it? : A German engineering-software group; that owns hyperMILL (5-axis CAM), SOFiSTiK(structural/BIM) and the largest Autodesk value-added reseller in Europe
Market pricing? : €600M market cap, €620M EV, 19x FY25 earnings, 17x FY26e
Balance sheet? €20M net debt, 6% dividend yield, €50M of an existing credit facility still earmarked for buybacks
Does it have a moat? Let’s look at some numbers. 3% churn, 95%+ renewal rates, 55% recurring gross profit, and high switching costs
What is the bet? Three years of one-offs, and German macro pressure compressed the P/E from 35x to 19x. Future earnings are set to compound in the low teens on a highly recurring, sticky revenue base.
So what’s the problem? Management has missed its own guidance two years in a row. The 72-year-old founder owns half the company (I like skin in the game, but too much skin can also lead to issues; remember Anexo!)
The potential outcome: This is not a binary play like some of our positions. So let’s say your earnings compound at 10%+; you get a dividend (which used to be a 3% yield, but now 6% because of the lower market price) and, if the price stays low, an opportunistic buyback by management. You’re looking at something like an 18% IRR with no re-rating; more if the multiple comes back or the founder sells
My stance: This is the kind of boring I like, and I will keep digging to decide if it’s worth a deep dive.
Description: Mensch und Maschine (Ticker: MUM) builds and sells engineering software in DACH (55%), the rest of Europe (32%), and internationally (13%). The business has two main parts. The Software segment is proprietary: hyperMILL, the CAM software that tells five-axis milling machines how to cut metal, at roughly 40% of company profits; SOFiSTiK, structural analysis for bridges and tunnels, at 15%; DATAflor, a Computer-Aided Engineering (CAE) product, for the rest. The Digitalization segment (VAR) customizes, implements, and trains. Half of its gross profit comes from reselling Autodesk licenses, and half from services built on top. The company was founded in 1984 and IPO’d in 1997. Founder Adi Drotleff still runs it and owns 48%.
Type: Hidden-Champion, dividend-paying, founder-controlled European microcap
Why it’s interesting: A high-quality compounder trading at half its own ten-year average multiple, for reasons that are mostly finished.
The rundown: What are you actually buying? MuM sells a family of engineering software, each product built for one narrow technical problem:
HyperMILL generates optimized CNC toolpaths for five-axis milling.
SOFiSTiK does structural analysis and reinforcement design for bridges, tunnels and buildings.
DATAflor runs planning, measurement, costing and invoicing for landscaping and groundworks.
eXs handles electrical circuit diagrams for projects running to thousands of sheets. Around these sit the digitalization toolboxes:
PDM Booster for product data
BIM Booster for construction
MAPedit for GIS and infrastructure
CustomX for variant configuration, plus the customization and training layer on top of Autodesk.
Every one of these products is a 2D or 3D model of physical reality, sold into a workflow where the output gets built, machined, or poured. None of them is generic enough to be bought off a shelf; all require deep industry knowledge to write and use, and all are customized to the individual customer before they produce anything. When your software plans the toolpath for a titanium aerospace part, or tells an engineer whether the bridge needs vibration dampers, a migration that goes wrong is not a bad quarter; it is physical damage. Switching costs in engineering software are of a different species than in normal enterprise software.
hyperMILL is the best example of this model. At about €30,000 per seat, it’s the most expensive CAM software on the market by a wide margin. It sells anyway, because it sits next to a milling machine that costs mid-six to low-seven figures and makes that machine 20 to 30% faster, in certain applications up to 10x. Gross margins from 90 to 95%. Group churn sits around 3%.
So why did the multiple compress from mid double digits to low double digits?
Three things happened.
Autodesk changed its billing model from resale to agent, which mechanically collapsed reported revenue while leaving gross profit untouched. It also pulled cash flow forward in 2023 and then unwound over the following three years.
On top of that came an ERP rollout, an efficiency program, and roughly €2.5M of non-recurring costs in 2025.
Underneath it all, Germany’s industrial economy has been flat on its back.
There is no demand problem.
The result: gross profit compounded at about 5% instead of the planned 7–10%, EPS at about 6% instead of 14–20%, and the multiple halved.
Most of that 3-part list is temporary and self-inflicted, not structural. The Autodesk working capital drag largely completes by Q3 FY26. The ERP is in. The cost base has been adjusted to a lower-growth setting. And the operating model is built for exactly this: 100 independently managed profit centers, compensation tied to gross profit growing 50% faster than OPEX, 20–25% of wages variable and around 50% for upper management.
The valuation math for this type of company is straightforward.
About 6% dividend yield, roughly 2% buyback yield, around 10% EBIT growth. So a 18% IRR with the multiple frozen exactly where it is. Management is funding the buybacks with debt at 2–4% to retire shares yielding 6–7%. Roughly half the dividend is paid in stock, which makes the buyback the better use of the euro.
This management missed guidance materially in 2024 (3.6% gross profit growth against 8–12% guided) and missed again on gross profit in 2025, hitting EPS only on an adjusted basis. Their 2030 target of €3.80 EPS assumes 8% gross profit growth and 15% EPS growth.
The other risk everyone will ask about is AI, and I think it’s mostly misplaced here. Only about 10% of the R&D headcount are generic developers; the other 90% are engineers, mathematicians, and physicists. Cheaper code doesn’t erode a moat made of 25 years of five-axis kinematics and jurisdiction-specific building codes. hyperMILL is licensed per machine, not per person, so productivity gains don’t compress seats. The risk is real at the low end of the services business: basic training, standard implementations, first-line support. That’s a small slice of a segment (c.10% of EBIT), not the company.
Current status: Interested. What tips this from “cheap German industrial” to something worth owning is the ownership structure. Drotleff is 72, holds 48%, and MUM sits in the Scale segment, where a delisting is straightforward. A take-private has been the consensus expectation for five years without happening; maybe not in the near term, and it would almost certainly happen at a higher multiple than today.
3. Best article of the week
Complexity Investing by Brinton Johns and Brad Slingerland
This week: Complexity Investing by Brinton Johns and Brad Slingerland (NZS Capital, 2014, updated 2021). A 46-page paper, free on their site.
The starting point is the same ergodicity problem I covered in our previous article.
Markets are complex adaptive systems, so outcomes follow power laws, not bell curves. Extreme events aren’t rare exceptions; they are the norm. Most economics, and most risk models, assume otherwise.
Their answer to that is a portfolio built on two things:
Resilience: companies optimized to survive and adapt rather than to squeeze out maximum short-term efficiency. Their example is ants: about half the colony sits idle at any time, which looks wasteful until a flash flood hits and the reserves matter.
Optionality: small positions with large asymmetric upside, structured like a VC fund. Most go to zero; a few pay for everything.
This is something I will do more and more in the future. But we did the inverse with our Nick Sleep portfolio, and with great success.
The rest of the framework:
Cut the middle. Anything that is neither resilient nor optional shouldn’t be in the portfolio. It won’t protect you, and it won’t pay you.
Size by how many predictions you need. The more specific the future a position requires, the smaller it should be. Big positions belong to companies where you only need current trends to continue. → This relates to how well you can predict the future.
Duration of growth beats rate of growth. The market chronically undervalues slow, long-duration compounding. Hyper-growth is fragile. This directly relates to what we wrote about the Competitive Advantage Period in the past…
Moats can be liabilities. In a world of free-flowing information, barriers turn into vulnerabilities. They prefer win-win economics over pricing power.
That’s it for this week
May the markets be with you, always!
Kevin