Projects
Side projects
Melvin/Projects/SFF Ventures

Three ventures in three months at Swiss Founders Fund Studio, all three closed

Three months in the studio's inaugural cohort, three ventures, and three problems validated with real buyers. All three are closed, and I can give the reason for each one.

Category
Software & AI · Ventures
Organization
Swiss Founders Fund Studio, Zurich
Status
Stopped
Dates
Jun – Sep 2026
3
ventures in 3 months
500+
applications, inaugural cohort
3
closed, reason on file
The landing pages of the three ventures side by side: Pullfacto over a night-lit Europe, Duops over a diagram linking ERP, CRM, calls, chat and spreadsheets, and REMVAL over a row of robot arms.
The three sites, June to September 2026. Left to right, Pullfacto, Duops, REMVAL.

Three months, three ventures

I was in the inaugural cohort of Swiss Founders Fund Studio in Zurich, picked from more than 500 applications. The format gives you three months to find a problem, prove it with real buyers and show traction, or stop. I ran three ventures in that time. All three problems were real and I validated them in the field. All three ventures are now closed, each for a reason I can name.

Pullfacto

Pullfacto was an AI copilot for investment promotion agencies. It reads expansion signals in job postings and public data, scores each company against the strengths of the territory, and drafts the first approach. Prospection officers told me they spend most of their week on exactly that research. It closed on market size. There are too few investment promotion agencies to support a venture-scale company.

The promise, in one line

Pullfacto's landing page: the headline "Find them before they choose", a short description of the product and a "Request a demo" button, over a photograph of Europe lit up at night from orbit.
The pitch was about timing. An investment promotion agency usually hears about a company's site decision after the shortlist is already closed.

REMVAL

Used industrial robots are sold on photos and the seller's word. The expensive failure is the one you cannot see. Reducer wear costs thousands to repair and does not show in a power-on demo. REMVAL was a standardised, signed condition report, read from the robot's controller on site. I built the site and confirmed the problem with both buyers and sellers. It closed on the market analysis. The second-hand robot arm market is small, and the willingness to pay per report was too low for that market size. The work was not lost. AIH Global, my own venture outside the studio and now paused, came directly out of it.

The failures a demo does not show

REMVAL's landing page: rotating headlines reading "A repair bill nobody mentioned", "Undisclosed crash history" and "Don't buy a robot blindly", a search field for a robot model, and a row of robot arms marked available or in development.
Search for your robot model and get the test protocol. The headline cycles through the problems a power-on demo does not show.

Duops

Companies rarely fail to automate because they lack tools. They fail because nobody knows where the time goes. Duops connected read-only to the ERP and the other tools already in place, then an agent interviewed the teams to capture what was never written down, like the workaround that keeps orders moving or the real lead time behind the official one. The output was one map of how the company runs, with the official process next to the real one. From that map, leadership picks what to automate first, and we build the agents that do the work. It closed on go-to-market. Duops needs a company big enough to run an ERP, and those ERPs, SAP and the others, do not give third parties access. Each connector is a large build before anything is proven. Winning a pilot inside a company that size, with no prior network and no track record, takes far longer than three months.

Read-only first, agents after

Duops' landing page: the headline "See how your company actually runs. Then decide.", and below it a diagram with the Duops mark at the centre, linked out to ERP, CRM, calls, chat and spreadsheets.
The map comes before the automation. Every system is read, nothing is changed.

What it taught me

The fastest way to learn a market is to try to sell into it. Two of the three ventures had a real problem with a market too small underneath it. The third had customers I could not reach in the time I had. I only learned that by putting a price and a pilot proposal in front of real buyers. No amount of desk research would have told me.