Most of what makes a product fail happens between the parts people buy separately. We run the whole distance because we run it on ourselves: eleven products, one shared production system, and the same studio responsible after launch as before it.
What people arrive with
Each hands over to the next, and each handover loses the reasoning behind a decision. The next supplier rebuilds that thinking from the artefact, which is how a product ends up with a design nobody can explain.
Without one, scope is settled by whoever is most insistent, and the first release is both late and too large to learn anything from. Choosing what the first version is not is the most valuable hour in the project.
A launch is the beginning of the part that matters: the listing, the first reviews, the crash in a locale nobody tested, the second release. A supplier whose contract ends at launch is not incentivised to care about any of it.
What the work includes
Scope is where a project is decided, so this list is specific rather than reassuring.
What it is, who it is for, what the first version deliberately leaves out, and what would have to be true for it to be worth building. Written down, so that the version six months later can be checked against it.
Each product gets to be itself. A cozy puzzle game and a sprint-analysis tool should not look like siblings, and in our own catalogue they do not - the studio shell is deliberately neutral so each product can bring its own world.
The hardest screen first, playable, before the easy ones are built. A prototype that avoids the risky part is a slide deck with animation on it.
Mobile, web, and the services behind them, with the shared parts built once. A large share of the implementation here is written by AI coding agents working under review against tests that have to pass - stated because it is true, and because it is why this pace is possible at this size.
Automated gates, a documented device and locale matrix, then a person. Store submission with the listing, screenshots and privacy answers derived from the shipped build rather than from a plan.
Release and performance signals return to where the work was made, so the second version is informed rather than guessed. The fix lands in the shared system, not in one product’s copy of it.
How it runs
Not a process diagram written for this page. It is the production loop the studio already runs, which is why it has a measure step and a human gate in it.
Evidence
Everything shown as evidence on these pages is a NOVEMADE product - our idea, our money, our name on the listing. There is no client logo wall here, no testimonials and no case-study metrics, because there have been no clients to take them from. What there is instead is software you can open.
A one-thumb arcade game about never standing still — 600 levels, all offline.
1002 early-learning activities for two- to six-year-olds, voiced in 29 languages.
Design a tattoo, turn it into a working stencil, and see it on your own skin first.
Check it yourself
Every word you hear is a file from Pivumo’s shipped voice package — the same recording, the same encode, at the same loudness the app plays it at. Six of the 29 locales are here, and the age toggle really does swap the take. The round itself is written for the web; the voice is not.
Where this stops
We take on a small number of these at a time, and we say no to the ones where this studio is the wrong shape for the work. We also do not run a product we hold no agreement in indefinitely - ownership, and who holds the keys, is settled at the start and in writing.
Also from the studio
Collaborate
Publishing, partnerships, investment, or a technical conversation about how we build. Tell us which one and we will skip the rest of the form.
