We got tired of building things that never shipped.
SMZone exists because too many good models die between the notebook and production. We work the other way round: start at the decision, go backwards to the data, and build only what the decision needs.

01How we work
Five
Five things we hold to.
The decision comes first
Before any modelling, we write down the decision, who owns it, how often it happens and what a wrong answer costs. If we cannot fill that in, there is nothing to build yet.
Smallest thing that beats the baseline
The baseline is your current process, not a held-out test set. Gradient-boosted trees beat transformers more often than anyone selling transformers will admit.
Boring infrastructure
We use the dull, well-understood parts of your stack. A system a new hire can read at 3am is worth more than one that scores two points higher.
We say no in writing
If the data cannot support the ask, that goes in a document with our reasoning, and the engagement stops. We would rather lose the build than deliver something that quietly fails.
Handover is not an afterthought
The runbook is written during the build, not at the end. If your team cannot operate it without us, we have not finished.
02What we refuse
The work we turn down.
Naming this up front saves everybody a call. We do not take on work where the output cannot be tied to a decision, where nobody on the client side is accountable for that decision, or where the honest recommendation would be to fix the data collection first and come back later. We also do not build systems whose purpose is to remove a team without telling them.

03The company
Plainly.
- What we are
- A small engineering team building decision systems for operational businesses.
- Where we work
- Remote, across European time zones, on your infrastructure.
- Sectors
- Logistics, manufacturing, insurance and field service — anywhere a decision repeats.
- Engagement length
- Two weeks to twelve, then an annual review.
- What we will not do
- Operate a system indefinitely so that you cannot leave.
PLACEHOLDER — confirm every line here against the real company before launch. See PRODUCT.md.
Tell us about one decision.
One decision your team makes often, and what it costs when it goes wrong.
Start a pilot