From a brilliant model trapped in notebooks to a product enterprises buy.
Two PhDs had a forecasting model that outperformed everything on the market — and no way to sell it. We turned research code into an enterprise-grade platform with the security, reliability and onboarding that procurement teams demand.
Challenge
Northstar's founders could predict demand better than anyone in their category, but their 'product' was a collection of Python notebooks run by hand for each client. Every new customer meant weeks of bespoke setup. Enterprise buyers loved the results and failed them on security review. The science was world-class; the product didn't exist yet.
Approach
We separated what was genuinely differentiated — the models — from everything that wasn't: ingestion, auth, tenancy, reporting. The founders kept full ownership of the science while we productized everything around it, designing for the procurement checklist as deliberately as for the end user.
Solution
A multi-tenant analytics platform where clients connect data sources in an afternoon, models retrain on schedule, and forecasts arrive in dashboards and APIs their teams already use. The founders' notebooks became versioned, monitored model services — same science, now with an SLA.
Outcome
“We knew the math. Gilva knew everything else — and never once made us feel like the science was the easy part.”