Istari is founded by three partners. The people you scope with are the people who build, ship and maintain the work — no subcontractors, no handoffs. Here is who we are.
Julian has over ten years in data analytics, five years in data engineering, and has led a data-engineering team for the past two years. He builds the pipelines and models the rest of the studio's work runs on.
His cost-optimisation work on data tooling saved one company more than €750,000 a year — the kind of quiet, compounding result the studio is built to find.
Edison is a technical leader focused on cloud and cloud security. He works in Microsoft Azure with a cybersecurity lens, modernising and securing enterprise IT.
Outside the studio he teaches in post-secondary programmes, equipping students with the technical skills and confidence to enter the Canadian workforce — closing the technical-debt gap one cohort at a time.
Nathan is a product leader who makes complex systems simple — especially in healthcare. Over the past decade he has shipped products across aging tech, consumer electronics and field operations.
He has helped nurses triage patients faster, and helped Coca-Cola cut more than $1M in retail waste with better demand forecasting.
Why public institutions rarely need more data — and almost always need less distance between a question and an answer they can defend.
Most of the data a ministry needs already exists. It accumulates inside administrative systems as a by-product of delivering services — claims, permits, wait-lists, inspections. The constraint is rarely collection. It is the weeks it takes to turn that record into an estimate a decision-maker can stand behind, and the trust that estimate must earn on the way. This paper sets out the studio’s position: smaller models, uncertainty kept in view, and findings written so a non-specialist can act on them the same afternoon.
Public decisions are paced by their reporting calendar, not by the questions in front of them. A figure compiled quarterly describes a cohort that has already moved on by the time it is read.
The delay is not neutral. Every week a service runs on a stale estimate is a week of effort pointed slightly wrong — staff deployed against last quarter’s demand, budgets defended with numbers no one quite trusts anymore.
Administrative data is the exhaust of public service: collected once to run a programme, then left at rest. Re-reading it answers most operational questions without a single new survey.
The work is connecting the record to the question — and closing the loop so an answer feeds the next decision rather than a binder on a shelf. The shorter that loop, the faster an institution learns.
We would rather hand a minister a range we can defend than a single number we can’t.
A lone point estimate hides the one thing a decision-maker most needs: how sure we are. We ship the interval, not just the midpoint.
A wide band is not a failure. It is an honest instruction — gather more before committing, or commit knowing the odds. The posterior widens with the horizon, and we show exactly where confidence runs out.
A model a person can follow is a model a public body can defend, audit and own. We don't ship what we can't explain in a meeting.
The range ships beside the estimate, every time. Confidence is never overstated, and a wide band is treated as an instruction, not an embarrassment.
If the finding can't be read by the person acting on it, it isn't finished. The method note is part of the deliverable, not an appendix.
