All use cases
Use case · Decision & risk models

Decision support

A forecast is useful when it changes what you do. We build systems that turn forecasts and their ranges into thresholds, alerts, allocations and plans your team can act on and defend afterwards. Each rule reads a range checked on data the model has not seen.

p10/p50/p90on every output a decision rule reads
77.2%vnappraise band coverage, against a nominal 80%
14 / 17vnpredict forecasts that beat the random walk

When you need this

Your dashboards report; nobody can say what to do. Capital, inventory, hedging, and staffing decisions ride on point estimates that hide the one thing that matters: how wrong they could be. Alert systems either cry wolf until they’re ignored or stay silent until it’s too late. And when a call goes wrong, there’s no trail showing it was the right decision given what was known.

What we build

  • Clear rules for action: act when the p10 clears your threshold, and hold when the band straddles it. Each rule is defined on a tested range.
  • Alerts people believe: early-warning thresholds set to a chosen false-alarm rate, so alerts stay rare enough to be trusted.
  • Plans that account for risk: portfolio, capacity and inventory optimisation that uses the full range of outcomes.
  • Stress tests on demand: Monte Carlo scenarios for tail risk, stress paths and the what-ifs a single forecast can’t answer.
  • A record of every call, scored on results: each recommendation logs its inputs, its range, its rule and then its outcome, so the rule is tuned on results and can be audited after the fact.

How it’s built

Calibrated predictions → decision rules · optimisation · Monte Carlo → recommendations · alerts · plans → logged & monitored

Methods: classical quantitative models · time-series foundation models

In our products

vnpredict forecasts 17 of the 28 Vietnamese macro indicators it tracks, each with an 80% band and a grade against a random walk, and refreshes its web-searched commentary on a 14-day cycle. vnappraise, a research preview, gives each property estimate a p10 to p90 band and publishes how often asking prices land inside it: 77.2% against a nominal 80%, measured out of sample.