Risk Forecasting
Simulate your business initiatives before you commit.
We simulate your proposed marketing or investment decision against your historical performance data and derive an expected value for the high stakes decision you haven't made yet.
Three steps, one afternoon
01
Upload
Drop a .csv, .parquet or .npz up to 256 MiB. Files go straight to encrypted object storage over a single-use URL — they never touch our web tier.
02
Name the target
Tell us which column to complete and which columns condition it. We fit an interpolant path between the observed marginal and the full joint on GPU workers.
03
Draw samples
Get the completed table back, plus per-cell uncertainty so you know which imputations to trust. Re-sample as many draws as you need.
Survives extreme sparsity
Classical imputation degrades badly past roughly half-missing — it collapses toward the mean and quietly destroys the tails. An interpolant learns the transport between distributions, so the completed column keeps its shape at 90% missing.
Correlations stay intact
Because we model the joint rather than each column alone, downstream models trained on the completed table see the same dependency structure as the real thing. That's the part per-column generators get wrong.
Plans
Free to set up. Pay when it runs.
Create an account and you can upload data, configure a scoring run and explore the whole product without paying. The Risk Scoring plan is what puts those runs on our compute, with a monthly allowance of scoring runs and scheduled re-scoring as your data drifts.
A scoring run is one dataset against one target column. Allowances reset on the 1st and don't roll over. Datasets are capped at 1M rows, and the target column must be at least 10% populated.
Bring the exposure you can't measure.
We'll benchmark the scoring against a holdout so you can see exactly where it beats what you're using today — and where it doesn't.
Score a dataset