Enter your market and the routes you are weighing. Each route is projected month by month, scored 0–100 on Velocity, Margin and Robustness, and stress-tested. The page opens with a worked example: a B2B software company entering a new European country.
Channel take is the reseller margin or partner revenue share, as a share of what customers pay. Concentration risk is a 0–100 judgment rating; partner dependency is the share of revenue that runs through the largest intermediary.
Each metric scores 0 at worst and 100 at best, capped at both ends. Metric weights are averaged within their dimension.
The tornado moves one input at a time. This check draws 2,000 scenarios with every input moved at once (±25%), labels each with the route the VMR model picks, and trains a logistic regression and a random forest to predict it. The models learn this tool's logic, not real market data: they show how robust the recommendation is and which assumptions decide it.
The machine-learning check trains in your browser: a softmax logistic regression on standardised inputs (L2, C = 1) and a 200-tree random forest (Gini splits over 64 quantile bins, √d inputs per split, at least 5 scenarios per leaf), with permutation importance on held-out scenarios. Method: revenue ramps on an S-curve from each route's first-revenue month to steady state (market size × target share × reach). Contribution is revenue × (gross margin − channel take); cash flow subtracts CAC on new and replacement customers, fixed costs and set-up. Breakeven past the horizon is projected at the final month's cash flow. Scores use fixed anchors, so one route's score never moves because another route changed. Same model as the open-source Python tool; the example figures are synthetic.