📊 Sales Forecasting with Machine Learning

About this simulation

Every gem drifting through the tunnel is a deal moving through five pipeline stages — Prospecting, Qualified, Proposal, Negotiation, Closed. Its color encodes a win-probability score, the output a gradient-boosted classifier (like XGBoost) would assign from features such as stage, engagement and deal quality: red is unlikely to close, green is likely. Behind the tunnel, a bank of purple bars is the Monte Carlo forecast — thousands of simulated "what if each deal closed or didn't, weighted by its probability" trials, binned into a distribution of possible total revenue outcomes.

How it works

• Win probability is modeled as a logistic function of pipeline progress and a per-deal "quality" factor, exactly like a calibrated classifier's predict_proba.
• Priority score multiplies win probability by deal value — the same logic sales-ops teams use to rank which deals a rep should work today.
• Monte Carlo simulation draws a Bernoulli outcome per deal per trial (win with probability p, at its full value) and sums revenue, revealing the shape of forecast uncertainty a single point estimate hides.

Controls

• Pipeline volume — how many open deals are in flight at once.
• Avg deal size — scales the dollar value driving expected revenue.
• Model confidence — how sharply the win-probability model separates good deals from bad (steepness of the logistic curve).
• Priority threshold — deals scoring below this priority cut-off dim to grey, like a CRM's "focus list" filter.
• 🎲 Run Monte Carlo — resample the revenue forecast distribution from the current pipeline snapshot.

Did you know?

Point forecasts ("we'll close $2.1M this quarter") hide a huge range of plausible outcomes. Monte Carlo forecasting instead reports a distribution — e.g. "$1.6M–$2.7M, 80% confidence" — which is far more honest, and far more useful for a finance team deciding how much to bank on.