This system is running live
Don't take
our word for it.
Try it.
We rebuilt a miniature version of this engagement and it's running right now. Click below — you are not looking at a screenshot, you are looking at the actual software.
▶ Open the live demo →SaaS · Client work · 6 weeks
Sales Forecasting Ensemble
Quarterly revenue predictions for a SaaS startup
- forecast error
- 22% → 8%
- backtest window
- 12 months
- used in Series A
- yes
The problem
A SaaS startup was forecasting revenue with a single-input linear regression on the previous quarter's pipeline. The forecast was consistently 15-25% wrong in either direction, which made resource planning unreliable and made investor conversations harder than they needed to be.
What we built
An ensemble of three models:
1. A gradient-boosted regressor on pipeline features (deal stage, age, value, owner historical close rate) 2. A SARIMA model on marketing-attributed lead volume by channel 3. A simple seasonal-decomposition baseline as a sanity check
The three are combined with weights tuned on a 12-month rolling backtest. The final output is a 50/80/95 confidence band, not a single number — because honest forecasts have uncertainty and pretending they don't damages decision-making.
Outcome
- Mean absolute percentage error dropped from ~22% to ~8% on the next quarter's actuals
- The forecasts were used as supporting material in a successful Series A raise
- The CFO stopped making weekly "is this number right?" calls to the data team
Tech stack
Python, statsmodels, LightGBM, dbt for the data prep pipeline, a Looker dashboard for the executive view.