Prove it earns its keep, or kill it
A model you do not measure is a model you are trusting on faith, and faith is not a growth strategy. The discipline that separates a working model from an expensive decoration is the willingness to put a number on it every month and to kill it if that number is bad. The point table earns nothing; the lift it produces is the only thing that does.
Measure lift against an unscored baseline, monthly
Every month, compute conversion lift: how the leads your model ranked highest actually convert, against an unscored baseline cohort. A working model lands between three and ten times the baseline. At one and a half times or lower, the model is barely separating signal from noise, and a model that does not separate is not a model, it is a ritual. This single number is your verdict. It does not care how thoughtful your weightings were or how clean your dashboard looks. It tells you whether the bet is paying, and if it is not, you change the model or you stop pretending it works.
Recalibrate on a trigger, not a calendar
The instinct to review the model quarterly because the quarter ended is a tidy habit that ignores how decay actually arrives. Do not recalibrate on the calendar, recalibrate when the signal moves. The trigger to watch is rejection rate: when sales acceptance shifts by ten or more percentage points, the model has drifted from the reality it was tuned on and needs attention now, not at the next scheduled review. A calendar review can be too late by months or wasted effort on a model that is still fine. The trigger fires exactly when the work is warranted.
Treat decay as a certainty, not a risk
Most B2B scoring models silently stop predicting conversion within about six months of launch. Not because they were built badly, but because the world they were trained on keeps moving: your ICP sharpens, your channel mix shifts, your buyers change how they evaluate. A model is a snapshot of a moment, and the moment passes. This means the setup is never finished. Decay activity scores over a thirty-to-one-hundred-and-eighty-day window so old behaviour fades, refresh firmographics through enrichment so the fit data stays current, and accept that a living model is the only kind worth running.
Close the feedback loop in a week
The model only learns if you tell it what actually closed, and you have to tell it fast. Close the sales-feedback loop within a week, feeding back which scored leads became real opportunities and which fizzled, so the model is updated against recent truth rather than last quarter's. A loop that takes a month to close is teaching the model about a buyer who has already changed. Tighten it to a week and the model keeps its grip on a moving target. Measure the lift, watch the rejection trigger, refresh the inputs, and close the loop fast, and the model stays an asset. Skip any of these and it quietly rots into a machine routing your time to the wrong people.