Every lead scoring model has a half-life. The exact number varies, but somewhere between 4 and 9 months after training, the accuracy drops below the threshold where the scores are actually useful. The sales team starts to say "the scores do not mean anything anymore" and they are right, even if they cannot articulate why.
The cause is drift. Buyer behavior shifts. Your product shifts. Your ICP shifts. The features that predicted a close six months ago no longer predict it today. The model still runs, the scores still populate, but the correlation to outcomes has quietly fallen off a cliff.
We covered this in how scoring models actually drift and why you should care. The short version is that drift is inevitable and invisible unless you measure it.
Measuring drift is not complicated. You compare predicted outcomes to actual outcomes on a rolling window. If the model says 90 and the lead closes, that is a hit. If it says 90 and the lead ghosts, that is a miss. Plot hit rate over the last 500 leads and watch the trendline. If it is stable, the model is fine. If it is declining, retrain.
The reason most teams do not do this is ownership. Scoring models get installed once, nobody owns the accuracy, and the slow degradation is visible only to whoever would have owned it if somebody had. Two years later the whole system has become ornamental. We have walked into sales orgs where the lead score field exists on every record but nobody on the team trusts it enough to use it.
The fix is boring and mandatory: monthly retraining on recent data, weekly accuracy monitoring, and alerts when drift exceeds a threshold. That is it. The tech is not hard. The discipline is. We build lead qualifier systems with this scaffolding on by default because asking clients to wire it up later never works.
There is a bigger lesson here that applies to every ML system in a business. A model is not an artifact. It is infrastructure. Infrastructure needs maintenance. If you do not assign an owner to the maintenance, the system rots. Build it in on day one or accept that you are on the two-year path to "this thing stopped working."