Two of your predictors have a correlation of 0.95. Does that matter, and what do you do about it?

It depends on your goal. For pure prediction, collinearity barely hurts accuracy; the model just splits weight between the twins. For inference it is serious: coefficients become unstable and their standard errors inflate, so signs can flip between samples and you cannot say which variable drives the outcome. Check variance inflation factors, then drop one, combine them, or use ridge.

My first question is what the model is for. If we are scoring leads and nobody reads the coefficients, a 0.95 correlation is mostly harmless. The fit is fine, the predictions are fine, the weight just gets split between the two variables in some arbitrary way. Where it bites is when someone plans to act on the coefficients. Then it is a real problem, because that split is unstable: refit on a slightly different sample and one coefficient swings positive while the other swings negative, and the standard errors are wide enough that neither looks significant even though the pair clearly matters. I check VIF, and anything over about ten I treat as suspect. The fix depends on the pair. If they are the same thing measured twice, like revenue and revenue in local currency, I drop one. If they are genuinely two facets of something, I combine them into a ratio or a single component. If I need both and only care about prediction quality, ridge handles it cleanly by sharing weight in a stable way instead of an arbitrary one.