Health Score in Renewal Forecasting

A health score becomes a forecasting input the moment its bands are attached to renewal rates you have actually observed. Until then it is a triage tool. The conversion is straightforward, and the discipline it requires is calibration rather than modeling.

Building the mapping

Take a snapshot of every account's health score as it stood 90 days before its renewal date, going back at least eight quarters. Bucket those accounts by score and calculate what share of each bucket renewed, and at what value. The finished table has one row per band and looks like this, with your own numbers in the last three columns:

The first time a team builds this, the bands rarely behave the way anyone assumed. Two adjacent bands with nearly identical renewal rates mean the score is not separating risk in that range, and the fix is a better signal rather than a redrawn boundary.

Apply the calibrated rates to the renewal base by band and the result is a probability-weighted renewal forecast built on behavior instead of on account manager sentiment.

Track value, not only logo

Renewal outcomes are not binary. An account that renews at 70% of its previous ARR counts as a save and as a miss at the same time. Run the calibration twice, once on logo retention and once on retained value, because a health score that predicts survival while missing contraction produces a forecast that is right about the customer list and wrong about the number.

Keep it calibrated

Two ORM positions govern the maintenance of this model.

The first concerns drift. ORM's view is that the most common reason a forecast fails is that something in the business or the market changed while the forecast was still built on old assumptions, and that a model unresponsive to changing dynamics will miss. Score-to-probability mappings age exactly this way. Recalibrate every two quarters.

The second concerns data quality, which is the objection most teams raise before they start. ORM's position is that everyone thinks their data is uniquely bad, that everyone is wrong, and that consistent data supports accurate predictions regardless of how messy it looks. Consistency is the requirement. Perfection is not.

One scope note. ORM's 95% accuracy target applies to new and expansion business rather than renewal. Renewal is a separate forecasting problem running on a different signal set, which is precisely why it deserves its own calibrated model instead of a haircut applied to the sales forecast. Done properly, the renewal line stops being the softest part of the plan and starts carrying real forecast accuracy, which is what makes net revenue retention something a board can be told in advance rather than after the fact.