Forecast Governance
Every rep's forecast accuracy, scored from their commits against the actuals that landed, so you can weight the roll-up by who is reliable.
Included with leadercron . A seat without it sees an upsell here instead of the page.
Overview
Forecast Governance gives a leader the one thing a roll-up forecast usually lacks: a track record. It scores each rep's forecast accuracy from the commits they submitted against the actuals that landed, so you know whose number to trust, whom to weight down, and whom to coach. It is metadata only: commit and actual figures per rep, never their content.
Bias and error
Two numbers per rep, both deterministic. Bias is the signed average of (actual minus commit) over commit: positive means the rep sandbags (beats the commit), negative means they over-forecast (miss it). The error rate is the mean absolute percentage error, how far off they are in either direction. A rep can have low bias but high error (wild but unbiased), or high bias with low error (consistently off by the same margin). Both matter, so both are shown.
Reliability bands
Each scored rep gets a reliability band from their error rate: tight (under 10%), moderate (under 25%), or loose. Reps are ranked most reliable first. The team tiles roll the same math across everyone: a team error rate and a team bias, so you can see the whole board's forecast discipline at a glance. Only closed submissions (where the actual has landed) count; an open commit is pending, not scored, so the accuracy is always evidenced.
Recording submissions
A commit is a number a rep says out loud on a forecast call, so there is no CRM field to import it from. Record them on the page instead: one line per rep and period, comma separated, as rep, period, committed, actual. Leave the actual off while the period is still open and add it when the period closes. Save & score stores the lines against your account and returns the scorecard in the same step; Show stored re-scores everything you have recorded so far without touching the box.
Rep plus period is the key, so recording the same rep and the same period again replaces that entry rather than stacking a second commit for the same quarter. That is what lets you fill in an actual later: paste the same line with the actual on the end. A line missing the rep or the period is skipped, and a rep with an open period is kept and simply not scored yet.
Where the data comes from
Governance scores the forecast submissions stored for your team, each a rep, a period, a committed number, and the actual once the period closes. A rep with no closed submission is not scored rather than guessed. As more periods close, the scorecard sharpens on its own. Turning a weak track record into a coaching plan stays with the leadercron coaching skills.
What the model says about the open pipeline
The second card on the panel is the other half of the same judgement. Governance asks how reliable the person is; this asks how reliable the deal is, from a regression trained on the deals your own team has already closed. It lists the committed deals the model lands after their committed date, worst slip first, with the value of what is late.
It is an estimate with an error band, not a date. Each row shows the predicted close, a plus-or-minus in days from the model's own residual spread, and the committed date it is being compared against. Treat a deal predicted three weeks late as a conversation to have, not as a number to re-forecast to.
Where it comes from. A daily job records a small set of activity features for each open deal: touches in the last 30 days, days since the last one, replies, meetings, days in the current stage, and how long you have worked the account. A weekly job fits those against how long your closed deals actually took. Nothing about deal content is read, and the model is yours alone: it is trained per tenant, never across customers.
When the card says nothing yet. Three different reasons, and it names the one that applies. No model has trained because there are not enough closed deals yet, and the daily job is already collecting toward it. A model exists but explains too little of your variance to be worth a date, in which case it retrains weekly and turns itself on when it earns it. Or there are no open deals with a recent snapshot to predict against.
Each row says what is pushing it. Up to three drivers, named in plain language, with whether each one is pushing the close sooner or later. A row predicted from a snapshot more than a day old says how old it is rather than presenting stale features as current.