Define what revenue impact means in your organization
Annual recurring revenue attached to an affected account is not the same as revenue at risk. The customer may have a workaround, the problem may affect one user, or renewal may be years away. Decide whether your model uses affected ARR, expansion influence, renewal risk, contraction risk, or a separately reviewed commercial assessment.
Keep each measure distinct. If account value is available but risk is unknown, label it affected ARR. Do not rename it revenue at risk simply because the larger number attracts attention.
Build the score from inspectable inputs
Start with a consistent account identifier so feedback can join to the correct commercial record. Define the revenue date, currency treatment, account hierarchy, and how duplicated reports from the same account affect frequency.
Then add the non-revenue factors that protect the model from commercial tunnel vision. Severity, number of affected users, product breadth, security, accessibility, contractual obligations, strategic fit, evidence confidence, and cost of delay can all matter.
- Confirmed affected accounts and a separate suspected-account list.
- Revenue values with source, effective date, and currency normalization.
- Severity based on customer outcome, not account size.
- Frequency counted by independent occurrence and affected account.
- Confidence based on evidence quality and completeness.
Avoid double-counting the same commercial signal
Account tier, ARR, renewal risk, and executive escalation often correlate. Adding full weight for each can multiply one commercial fact several times. Map dependencies between factors and choose either a primary measure or capped contributions.
The same caution applies to ticket volume. Ten follow-ups in one incident do not equal ten independently affected customers. Deduplicate conversations at the account and problem level while preserving total interaction burden as a separate service-cost measure.
Review rankings with the underlying evidence visible
A score is a compact explanation, not a substitute for one. Reviewers should see which factors drove the ranking, which values are estimates, and how the order changes when uncertain inputs are removed.
Calibrate the model against actual decisions and outcomes. If severe broad-impact problems repeatedly rank below narrow high-ARR requests, adjust caps or weights. Preserve the change history so old decisions remain understandable under the policy that existed at the time.
- Run sensitivity checks on uncertain revenue and frequency values.
- Use separate queues for security, legal, safety, and critical reliability obligations when appropriate.
- Review stale scores after account, renewal, severity, or product evidence changes.
- Document exceptions and the manager who approved them.