Skip to main content

Co-occurrence Warnings

You may notice an amber triangle icon next to certain rules on your scoring rule list pages. That's a co-occurrence warning. It tells you that two (or more) rules are firing on largely the same set of leads, which can throw off calibration's lift estimate. This page explains what the warning means, when to act on it, and how to fix it.

What the Warning Means​

In plain language: two rules that fire on the same leads can't both claim full credit for those leads converting. When kenbun calculates lift for a rule, it compares "leads who triggered this rule" to "all leads". But if 80% of the leads who trigger Rule A also trigger Rule B, then the lift for Rule A is partly being driven by Rule B's signal — and vice versa.

This is called confounding. The amber triangle warns you that a rule's calibration suggestion may be inflated because the rule overlaps heavily with another rule. The threshold is 70% lead overlap.

Why It Matters​

If you apply a confounded suggestion without thinking, you can end up over-rewarding both rules — effectively double-counting the same signal. Your scores get noisier, hot leads get even hotter for the wrong reasons, and your calibration data slowly degrades.

When to Act on the Warning​

A co-occurrence warning is a flag, not a stop sign. It says "look closer before applying." Here's how to think about it:

  • Two rules that mean essentially the same thing: consolidate. Pick the more meaningful one and delete the other. For example, if "Visited Pricing Page" and "Clicked Pricing Plan Card" almost always fire together, you don't need both.

  • Two rules that genuinely measure different things but happen to overlap: keep both, but apply suggestions cautiously. Maybe halve the suggested change, or only apply the higher-confidence one.

  • Two rules where one is a strict superset of the other (e.g., "Demo Booked" implies "Demo Request Submitted"): keep the more specific one with higher weight, lower the broader one's weight, or remove it.

  • Two rules in different dimensions (e.g., a profile rule and an engagement rule that both correlate with the same buyer persona): this is usually fine. Cross-dimensional confounding is expected and less harmful — the dimensions are scored separately.

Common Patterns​

A few patterns we see often:

Pricing-page family "Viewed Pricing Page", "Clicked Pricing Card", "Viewed Pricing FAQ" all fire on the same in-market shoppers. Consolidate to one or two rules.

Demo-related cluster "Demo Request Form Submit", "Demo Confirmation Page View", "Demo Request Email Open" all fire around the same demo-booking moment. Pick the most actionable signal.

Form-submit overlap "Form Submit" (a generic event) and "Contact Sales Form Submit" (a specific one) often overlap because the specific one fires through the generic. Decide which level of specificity you want to score.

Account-level twins Two account rules that both rely on company size — e.g., "Mid-market Account" and "100-500 Employees" — will naturally overlap. Pick one definition and stick with it.

How to Fix It​

There are three reasonable responses to a co-occurrence warning:

  1. Consolidate: delete the redundant rule and keep the more meaningful one.
  2. Reduce weight on one: if both rules are useful, reduce the weight of the less-specific one so you're not double-counting.
  3. Accept it and move on: if the rules genuinely measure different things and you understand the overlap, you can leave them alone — just apply calibration suggestions for confounded rules with extra skepticism.

To see what other rule(s) a rule is confounded with, hover over the amber triangle. A tooltip will show the overlap percentage and the conflicting rule name.

What This Warning Is Not​

A few things the warning is not telling you:

  • Not a bug. The rules are firing correctly; they just overlap.
  • Not a data quality issue. Your events are fine.
  • Not a hard error. Calibration still produces a suggestion — it just flags that you should consider the suggestion in context.
  • Not a scoring problem. Your scores still compute correctly. The warning is purely about calibration accuracy.

Acting Cautiously on Confounded Suggestions​

When you go to apply a suggestion for a rule with the amber triangle:

  1. Check what it's confounded with. Hover the triangle in the rule list, or check the calibration panel inside the rule slideout.
  2. Decide whether the overlap is intentional. If both rules are needed, you accept some calibration noise. If one is redundant, fix the rule list before applying weights.
  3. Apply more conservative changes. If the suggestion says go from 5 to 25, consider 5 to 15 instead. The true lift is probably less dramatic than the raw number suggests.
  4. Re-calibrate after consolidating. If you delete or merge rules, the next calibration refresh will recompute lift cleanly. Old suggestions become stale.