Applying Suggested Weights
Once calibration has data to work with, you'll see suggested weights next to your scoring rules. This page covers how to review them in bulk, what the audit-trail rationale tells you, and how to apply changes safely.
Why Bulk Review
You could review suggestions one rule at a time inside each rule's slideout — and that works fine for occasional tweaks. But if you have dozens of rules across engagement, profile, account, and deal dimensions, going one-by-one is tedious. The Calibration Review page gives you a single sortable table of every calibrated rule, with bulk-select and a single Apply confirm dialog.
Most teams use the bulk page for monthly or quarterly tune-ups, and the inline panel for one-off changes when they're already editing a specific rule.
Navigating to the Bulk Review
Go to Configure > Scoring > Calibration Review. You'll see a table with one row per calibrated rule.
Reading the Table
Each row shows:
| Column | What it means |
|---|---|
| Rule | The rule name and its dimension (engagement, profile, account, deal) |
| Current | The weight you've set today |
| Suggested | What calibration recommends, based on observed lift |
| Δ (Delta) | The size of the change. Positive means up, negative means down. Color-coded for quick scanning |
| Confidence | How trustworthy the suggestion is, given sample size |
| Sample | How many leads the suggestion is based on |
| Last refresh | When this suggestion was last computed |
How to Read the Delta Column
The delta column is your fastest way to find high-impact changes. Sort by absolute delta, descending, and you'll see the biggest weight changes at the top.
A large positive delta (e.g., current 5, suggested 25) means: "this rule predicts conversion much better than your current weight reflects — give it more weight." A large negative delta (e.g., current 30, suggested 8) means: "this rule isn't pulling its weight — you've been over-rewarding it."
Small deltas (under 2-3 points) are usually fine to leave alone. The big signal is in the outliers.
Filtering and Sorting
Above the table you'll find:
- Dimension filter: show only engagement, profile, account, or deal rules
- Confidence filter: hide low-confidence suggestions if you only want to act on solid data
- Staleness filter: surface rules whose suggestion hasn't refreshed recently
- Search: find rules by name
Sorting by delta descending is the most common workflow. Filter to "high confidence only" and sort by delta to find the safest, highest-impact changes first.
Selecting and Applying
Each row has a checkbox. Select one, several, or all rows; then click Apply selected. A confirmation dialog appears that summarizes what you're about to change.
You can also apply a single row inline by clicking its row-level Apply button. The same confirm dialog appears, just for one rule.
Reading the Apply Confirm Dialog
Before any weight changes, kenbun shows you a confirmation dialog with the full rationale:
- Conversion event in use (e.g., "Closed-won deal", "Engagement persistence")
- Lookback window the calibration was computed over
- Sample size: how many leads triggered the rule in that window
- Conversions: how many of those leads converted
- Baseline conversion rate: how often a typical lead converts in your account
- Observed lift: the rule's conversion rate divided by baseline
- Confidence: high / medium / low
- Current weight → Suggested weight: the change you're about to make
This is your last checkpoint. Read it. If anything looks off — sample is tiny, lift seems too good to be true, confidence is low — cancel and dig in. If it all looks reasonable, click Apply.
What Gets Logged
Every Apply is recorded in the rule's audit log. The log entry includes:
- Who applied it (your user account)
- When it was applied
- The previous and new weight
- A snapshot of the calibration data: conversion event, lookback, sample size, conversions, lift, confidence
- A sample of lead IDs whose behavior drove the suggestion (up to 100, for spot-checking)
This means three months from now, when someone asks "why is the demo-request weight 35?", you can pull up the audit log and see exactly which conversions justified the change.
See Audit Logs for more on how to read audit log entries.
Reverting a Change
There's no automatic undo. If you applied a suggestion and want to roll back:
- Open the rule in Configure > Scoring > [dimension] > [rule].
- Set the weight back to its previous value (you can find the previous value in the audit log).
- Save.
The next calibration refresh will likely surface the same suggestion again, since the underlying lift hasn't changed. If you keep reverting, that's a signal your business judgement disagrees with the data — worth a conversation with your team about whether the conversion event is the right one, or whether there's something the model isn't seeing.
Tips
- Sort by absolute delta, descending. This puts the biggest changes at the top. Tackle those first — they're where calibration moves the needle most.
- Filter out low confidence. Low-confidence suggestions are directional only; don't make weight changes based on them until more data arrives.
- Re-review after big funnel changes. New ICP, pricing, or pipeline structure? Wait one full lookback window, then revisit.
- Don't apply everything at once on day one. Apply 5-10 of the most impactful changes, then watch for a few weeks before doing another batch. Big simultaneous changes make it harder to attribute results.
- Ignore noisy single-point changes. A delta of 1 or 2 points usually isn't worth thinking about.