Cold Start
Calibration is a data-driven feature, which means it needs data before it can recommend anything. This page explains what to expect in the first weeks of using kenbun, why some rules show suggestions before others, and what "cold start" really means in practice.
What "Cold Start" Means
When you first connect kenbun, calibration kicks off immediately — but with no historical data, there's nothing for it to measure. Every rule's lift estimate would be based on zero conversions, which is the same as no information at all. Rather than showing meaningless numbers, kenbun waits until enough data accumulates to give a confident answer.
The good news: data starts accumulating from your very first event. The cold start period is finite and predictable.
When Will I See Suggestions?
A rough rule of thumb:
| Days after setup | What you'll see |
|---|---|
| 0-7 | Calibration panel shows "Collecting data" for most rules |
| 1-2 weeks | Suggestions start appearing for high-volume rules |
| 3-4 weeks | Most rules have suggestions; confidence is still building |
| 6-8 weeks | Calibration is fully warm; confidence bands stabilize |
The exact timing depends on three things:
- Lead volume: more leads = faster data accumulation
- Conversion volume: you need enough actual conversions to compare against
- Rule popularity: rules that fire on every lead reach a useful sample size much faster than rare rules
A rule that fires on 5,000 leads per month will be calibrated within days. A rule that fires on 30 leads per month may take months to reach high confidence — and that's fine.
Confidence Bands Explained
Each suggestion comes with a confidence indicator. In plain language:
- High confidence: large sample, lots of conversions, the lift estimate is statistically meaningful. Trust it.
- Medium confidence: enough data to suggest a direction, but the exact number could shift as more data arrives. Use as a guide, not gospel.
- Low confidence: small sample. The suggestion shows a direction (this rule probably matters or probably doesn't), but the magnitude is uncertain.
- Needs more data: not enough leads or conversions yet. No suggestion shown.
Higher-confidence suggestions are safer to apply. Lower-confidence ones are more like "watch this space — check back in a few weeks."
"Why Don't I See a Suggestion for This Rule?"
A few common reasons:
The rule is brand new. You created it last week. Wait. Suggestions appear once enough leads have triggered the rule.
The rule fires on very few leads. Niche events take longer to reach a useful sample size. If a rule fires on only 10 leads per month, it'll take time to gather enough data — and the suggestion will likely be lower-confidence even when it appears.
Your conversion event hasn't fired enough. If you picked "Closed-won deal" but only have 5 closed deals in your lookback window, calibration doesn't have enough conversion examples to compute lift reliably for any rule. Either widen the lookback window, switch to a higher-volume conversion event (like Engagement persistence), or just give it more time.
You just changed the conversion definition. Calibration recomputes from scratch when you change the conversion event. Suggestions will reappear after the next scheduled refresh.
A Patience Note
Calibration is a long-game feature. The best signal comes from steady, consistent data collection over weeks. Resist the temptation to apply low-confidence suggestions in the first week — you'll just be reacting to noise.
A useful workflow for new accounts:
- Week 1: Set up your conversion event. Do nothing else with calibration yet. Let data accumulate.
- Week 2-3: Check the Calibration Review page. Note which high-volume rules are starting to show suggestions. Don't apply yet.
- Week 4: Sort by largest delta. Review the top 5-10 suggestions with high confidence. Apply the ones that align with your business judgement; defer the rest.
- Month 2 onward: Make calibration review a regular cadence — weekly or biweekly. By now, you'll have stable suggestions across most rules.
This pacing keeps your scoring system stable and gives you time to understand what calibration is telling you before you act on it.
What Calibration Doesn't Need
A few common worries that don't apply:
"Do I need to backfill historical events?" No. Calibration uses whatever events kenbun has captured since you started. Backfilling helps but isn't required.
"Do I need a perfect CRM integration?" No. The default conversion event (Engagement persistence) works with zero integrations. You can layer richer data on later.
"Do I need to pause my scoring rules during cold start?" No. Your scoring continues to run normally on whatever weights you've set. Calibration is a recommendation layer that runs alongside — it doesn't gate your existing scoring.