Skip to content

Confidence Scoring

Each dormant detection includes a confidence score (0–100%) that tells you how reliable the signal is. Scores are calculated deterministically from your job data — no AI, no black boxes.

Fires when: No completed job in 12+ months.

Confidence is based on recency and history depth:

ConditionBase score
Last job 12–15 months ago70%
Last job 15–24 months ago85%
Last job 24+ months ago95%
+10% if client had 3+ jobs before going dormantBonus
+5% if client had 5+ jobs before going dormantBonus

A client with 6 jobs who last booked 18 months ago scores 90% (85 + 5 for 5+ jobs).

Fires when: Average job value dropped 30%+ compared to earlier jobs.

Confidence considers the size and consistency of the decline:

ConditionBase score
30–50% decline from peak65%
50%+ decline from peak80%
Decline sustained across 3+ jobs+15%
Only 2 total jobs (one big, one small)Cap at 50% (may be scope change, not erosion)

A client with 5 jobs averaging $2,000 who last two averaged $800 (60% drop) scores 80% (80 + 0, since sustained needs more jobs).

Fires when: Client who previously booked in a specific season missed their usual window.

Confidence depends on how many seasons of data exist:

ConditionBase score
Missed 1 expected season, 2+ years of history60%
Missed 2+ consecutive expected seasons85%
Only 1 prior season on recordCap at 50% (not enough history)
Client had exact month +-1 week multiple times+10%

A client who booked every April for 4 years but missed this April scores 85% (60 + 25 for 4 years of data).

If a detection doesn’t make sense for your market:

  • Dismiss the detection — it stays hidden unless you manually re-run detection.
  • Dismissed clients may re-appear on future scan if the signal strengthens.
  • You can also adjust detection parameters (coming in a future update).

No AI means:

  • Results are identical every run — no hidden prompt changes
  • Zero cost — no LLM tokens burned on detection
  • Auditable — you can verify the math against your own job history
  • Instant — detection runs in milliseconds, even at 5,000 clients