Musubi

Fixed ML Classifier False-Positive Cost Calculator

For teams running fixed ML classifiers (Hive, Rekognition, etc.)

How to use this calculator
  1. Pull a random 100 classifier flags from the last 7 days. Have a senior moderator (not the queue team) reclassify each as correct, false positive, or borderline.
  2. Plug your real numbers into the fields below. Defaults shown are typical mid-market values.
  3. Read your total annualized FP cost at the bottom.

Inputs

Total items moderated per month across all surfaces.
items
% of content your classifier flags. Check your dashboard.
%
% of classifier flags that are wrong (from the 100-item audit). Most customers see 50-80%.
%
Minutes per FP review. Typical quick visual scan is 10-15 seconds (0.15-0.25 min).
min
Per hour. Salary + benefits + overhead. Typical $30-50/hr.
$/hr
"Why was my content removed" tickets per 100 FPs. Most users don't bother filing; typical 2-8.
per 100
Minutes to handle a moderation-related ticket end to end.
min
Per hour. Typical $25-35/hr.
$/hr
% of FP'd users who eventually churn. Very conservative: 0.1-0.5%.
%
Lifetime value of a user. Use a conservative number.
$

Outputs

Monthly FP volume

374,400FPs/mo

Annual moderator override hours

14,976hrs/yr

FP'd users at churn risk

4,493/yr

Annual moderator override cost

$569,088

Annual support cost from FP tickets

$494,208

Annual churn cost (conservative)

$1,123,200

Total annual FP cost

$2,186,496/year