kpi-engine

KPI Engine - universal DataSift KPI reporting

Pulls your real calling activity from DataSift's per-record activity log (every call, text, and disposition event, caller-attributed and timestamped) and turns it into a graded KPI report: per caller, per day, and account-wide, with funnel pacing toward deals.

Why the activity log: DataSift's dashboard widgets are not exposed to scripts and cannot cleanly isolate "what happened on the phones this week." The per-record log is the one date-accurate, caller-attributed source.

Quick start

1. Log into app.reisift.io. Open DevTools (F12) -> Network tab -> click any request to apiv2.reisift.io -> copy the authorization: Bearer <token> value (the long JWT, without the word Bearer).
2. Save it: set env var REISIFT_TOKEN, or paste it into a file named reisift_token.txt next to the script. Tokens last about 48 hours; repeat when it expires.
3. Run:

python scripts/pull_kpis.py --days 7                 # trailing week
python scripts/pull_kpis.py --from 2026-07-06 --to 2026-07-16
python scripts/pull_kpis.py --days 1                 # today
python scripts/pull_kpis.py --days 7 --xlsx          # also build an Excel workbook
python scripts/pull_kpis.py --days 7 --detail        # + record-level CSV (one row per record worked)
python scripts/pull_kpis.py --days 7 --slack <webhook-url>   # post digest to Slack

No dependencies for markdown/CSV output (pure standard library). Excel output needs pip install openpyxl. Expect roughly 2-5 minutes per week of data (it reads each worked record's log).

What you get

• Volume: dials, answered, no-answer, records touched, talk time, first/last call window, dials per hour.
• Three rates, never collapsed into one "connect rate":
  • answer rate = answered / dials (loosest - includes voicemail pickups)
  • conversation rate = answered calls of 60s+ / dials (meaningful = 120s+)
  • contact rate = correct numbers / dials (right party confirmed)
• Dispositions: correct / wrong / dead / DNC numbers; not interested, follow-ups, dead leads.
• Leads that actually count: most reports only count Cold/Warm/Hot Lead statuses and miss that first-touch leads land in new_lead and No Contact New Lead - so leads read 0 while your callers produce 2-3 a day. This skill counts the full lead-status set (configurable in benchmarks.json).
• Funnel pacing: dials per correct number vs the ~9 (phone-scored) / ~32 (blind) benchmarks, correct numbers toward the 100-correct ~= 1-deal ratio, projected appointments from your appointment-take rate, projected contracts from your leads-per-contract ratio.
• Per-caller scorecards: dial floor (150/day) and conversation floor (5/day) MET or BELOW per caller, scaled by days active, plus lead targets.