Reports you can open before you run anything

How likely are customers on each plan to recommend us?

The data. standard_nps.csv, the published sample file for NPS Analysis: 480 rows; columns response_id, plan, likelihood_to_recommend. Sample data, not a customer's.

The calculation. Python, standard library only. This exact code ran against the file when this page was built, and printed the output below it.

import csv
rows = list(csv.DictReader(open("standard_nps.csv")))
def nps(v):
    pro = sum(s >= 9 for s in v) / len(v)
    det = sum(s <= 6 for s in v) / len(v)
    se = ((pro + det - (pro - det) ** 2) / len(v)) ** 0.5
    return round(100 * (pro - det), 1), round(196 * se, 1)
plans = {}
for r in rows:
    score = int(r["likelihood_to_recommend"])
    plans.setdefault(r["plan"], []).append(score)
allv = [s for v in plans.values() for s in v]
print(len(allv), "responses: NPS %s +/- %s" % nps(allv))
for p, v in sorted(plans.items()):
    print(p, len(v), "responses: NPS %s +/- %s" % nps(v))

480 responses: NPS 11.0 +/- 7.0
Business 130 responses: NPS 35.4 +/- 11.5
Starter 150 responses: NPS -22.0 +/- 12.5
Team 200 responses: NPS 20.0 +/- 10.7

What it says. Net Promoter Score is the share of 9 and 10 ratings minus the share of 0 to 6. Overall it is 11, but the plans differ widely: Business 35.4, Starter -22, Team 20. The plus-or-minus figure is a 95% margin of error: 7 points overall and up to 12.5 for a single plan, so a move of a few points between surveys is often noise.

What it does not say. NPS throws away the difference between a 0 and a 6, and between a 9 and a 10. It says how people feel, not why; the library analysis breaks the score down by segment and tests whether the differences between segments are significant.

Run NPS Analysis on this file or your own, or open its stored sample report.