Level 2 · Which traffic source converts

One rate per source

A channel slide needs the conversion rate for each traffic source. It is the same formula as before, computed per group:

s.groupby("source").agg(sessions=("session_id", "size"), conversions=("converted", "sum"))

converted is a boolean column, so summing it counts the converted sessions. Dividing conversions by sessions within each row gives each source’s rate. Over 2023 to 2025, email converts 9.41% and social 0.62%: a fifteen-fold gap.

The denominator decides the story

The agency’s 5% comes from a different denominator. In this data a conversion is only recorded for a known, logged-in customer, so anonymous sessions can never convert. Divide by known sessions only and every rate goes up, by very different amounts:

source	anonymous share	all sessions	known only
email	10.7%	9.41%	10.54%
organic	65.3%	1.83%	5.28%
social	71.3%	0.62%	2.18%

Email traffic is almost all known customers clicking a newsletter, so the two rates are close. Organic and social bring mostly new, anonymous visitors, so removing them nearly triples the rate.

Which one goes on the slide

The all-session rate answers “of the traffic this source brings, how much buys?” That is the marketing question, and it is the KPI. The known-only rate answers “once a visitor is logged in, how likely are they to buy?” That is useful for diagnosing the checkout, and misleading as a channel comparison, because it hides how much of each channel’s traffic never gets that far.

Give both if you like, but label the denominator in the column name, as this task does.

Common mistakes

• Averaging per-session flags in the wrong place. s["converted"].mean() is the overall rate; per source you need the group mean or the two counts.
• Silently filtering customer_id.notna() early in the analysis “to clean the data”. Anonymous sessions are not dirty, they are most of the traffic.
• Sorting without resetting the index, which leaves the old row labels behind.