The idea
A conversion rate answers one question: of the people who showed up, how many did the thing we wanted? It is a fraction, and a fraction has two parts that you choose. The top is what counts as "did it". The bottom is who counts as "showed up". Both choices are part of the metric.
At Harbor & Pine the top is a session with converted = true, and the bottom is every session
that started in the period. With that definition 2025 gave 500 conversions out of 16,225 sessions,
3.08%. Move the denominator to engaged sessions and the same 500 become 6.93%.
Figure 1
The 2025 web funnel: three counts, three different rates
- All sessions 16,225
- Engaged 3+ pages: 7,213
- Converted 500
- Count every session that started in 2025: 16,225. This is the widest part of the funnel, and the usual denominator of 'the conversion rate'.
- 7,213 of them viewed three or more pages, 44.46% of all sessions. The other 55.54% left after one or two pages.
- 500 sessions converted. That is 6.93% of engaged sessions, but 3.08% of all sessions. Same 500, two honest rates: only the denominator changed.
Why it exists
Marketing has to decide where to spend. The rate tells you whether a source, a device or a landing page turns visits into customers. A raw count cannot do that: more traffic produces more conversions even if the site got worse.
The web data comes with limits that the definition must respect. web_sessions is its own funnel
dataset. A session carries no order id, a conversion is only recorded for a known customer who
viewed at least three pages, and app and marketplace orders never pass through it. A session
conversion rate is therefore not "orders divided by visits". If you divide orders by sessions you
are mixing two datasets that were never meant to be joined, and the answer has no meaning.
How it works under the hood
A funnel is a chain of denominators
Each stage divides by something different. From all sessions to engaged: 44.46%. From engaged to converted: 6.93%. From all to converted: 3.08%. The first two are step conversions, and multiplying them gives the third: 0.4446 × 0.0693 = 0.0308. The place where most people leave is the step with the lowest step rate, so reading each step against the one before it shows where the funnel leaks. Reading every stage against "all sessions" shows only how small the end is.
A related number is the bounce rate: sessions with fewer than three pages, as a share of all sessions. Here that is 55.54%, simply the other side of the engaged step.
Segments and the weighted average
Split the sessions by source or device and every segment gets its own rate. Those segment rates are not independent of the total. The total rate is the weighted average of the segment rates, and the weight of each segment is its share of sessions:
overall = Σ (segment share × segment rate)
It follows that the total can change for two different reasons: the segment rates moved (the rate effect), or the shares moved (the mix effect). Mix does not mean anyone did better or worse. It means the audience changed.
Figure 2
Every device improved, the total barely moved
| device | share_2023 | rate_2023 | share_2025 | rate_2025 |
|---|---|---|---|---|
| desktop | 41.9% | 3.78% | 32.4% | 4.13% |
| mobile | 47.8% | 2.30% | 58.0% | 2.52% |
| tablet | 10.3% | 2.93% | 9.7% | 2.94% |
| view | rate |
|---|---|
| 2023 actual | 2.99% |
| 2025 actual | 3.08% |
| 2025 rates, 2023 mix | 3.24% |
- Conversion by device, with each device's share of sessions. Rates are conversions divided by that device's own sessions.
- Between 2023 and 2025 every rate went up: desktop 3.78% to 4.13%, mobile 2.30% to 2.52%, tablet 2.93% to 2.94%.
- But traffic moved. Mobile, the lowest converter of the three, grew from 47.8% to 58.0% of sessions. Desktop shrank from 41.9% to 32.4%.
- The overall rate only rose from 2.99% to 3.08%. Apply the 2025 device rates to the 2023 traffic mix and you get 3.24%: the missing 0.16 points are the mix shift.
A worked example
Take 2025. Desktop is 32.4% of sessions and converts at 4.13%, mobile is 58.0% at 2.52% and tablet is 9.7% at 2.94%. Multiply and add: 0.324 × 4.13 + 0.580 × 2.52 + 0.097 × 2.94 = 3.08. That is exactly 500 conversions divided by 16,225 sessions, reached the long way round. The weights are what make the two routes agree. Use any other weights, such as one third each, and the answer is 3.20%, a number that describes no real traffic.
Splitting a change in two
To separate the two effects, recompute the new period with the old mix. That is the 3.24% in the diagram: what 2025 would have scored if the traffic still looked like 2023's. The gap between 3.24% and the real 3.08% is the mix effect, caused by mobile growing. The gap between 2.99% and 3.24% is the rate effect, the genuine improvement on each device. A useful report states both.
Small segments are noisy
A rate is only as steady as the count under it. Tablet had 1,566 sessions in 2025 and 46 conversions. One extra conversion moves its rate by 0.06 points, which is about as much as the whole 2023 to 2025 change on that device (2.93% to 2.94%). Before you celebrate or panic about a segment rate, look at the number of sessions behind it. A rate on a few hundred sessions can move by itself from one month to the next.
Attribution is a rule, not a fact
Marketing splits conversions by source. Last-touch attribution gives the credit to the source of
the converting session. That is a rule and it flatters whatever source customers use at the end of
the journey. Anonymous sessions cannot convert in this dataset at all, so a source that mostly
brings new, not-yet-logged-in visitors will always look poor.
Write the definition next to the number
A conversion tile should carry its recipe: converted sessions ÷ all sessions, for sessions started in the period, known customers with 3+ pages only. That one line stops the 3.08% and the 6.93% from ever being quoted in the same meeting as if they were the same thing.
Common misconceptions
"The conversion rate is orders divided by visits"
Orders and sessions come from different datasets here and cannot be divided by each other. Say what you counted: converted sessions over all sessions. Check the numerator and denominator before you compare two rates, because a rate whose bottom changed is a different metric.
"The total rate is the average of the segment rates"
The average of 4.13%, 2.52% and 2.94% is 3.20%, but the real total is 3.08%. The simple average treats tablet, with under 10% of traffic, as if it were as big as mobile with 58%. Weight each rate by its sessions, or just divide total conversions by total sessions.
"If every segment improved, the total must have improved by as much"
Not necessarily. When traffic moves toward the weaker segment, the total improves less, stays flat or even falls. Before you blame or praise a result, compare the segment rates with the segment shares: if the shares moved, part of the change is mix.
Check yourself
One question. Answering it ticks this lesson off on the map; it never locks a level.
Not quite. Sessions carry no order id, so a session conversion rate counts converted sessions. Orders belong to a different measure.Try another answer.
Not quite. Device is a segment, not the definition. The total rate uses all sessions of the year.Try another answer.
Not quite. Averaging device rates gives tablet the same weight as mobile. The total is sessions-weighted, which is conversions divided by sessions.Try another answer.
Right. The tile mixes two definitions. 6.93% is the step conversion from engaged to converted; the conversion rate uses every session in the denominator.
Lesson done: it is ticked on the map.