Why is our overall conversion rate falling when every location improved?
Mix shift. If a lower-converting location grew its share of total volume, the blended rate can fall even though every location's rate rose. This is Simpson's paradox, and it appears whenever you average across segments of different sizes. Always read segment rates alongside each segment's share of volume before concluding that a rate moved at all.
How the arithmetic produces the illusion
A blended rate is a weighted average, and the weights are volume shares. If the segment with the lower rate grows faster than the others, its weight rises and pulls the blend down. Nothing in any segment has to get worse for the headline to get worse.
The reverse happens just as often and is more dangerous, because it looks like success. A blended rate can improve purely because a high-converting segment took share while every segment individually declined.
The clue that mix is involved is a headline moving in the opposite direction from most of its parts. Whenever the summary disagrees with the majority of the segments underneath it, the weights are doing the work, and no amount of examining the segments individually will explain the summary.
Where it shows up in service businesses
- Multi-location rollups. Opening a new market drags every blended rate down for a year, because ramping locations convert worse than mature ones.
- Channel mix. Shifting spend toward a broader, cheaper source lowers blended booking rate even if each source held steady.
- Job type mix. A maintenance-heavy month has lower average ticket and higher booking rate than a replacement-heavy month, regardless of performance.
- New versus repeat customers. Repeat customers book at a different rate; a change in their share moves the blend on its own.
The diagnostic is decomposition
Split the change in the blended rate into two parts. Recompute the current period using the prior period's volume weights, which gives you the rate effect: what the blend would have done if mix had not moved. The remainder is the mix effect.
This takes about ten minutes in a spreadsheet and it settles the question permanently. Reported next to the headline, it turns a confusing number into a sentence: the blended rate fell, all of it from mix, and here is which segment grew. Building that decomposition into the reporting layer is standard marketing intelligence work.
A reporting rule that prevents the whole class of error
Never show a blended rate without showing the segment shares that produced it. A small stacked bar of volume share next to the rate line is enough. Once the reader can see the weights, they stop misreading the average, and you stop having the same conversation every quarter.
The same rule applies to automated analysis. Any system that narrates a rate change should be required to check mix before it attributes the movement to performance.
The same caution applies to year-over-year comparisons. If your footprint, channel mix or job mix changed materially, the blended figures are comparing two different businesses, and the honest comparison is same-location, same-channel or same-job-type.
Topics: Simpson's paradox · mix shift · rollups · multi-location
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.