The average that lied to me
The first time a plain average burned me, I was looking at conversion rate across three traffic sources. Simple average of the three rates came out looking healthy. Then I pulled the actual revenue numbers and they didn’t match up at all. The problem wasn’t the math. It was that one of those three sources sent me ten visitors a month and another sent me two thousand. Averaging the three rates like they were equally important gave the tiny, noisy source the same voice as the source doing almost all the work.
That’s the whole reason weighted averages exist. A regular average treats every number in the group as equally important. A weighted average lets you say “this number matters more because it represents more” and adjusts the result accordingly. Once you see it, you can’t unsee how often a plain average is quietly misleading you.
What a weighted average actually is
A simple average adds up your numbers and divides by how many there are. Every value gets the same vote.
A weighted average multiplies each value by a “weight” first, usually something like the size, volume, or dollar amount behind that value, adds up those products, then divides by the sum of the weights. The formula looks like this:
weighted average = (value1 × weight1 + value2 × weight2 + …) ÷ (weight1 + weight2 + …)
The weight is whatever makes sense as the “size” of that data point. It could be number of customers, dollars of revenue, number of sessions, or days in a billing period. The point is that a value backed by more volume pulls the average harder toward itself, and a value backed by almost nothing barely moves it at all.
A worked example: blended conversion rate
Say you run paid ads on two platforms. Platform A sends 100 visitors a month and converts 10 of them, a 10% rate. Platform B sends 1,000 visitors a month and converts 30 of them, a 3% rate.
A simple average of 10% and 3% is 6.5%. That number describes neither platform and doesn’t describe your business either, because it pretends both platforms contributed equally to your traffic.
The weighted average, using visitor count as the weight, looks like this:
(10% × 100 + 3% × 1,000) ÷ (100 + 1,000) = (10 + 30) ÷ 1,100 = 3.6%
That 3.6% is your actual blended conversion rate, the number you’d get if you just divided total conversions by total visitors. It’s dragged down close to Platform B’s rate because Platform B is where almost all your traffic actually comes from. If you’d used the 6.5% simple average to forecast next month’s signups, you’d have overestimated by nearly double. That’s not a rounding error, that’s a decision made on a fictional number.
Where this shows up in the metrics you already track
Weighted averages aren’t a separate topic from churn, MRR, or CLV. They’re baked into how those metrics are supposed to be calculated correctly, and it’s easy to calculate them wrong without noticing.
Blended churn rate. If you sell three plans with different customer counts, your overall churn rate isn’t the average of the three plans’ churn rates. It’s the weighted average, weighted by how many customers are on each plan. A plan with five customers and 40% churn shouldn’t move your overall number nearly as much as a plan with 500 customers and 4% churn, even though 40 is a bigger number than 4.
Average revenue per user across cohorts. If you’re comparing ARPU across signup cohorts of very different sizes, weighting by cohort size keeps a tiny early cohort from distorting the picture. A cohort of 8 people from your first month can have wild average numbers just from small-sample noise. Weight it fairly and it won’t hijack the trend line.
Customer lifetime value blended across segments. If you serve both a low-price self-serve tier and a higher-price done-for-you tier, a single blended CLV number is only honest if it’s weighted by how many customers actually sit in each tier, not just averaged straight across the two tiers as if they had equal headcount.
Time-weighted MRR. If a customer was on a $50 plan for 20 days of the month and then upgraded to a $90 plan for the remaining 10 days, a fair monthly average for that customer weights each price by the number of days it was in effect, not by treating the two prices as equally represented.
In every one of these, the mechanics are identical to the ad-platform example. Pick the right weight, meaning the thing that represents how much each value actually contributed, and the average tells you the truth about the whole instead of the truth about an arbitrary subgroup.
The weight is a judgment call, not a fact
Here’s the part people skip. The result of a weighted average is only as honest as the weight you chose, and picking the weight is a judgment call, not a neutral math step.
If you weight churn by customer count, a handful of high-value enterprise customers churning gets buried under a much larger number of low-value free-tier accounts staying put. If you weight by revenue instead, that same churn event looks a lot scarier, because it’s weighted by the dollars walking out the door rather than the headcount. Neither weighting is “the correct” one in some absolute sense. They answer different questions. Customer-count weighting tells you about retention across your base. Revenue weighting tells you about retention of the dollars that keep the lights on. If you only ever look at one, you’ll miss what the other one would have told you.
This is also where a weighted average stops being purely mechanical and starts requiring you to know your own business. I weight my churn numbers by revenue first because a handful of larger accounts leaving hurts a lot more than the same count of small accounts leaving, and I want that reflected in the headline number I look at each week. But I also keep the customer-count-weighted version around, because if that one starts moving badly it usually means something is wrong with the product experience for smaller customers even while the revenue number still looks fine.
How I actually use this when I check my own numbers
Practically, I ask two questions before I trust any average I’m looking at. First, are the underlying groups actually different sizes? If channel, plan, or cohort sizes are close to equal, a simple average and a weighted average will land close to each other and it doesn’t matter much. Second, if the groups are unequal, what am I actually trying to measure, the experience of a typical individual unit or the total behavior of the whole? Those two questions point you to different weights, or sometimes tell you a weighted average isn’t even the right tool and you should be looking at the raw totals instead.
What a weighted average can’t tell you
A weighted average, done correctly, tells you what’s true about the blended whole given the weight you picked. It doesn’t tell you why the underlying numbers are what they are, and it doesn’t tell you which weighting scheme is the “right” one for a decision you’re about to make. It’s also still sensitive to how you define the groups in the first place. Two people can build technically correct weighted averages from the same raw data and land on different headline numbers just by choosing different weights, and both can defend their choice. Treat the number as a starting point for a conversation with yourself about what actually happened, not as a verdict.
If you want more explainers like this on the metrics and methods behind running a small operation on your own numbers, you can find the rest on the home page.
Get new guides and videos first — join the Telegram channel.