The short version
A conversion rate is a ratio between the top of a funnel and the bottom, so it averages away everything in between. When it falls, at least three unrelated causes produce an identical number: traffic that was never human, a handoff that loses the attribution, and a last step that is simply broken. Measured top to bottom they cannot be told apart. Measured step to step they are obvious, because each one collapses a different join in the chain. The rule we work to now is that no report shows a conversion rate without the step-to-step ratios printed underneath it, because the total tells you something is wrong and never where.
What a single conversion rate actually averages
Conversion rate is sessions divided by outcomes. Every step between those two ends - the page that loads, the tap that responds, the form that submits, the phone that gets answered - is compressed into one ratio. That is fine while the number is healthy and useless the moment it is not, because a fall of the same size can come from the top of the funnel, the middle, or the last join before money changes hands, and the ratio reports all three identically.
This matters more than it used to, because the denominator has become unstable. Sessions now include a growing share of automated traffic, and impressions are inflated by AI surfaces issuing several machine queries where a person asked one question. A ratio built on a denominator that moves for reasons unrelated to your marketing will move too - and you will go looking for the cause in the wrong half of the site.
Measure the steps, not the total
The fix is unglamorous and takes about an hour. Instead of one number, record the ratio between each adjacent pair of steps. For a contractor that chain runs session, service page, form start, form submit, call answered, job booked. For a store it runs session, product view, cart, checkout, paid. Either way, five or six ratios rather than one.
What you are looking for is not the smallest ratio - some steps are naturally lossy and always will be - but the one that is out of line with the steps on either side of it. A healthy chain degrades gradually. A broken one has a cliff, and the cliff is always adjacent to the thing that is actually wrong. That is the entire diagnostic, and it is why the total is worse than useless: averaging is precisely the operation that removes a cliff.
Failure one: the traffic was never real
Automated traffic arrives, registers a session and leaves. It never reaches a second step, so it inflates the denominator of every ratio computed from sessions and touches none of the numerators. Your conversion rate falls and nothing about your site has changed.
It is easy to spot once you look for it: a city in your top locations where you have no customers and no service area - usually somewhere that hosts data centres rather than people; sessions with no second page; a device or browser mix that does not match the rest of your traffic. The danger here is not the traffic itself, which costs you nothing. It is that the depressed top-line rate reads as a quality problem, so people start cutting the channels that were working.
Failure two: the handoff loses the attribution
The second failure is not lost business at all. It is lost knowledge about the business, which is worse in a specific way: it does not show up as a smaller number anywhere, so nothing prompts you to look.
Any time a visitor crosses to another host mid-journey - a booking widget, a payment page, a scheduling tool on a different domain - the session can restart, and the new session is credited to the referrer it just arrived from. Which is you. The symptom is unmistakable once you know it: your own domain appears in your own referrer report, sometimes with a serious share of revenue behind it. That revenue was earned by some channel and is now filed under nothing, and every decision about where to spend next is being made with that share missing.
Failure three: the last step is broken and nobody owns it
The third is the expensive one. Every step up to the final action reads normally, and then the last join collapses - people arrive at the point of committing and do not commit, at a rate far worse than any step before it.
For a contractor that last step is the quote form and the phone behind it. For a store it is the payment screen. The common feature is that it sits past the boundary of whoever was hired to bring the traffic. The SEO retainer reports sessions and positions and stops. The booking tool is somebody else's product. The phone is the office. So the one step where the money is actually lost is the only step in the chain with no owner - and it can stay broken for a very long time without anyone raising it, because nothing anyone is reporting looks wrong.
Why the total conceals it so well
Run the arithmetic and the concealment is obvious. A chain that loses a normal amount at every step and almost everything at the last one produces the same headline rate as a chain that loses slightly more than normal everywhere and nothing catastrophic anywhere. The first is an afternoon's work to fix. The second is a year of grinding improvement. The single number cannot distinguish them, and it is the number most reports lead with.
What we changed in our own reporting
Two rules, both boring. No report shows a conversion rate without the step-to-step chain printed underneath it - and the last step is always included, even when it runs on somebody else's software, because a step nobody reports is a step nobody fixes.
The second rule is that a percentage never appears without the two absolute numbers it was computed from. A rate on its own cannot be checked, and rates built on unstable denominators are exactly the ones that most need checking. It is also the fastest error-catcher we have found: a percentage that does not follow from the numbers printed beside it is visible at a glance, and it is remarkable how often one does not.
Check it on your own site
Write the chain down before you open analytics
Name every step between a stranger arriving and money changing hands. If you cannot name them, you cannot measure them, and you will default to the single ratio again. Most contractors land on six: session, service page, form start, form submit, call answered, job booked.
Record each adjacent ratio, never the total
Five or six numbers, each one dividing a step by the step directly before it. Write the two absolute counts next to each ratio. Take the same window last year if you have it, because a single period tells you the shape of the funnel but not whether it changed.
Look for the cliff, not the lowest number
Some steps are naturally lossy and a low ratio there is normal. What you want is the step that is badly out of line with its immediate neighbours. That discontinuity is adjacent to the actual fault, and it is the only thing in this exercise that tells you where to spend the next week.
Search your referrer report for your own domain
If your own hostname appears as a referrer, the journey is crossing hosts and restarting the session somewhere - usually at a booking, scheduling or payment step. Whatever revenue sits behind that line is currently credited to nobody, and fixing it changes what the rest of your reporting means.
Sort sessions by city and find the one you do not serve
A city in your top locations that you have never taken a job in is almost always automated traffic rather than a new market. Segment it out and recompute the chain. If your ratios move materially, your headline rate was measuring the internet rather than your customers.