The period
One window, one motion. The window length matters more than people expect, and the timing card further down explains why.
The stages
Top to bottom, with the count that reached each one. Rename them to whatever your team actually says. Typical lag is how long it usually takes someone to get from the stage above to this one.
The picture
Volume on the left, so you can see the collapse. Conversion on the right, so you can compare the junctions to each other. The pale bar behind each rate is the range the number could honestly be.
Step by step
Each rate with its uncertainty. A rate computed on a handful of conversions is a guess wearing a percentage sign, so the range matters more than the number.
Timing and lag
The question this answers: an email goes out in January and the reply lands in March. Which month owns that reply, and what does it do to the rate you are reading?
The two ways to count, and what each one breaks
Cohort basis. You take everyone who entered the funnel in the window and follow them wherever they end up, even if that is months later. January's email and March's reply both belong to January, because January is when the cohort started. The rate is then honest about who those people were, but it is only finished once enough time has passed. Ask a cohort about closed-won six weeks after it entered, when deals take three months, and you will read a low number that is mostly just youth.
Period basis. You count every event that happened inside the window, no matter when the person entered. March's reply belongs to March. Nothing is unfinished, so the numbers settle immediately. The cost is that the numerator and the denominator describe different groups of people. If volume at the top is growing, the bottom rate is divided by a top that has swollen since those deals started, and it reads worse than reality. If volume is shrinking, it flatters you.
Neither is wrong. Pick one, write down which, and never compare two periods counted different ways. Most arguments about a conversion rate turn out to be two people using different bases and neither of them saying so.
Benchmarks
A benchmark is only useful if it was measured the way you measured. That rules out most of them, which is why this offers three and puts the weakest one last.
Saved periods
Save this week or this month, change something, and save again. The comparison above then tells you whether a move is real or whether you are reading noise and calling it a lesson.