Your quiz funnel has seven stages. You are probably watching two.

Ask ten quiz marketers what a good completion rate is and you will get ten numbers, all confidently stated, none of them useful to you. The number depends on where the traffic came from, whether the quiz was embedded or standalone, how many questions it had, and whether the person arrived intending to find something out or arrived because they misclicked. An average across all of that describes nobody.

What does transfer between accounts is the shape. RooQuiz records seven stages for every session, and the ratios between adjacent stages are diagnostic in a way that absolute percentages never are. Each drop tells you about a different screen, and each one has a different fix.

The seven stages, and what each drop actually means

Stage	Recorded when	A drop here means
viewed	The quiz page or embed renders	—
started	First question answered	Your promise did not land, or the entry screen asked for effort before offering a reason
submitted	Last question answered	Length, question quality, or a specific question people refuse
leadCaptured	Contact details captured	The gate is badly placed or badly framed
reportViewed	Result screen actually rendered	Rare — usually a technical fault or a slow report
ctaClicked	The next-step button is clicked	The result screen described them but did not tell them what to do
shared	Share action taken	The result was not interesting enough to be worth attaching your name to

Two things are worth noticing in that column on the right. First, every drop points at exactly one screen, which means the funnel is a work queue and not just a scoreboard. Second, viewed → started and ctaClicked sit at opposite ends and are the two most commonly reported numbers — which is why so many quizzes get optimised in the middle by guesswork.

Why the absolute numbers travel so badly

Consider two funnels that both produced 40 leads last month.

A  paid social, standalone page
   2,000 viewed → 600 started → 320 submitted → 40 leadCaptured

B  embedded on a pricing page
   180 viewed → 150 started → 120 submitted → 40 leadCaptured

Funnel A converts 2% of views into leads. Funnel B converts 22%. On a benchmark table, A looks broken and B looks excellent. In reality A is doing what paid social does — buying a lot of cheap, weakly-intentioned attention — and B is doing what a pricing-page embed does, catching people who were already halfway to a decision.

Neither number tells you what to fix. The ratios do. A loses 70% of viewers before the first question, which is an entry-screen problem. B keeps 80% of viewers through to submission and then loses two-thirds of them at the gate, which is a lead-capture problem. Same lead count, completely different week of work.

This is why RooQuiz splits the funnel by utm_source and flags whether the session was embedded. A blended number across both of those funnels would have averaged out into something that describes neither.

Read the ratios, not the totals

Take the ratio between each adjacent pair and you get five numbers that mean something on their own terms. Using funnel B above:

• started / viewed = 150/180 = 83%
• submitted / started = 120/150 = 80%
• leadCaptured / submitted = 40/120 = 33%
• reportViewed / leadCaptured — should be near 100%
• ctaClicked / reportViewed — the number that actually pays

Note what the fourth one is for. reportViewed almost never has a legitimate reason to drop; if it does, something is broken, not underperforming. Treating it as a conversion metric to improve is a category error. Treat it as an alarm.

The lines worth drawing

These are not industry averages. They are the points at which the ratio is telling you something specific enough to act on, and they hold regardless of your traffic mix.

Ratio	When to investigate	What to look at first
started / viewed	Below half	The entry screen. Is the promise concrete, and is question one the easiest one?
submitted / started	Below two-thirds	The dropOff data, not the whole quiz — see below
leadCaptured / submitted	Below a quarter	Gate placement, and whether the ask is framed as a continuation or a toll
reportViewed / leadCaptured	Below 95%	A fault. Check report generation time and errors
ctaClicked / reportViewed	Below a tenth	The result screen's next step, and whether there is more than one

The leadCaptured row is the one most often misdiagnosed as a copy problem when it is a placement problem — a gate in front of a result people have already earned reads very differently from a gate in front of a result they have not seen yet. That mechanic is worth its own read: why your quiz has no leads.

The drop-off data changes the answer

A weak submitted / started ratio invites the wrong reflex, which is to cut the quiz shorter. Before doing that, look at the per-question stall data. RooQuiz reports which question unsubmitted sessions were sitting on when they left, with its index and the count.

If the stalls are spread evenly across every question, length genuinely is the problem and cutting helps. If they pile up on one question, length is not the problem — that question is. In practice the concentrated pattern is the common one, and the offending question is usually one of a small number of types:

• It asks for a number the person does not know off-hand (revenue, headcount, conversion rate).
• It forces a choice between options that all feel wrong, with no escape option.
• It is the first question that reveals the quiz is really a sales qualification form.
• It asks something that feels like it will be used against them.

Cutting three good questions to compensate for one bad one leaves you with a shorter quiz that still stalls.

Setting your own line

The thresholds above are starting points for investigation, not targets. Your actual line comes from your own baseline, and establishing one takes more traffic than people expect. A ratio computed on 40 sessions moves several points if three people behave differently, so a week-over-week change on small numbers is usually noise rather than a result.

A workable order of operations:

1. Let one funnel accumulate enough sessions that a single person cannot move a ratio by more than a point or two.
2. Record all five ratios as your baseline, per channel — never blended.
3. Change one screen at a time, and watch only the ratio that screen controls.
4. Re-baseline whenever the traffic mix changes, because a new channel invalidates the old comparison entirely.

That last step is the one people skip. Launching a paid campaign into a funnel whose baseline came from organic traffic makes every ratio drop at once, and the resulting fire drill produces changes to screens that were never the problem.

The benchmark you were looking for does not exist. What exists is a funnel where every stage points at one screen, and a set of ratios that tell you which screen to open.

Photo by Immo Wegmann on Unsplash