The problem both models solve
You want to ask a returning visitor something new rather than the same three fields again. Progressive profiling handles that, and there are two ways to think about it.
The queue model
You define an ordered list of questions and the form serves the next unanswered ones. Simple to reason about, easy to explain to a client, and it degrades gracefully — if someone answers out of order, the queue just moves on.
- Good when the questions are independent of each other.
- Good when you care more about total fields collected than about which ones.
- Falls over when a later question only makes sense given an earlier answer.
The dependency model
Questions are conditional: ask about fleet size only if they said they run vehicles. More relevant, better completion rates, and considerably more to maintain.
- Good when the questions genuinely branch.
- Good when a wrong question would damage credibility.
- Falls over when the dependency chain grows past a handful of branches and nobody can remember what asks what.
The failure mode is not technical. It is that six months later nobody on the team can say with confidence what a given visitor will be shown.
How to choose
Count the branches. If you have fewer than three genuine conditionals, use the queue — the extra relevance is not worth the maintenance. If the questions truly depend on each other, use dependencies, and document the tree somewhere a human will find it.
Pick the model your team can still explain in a year, not the one that demos best.