Every field carries an abandonment cost

What it is

Each extra question loses some people at that question, even when it “only takes a few seconds.” Every field has an abandonment cost means drop-off is not a single tax on the whole form. It is marginal loss accumulated along fields: arrival rate steps down after each ask. How many boxes are visible on one screen shapes effort estimates at entry. This entry is about asking versus not asking the item at all—even if it lives alone on step 5 as a single input. Whether to delete for lack of purpose, and whether asking later is better, are the next two steps. Here the only claim is that one more ask is one more chance to leave.

Why it happens

Each item is a micro-decision: understand the label, have an answer, be willing to give it. The exit from a failed decision is leaving. Fields contribute that exit independently, so leave at item n can happen after n−1 are already filled—sunk cost will not stop “this one means opening email.” A second layer is nonlinear cost. ID numbers, open questions, items that need another page to look up, have far higher marginal leave than a recognition item like gender. An average “2% leave per field” hides the peaks. Paging does not zero the marginal cost; it only moves that item’s leave from the first screen to screen 5; the funnel still drops there. Splitting one item into two smaller boxes doubles the number of asks, and doubles the chances to leave. Cost hangs on the ask, not on pixels.

Studying it

Build a field-level funnel: arrival, start typing, complete, leave for each item. Compare whole-form completion before and after removing an item, and whether leave merely shifts to the neighbors.

Independent variables: whether the field exists, how hard it is to obtain (recognition / recall / lookup), which step it sits on. Dependent variables: leave rate on that item, whole-form completion, whether leave moves to the next item, completion time.

Labs that require finishing almost never see marginal leave. Use live traffic or unincentivized remote tasks. An experiment that only reduces fields visible on one screen is not “ask one fewer time”—the former measures visible density, the latter the number of questions.

Where it stops holding

Legally required items still have marginal cost that cannot be paid by deletion; pay by lowering obtainment cost (upload instead of retype, read from an ID already on file). Expert internal forms have low marginal leave; repetition and opaque labels are what raise it. High-yield tasks with no alternative (a visa) tolerate more accumulated cost if every item can be understood as a necessary step toward the result. Making an item skippable lowers marginal cost, not to zero—seeing it still requires a skip decision.

Applying it

• Tag every field with arrival and leave; rank by leave peaks, not by importance on the mock.
• At a peak, ask first: can it become recognition, can it be fetched, can it leave the flow. Do not start by making it visually shorter.
• Before splitting an item, count asks: two boxes are two chances to leave, unless the original box peaked because it was hard to fill.
• Verify by removing or rewriting the highest-leave item and watching whole-form completion in the same traffic window, and whether leave only moved to the next item. If completion is flat and the peak merely shifted, that item was not the true cost source; keep walking the funnel to the next peak.

Related

• Within the group: H1.12.2 Fields whose purpose cannot be explained should be deleted · H1.12.3 Collecting later beats collecting everything now
• Adjacent: H1.01 Form length and multi-step · H1.03 Required and optional markers · O1.02 Data minimization
• Search terms: field-level drop-off · marginal cost · form funnel