Call-to-Action Placement Test¶

Placement experiment — instantiates Latent Affordance Surfacing

Runs controlled variants of where and how prominently a cue for an available action appears, then keeps the version that most raises discovery without drowning the surrounding surface in noise.

Once you know an affordance is hidden and you have a cue for it, two questions remain that opinion answers badly: where on the surface it should sit and how loud it should be. Call-to-Action Placement Test answers both empirically. It exposes randomized variants — the cue below the fold versus beside the relevant field versus a prominent chip at the top; understated versus bold — and measures which variant most raises the rate at which the intended action is discovered, while watching guardrail metrics so a louder cue doesn't degrade the rest of the surface. Its defining move is that placement and salience become measured questions with a noise budget, not a designer's guess: the winner is the version that maximizes discovery up to the point where added prominence starts costing more in clutter and annoyance than it returns in finds.

Example¶

A nonprofit's donation page offers a recurring-gift option — "make it monthly" — that genuinely exists but sits below the fold, so few donors ever see it. The team runs a Call-to-Action Placement Test. Variant A (control) leaves it below the fold; B places a toggle directly beside the amount selector; C floats a bold chip at the top of the form. The primary metric is the share of donors who select the recurring option; the guardrails are overall completion rate and a later opt-out rate, so a pushier cue can't quietly cannibalize one-time gifts or feel like a bait-and-switch.

Variant B wins: recurring selection rises by roughly a third while completion holds flat and opt-outs don't move. The bolder C lifts selection slightly more but nicks completion — past B's prominence, the extra salience buys noise, not discovery. They ship B as the placement-and-timing rule and hold salience at the point where the discovery curve flattens, before the guardrails complain.

How it works¶

Define a discovery metric and its guardrails. Pick the rate that means "the affordance was found and used," plus the surrounding metrics a louder cue could harm (completion, opt-outs, task time).
Vary placement, prominence, and timing — not the offer. Randomize where the cue sits, how salient it is, and when it appears, holding the underlying action constant so the test isolates surfacing.
Find the salience knee. Read where added prominence stops raising discovery and starts tripping a guardrail; that inflection, not the loudest variant, is the calibrated setting.
Ship the rule and keep measuring. Adopt the winning placement/timing as a rule and leave the loop running, since the best surface shifts as the page and its traffic change.

The test is distinguished by being empirical and comparative: it decides placement and loudness by evidence against a metric, rather than authoring the cue itself.

Tuning parameters¶

Primary metric — immediate find-rate versus downstream completion or retention. Optimizing the click is easy and can reward a cue that wins attention but not genuine value.
Variant space — position, size, color, motion, timing — how many dimensions you vary at once. Wider spaces find better optima but dilute statistical power and invite false winners.
Salience ceiling — how loud a variant is allowed to get before a guardrail vetoes it. Set it high and you can manufacture clicks by making the cue impossible to ignore.
Traffic and duration — how much traffic and time each variant gets. More buys confidence; less ships faster but risks calling noise a result.
Segmentation — whether new and returning users are tested and tuned separately, since a cue that teaches a newcomer only nags a veteran.

When it helps, and when it misleads¶

Its strength is that it replaces placement debate with evidence, and the guardrail metric is what keeps a local click-win from poisoning the whole surface — a cue that raises one number while lowering three is correctly rejected. It is also self-correcting: the loop notices when a once-good placement decays.

Its failure modes are the ones every optimization inherits, sharpened by attention being a shared resource. It optimizes the metric it is given, so a short-term find-rate can be won at the cost of long-term trust — the boundary where surfacing shades into a dark pattern. Cranking salience to move the number, repeated across a product, breeds banner blindness: users learn to tune out the loud region entirely, and the next genuine cue placed there is invisible from birth.^[1] And with enough variants the test will hand you a "winner" that is really noise. The discipline that keeps it honest is a protective guardrail metric, a salience ceiling set before the test, and stopping rules fixed in advance rather than after peeking.

How it implements the components¶

Call-to-Action Placement Test realizes the placement-and-calibration side of the archetype — deciding, by measurement, where and how loudly a cue should appear:

surface_placement_and_timing_rule — its primary output: the evidence-backed rule for where on the surface, and when in the flow, the cue belongs.
salience_and_noise_calibration — it finds the prominence that maximizes discovery before added loudness turns into clutter, setting the noise budget explicitly.
discovery_measurement_loop — the metric-plus-guardrail comparison is a running discovery loop that both decides the winner and detects later decay.

The cue itself and the gap it addresses — affordance_signifier_design, latent_capability_inventory, effective_option_set_gap — come from Contextual Hint or Tooltip and Affordance Visibility Audit; this test optimizes the placement of a cue it does not design.

Instantiates: Latent Affordance Surfacing — it tunes where and how loud a surfacing cue should be.
Consumes: a cue to place — Contextual Hint or Tooltip or a plain call-to-action supplies the signifier this test positions.
Sibling mechanisms: Contextual Hint or Tooltip · First-Attempt Discovery Test · Search and Autocomplete Suggestion · Signage and Wayfinding Revision · Failed-Search and Helpdesk Query Analysis

Notes¶

The test presupposes a cue: it optimizes placement and salience but does not create the signifier being placed. Run against a weak or unclear cue, it will faithfully find the least-bad location for a signal that was never going to be understood — so a cue worth placing (from Contextual Hint or Tooltip or an audit-motivated CTA) is a precondition, not an afterthought.