The Heuristic Walkthrough: Reading Pages Like a Stranger

A heuristic walkthrough inspects a page against a short list of proven usability principles - Nielsen Norman Group's ten heuristics are the canonical set - and records where the page violates each one, with the screen evidence attached. For conversion work the load-bearing principles are: the system keeps the visitor informed of state (cart, progress, errors); the language is the visitor's, not the catalogue's; errors are prevented and forgiven; the design is consistent; and help appears where the task gets hard. The walkthrough finds comprehension and mechanics problems cheaply. It cannot find offer problems - no heuristic indicts a price - which is why it complements measurement rather than replacing it.

Visibility of status: does the visitor always know what the cart holds, what step they are on, what just happened. Match with the real world: product language, delivery language, the visitor's words. User control: back works, edits undo, nothing traps. Consistency: the same action looks the same everywhere; terms do not change names mid-funnel. Error prevention: the design stops the postcode mistake rather than punishing it. Recognition over recall: options are visible, not remembered from three screens ago. Flexibility: returning visitors get the shortcuts (saved details, reordering). Minimalist design: every element earns its attention. Error recovery: messages that repair. Help at the point of need. Score each 0 (violates) to 2 (honours) per page, with evidence.

Walk the page as the visitor: phone in hand, the task the page exists for, from the arrival source's promise. Screenshot every violation at the moment it occurs; a finding without its screen becomes an opinion by Friday. Score independently before discussing if two people walk - the disagreement between scorers is itself a finding about the page's ambiguity. And keep the walkthrough's scope to what heuristics can see: comprehension, mechanics, states, language. The moment the notes say 'this offer feels weak', the method has been left; log it as a lead for the data to try, not as a walkthrough finding.

Not all violations are equal. Rank findings by what they do to the page's task: a violation that blocks the task (an unrecoverable error state) outranks one that slows it, which outranks one that merely looks inconsistent. Multiply by the page's traffic for the fix list's sake, and record severity on the finding - audit lists fail when a cosmetic inconsistency and a checkout-blocking error sit in the same typeface. The walkthrough's output joins the same fix list as every other evidence stream, scored by the same prioritisation method.

Evaluators find the problems they are primed to find; two walkthroughs of the same page overlap imperfectly, and both miss what neither thought to look for. Heuristics age: the set was written for software generally, and store-specific failure modes (total-cost opacity, variant ambiguity) need the bench's own doubt lists beside it. Treat walkthrough findings as indicated, not proven: where a repair is cheap, fix on the walkthrough's word; where it is expensive, let recordings or a test try the finding first. Judgement methods are the audit's scouts - fast, cheap, fallible - and scouts are not the army.

Run walkthroughs on the highest-traffic templates after any redesign, theme change or new page type launch, and once a year on the core set regardless. New pages benefit most: the walkthrough is the cold read with a checklist, catching the comprehension failures their makers are blind to because they know what the page means. Keep the scores per template over time; a template whose scores decay release by release is telling you about your change process, not about the page.

Beside the canonical ten, this bench keeps five store-specific checks on the same card, because the generic set underweights the failures that cost orders: total cost knowable before checkout; delivery date or honest range visible where the decision happens; variant state (chosen, available, priced) always visible; the deal unchanged at every click from the visitor's side; and help reachable at the moment of the hardest question on the page. Score them with the ten, on the same walk. A page that honours Nielsen and fails the five is a usable page that does not sell; a page that passes the five and trips the ten sells awkwardly and leaks goodwill. The audit wants both columns green, and the fix list takes the reds in task-damage order.