Before You Buy More Traffic: Diagnose the Ecommerce Journey

Traffic multiplies the journey you already have

When growth slows, more traffic is an attractive prescription. It is measurable, purchasable, and easier to brief than a cross-functional repair. A campaign can start next week; fixing an unclear proposition, fragile checkout, or unreliable fulfilment can involve several teams.

But acquisition does not correct the journey. It sends more people into it. If the first material constraint is a confusing offer, an unusable product selector, an unexpected delivery term, or a failed payment recovery, paid traffic increases the number of customers exposed to that constraint. The campaign may even look efficient inside its own dashboard while margin, support load, refunds, and repeat purchase deteriorate elsewhere.

The better question is not “How do we increase conversion?” It is “Where does a suitable customer first lose the confidence or ability to continue, and what evidence would distinguish that constraint from the others?”

That is a diagnostic question. It needs a staged model, several kinds of evidence, and the discipline to repair the first consequential loss before optimising the end of the funnel.

The Friction-to-Spend Gate

The Friction-to-Spend Gate divides an ecommerce journey into six conditions: message, findability, confidence, transaction, fulfilment, and return. More acquisition spend is justified only when the earlier gates are sufficiently healthy for the decision at hand.

This is not a claim that every customer moves through a neat linear funnel. People compare tabs, leave, ask somebody, return on another device, encounter an advert, and buy later. The gates describe commercial conditions, not a single clickstream.

Message

Can the right customer quickly understand what is offered, for whom, and why it is a credible choice? Inspect campaign promise, landing headline, product language, price framing, and the match between the advert and destination. A high bounce rate may indicate mismatch, but it may also reflect accidental visits, tracking, slow performance, or a customer who found an answer immediately. Read behaviour alongside the page itself.

Ask five people from the intended audience to explain the offer after a brief exposure. Do not ask whether they “like” the page. Ask what they think is being sold, what makes it relevant, what it costs, and what they would do next. Divergent answers reveal an interpretation problem that another creative variation may simply amplify.

Findability

Can customers locate the right product, variant, information, and next action? Search terms with no useful result, filters that eliminate valid options, ambiguous category labels, and inaccessible controls all belong here. Mobile navigation and keyboard operation often reveal constraints hidden in a desktop review.

WCAG 2.2 supplies testable criteria for many interaction fundamentals, including focus, labels, target size, error identification, and reflow. Meeting a criterion does not prove that the taxonomy makes sense, but failing basic interaction can prevent customers from testing the taxonomy at all.

Confidence

Can the customer form a reliable expectation about the product and the commitment? Look for specifications, imagery, sizing, availability, delivery timing, total price, returns, warranty, reviews, and evidence behind claims. The issue is not “add more trust badges.” It is whether the questions that govern this purchase are answered where they arise.

Analyse pre-purchase contacts and on-site search language. Customers often describe uncertainty more precisely than a generic exit survey. Group questions by decision: suitability, risk, timing, price, compatibility, legitimacy. Then inspect whether the page provides a usable answer or merely adds reassurance without proof.

Transaction

Can the customer complete, verify, and correct the order? Baymard’s 2024 checkout research reports recurring usability problems from a substantial commercial research programme. Its findings are useful hypotheses, not universal facts about your store. Test guest checkout, account creation, address entry, promotion codes, delivery selection, payment, authentication, errors, review, and confirmation with your own conditions.

Observe what happens after failure. Does a declined payment preserve the basket and entered data? Does an error identify what needs correction? Can the customer change the delivery method without starting again? A nominal checkout conversion rate can conceal repeated attempts, duplicate orders, or support-assisted completion.

Fulfilment

Does the operation deliver the promise that won the order? Join web data with stock substitutions, dispatch delay, carrier exceptions, damaged goods, cancellations, and “where is my order?” contacts. A campaign may acquire profitable-looking orders that the operation cannot fulfil at the promised cost or speed.

Segment by product, delivery method, geography, and campaign where volume permits. Avoid treating small groups as precise. The objective is to find patterns worth investigating, not to manufacture certainty from sparse cells.

Return

Can a customer resolve disappointment, return an item, obtain a refund, and choose the brand again? Returns are not automatically failure: they can be a legitimate part of categories where fit is uncertain. The diagnostic issue is avoidable mismatch, unexpected effort, recovery time, and the effect on contribution margin and trust.

Read return reasons cautiously. Fixed dropdown categories often reflect the options a system offered, not the customer’s full explanation. Pair them with support transcripts, product reviews, repeat behaviour, and direct observation of the returns journey.

Combine behavioural and operational evidence

No single dashboard locates the constraint. Build a compact evidence table for each gate.

• Behavioural evidence: entrances, progression, search, filters, errors, exits, repeat attempts, device and browser patterns.
• Observed evidence: task-based usability sessions, support shadowing, session replay used with appropriate privacy controls, and direct walkthroughs.
• Customer language: search queries, questions, complaints, reviews, cancellation notes, and research interviews.
• Operational evidence: inventory, dispatch, delivery exceptions, payment failure, refunds, contact demand, and handling time.
• Commercial evidence: contribution margin, promotion cost, media cost, return cost, repeat purchase, and cash timing.
• Technical evidence: availability, error logs, accessibility checks, and field performance.

Google’s Web Vitals guidance distinguishes field data from laboratory measurement. Field data shows experienced performance across real visits where coverage exists; lab tests support diagnosis under controlled conditions. Neither tells you whether a customer believed the offer. Treat performance as one possible constraint in a larger journey.

Write findings as testable statements. “Checkout is bad” is too vague. “On mobile, customers who select collection cannot discover how to change location after the basket, and support contacts show repeated requests to restart the order” connects a condition, an audience, a behaviour, and corroborating evidence.

Use the friction-to-spend matrix

For each gate, score two dimensions using bounded language: customer consequence and evidence confidence.

Customer consequence asks what happens if the suspected friction is real. Does it block purchase, increase material risk, cause recoverable effort, or create a minor delay? Evidence confidence asks whether the conclusion rests on a consistent combination of observed, behavioural, operational, and customer evidence, or on one ambiguous metric.

The matrix creates four action types:

1. High consequence, strong evidence: repair before increasing spend.
2. High consequence, weak evidence: investigate quickly with targeted observation or instrumentation.
3. Low consequence, strong evidence: schedule a bounded improvement if its cost is justified.
4. Low consequence, weak evidence: do not let it displace material work.

This is not arithmetic that removes judgement. A safety, legal, privacy, or accessibility issue may require action regardless of measured frequency. A new store may lack enough behavioural data and need more direct research. Record why the priority was chosen.

Sequence tests around the first constraint

Optimisation programmes often scatter small tests across the site because each is easy to launch. That produces activity without resolving the main loss. Instead, identify the earliest gate where a suitable customer experiences material friction.

If message is unclear, test proposition and evidence before button colour. If findability fails, repair taxonomy or product selection before checkout urgency. If fulfilment breaks the promise, control the promise and operation before adding traffic. Later improvements cannot compensate reliably for an earlier inability to proceed.

A useful repair test has four parts:

• A decision: what will the result change?
• A mechanism: why should this repair change customer behaviour or operational performance?
• A primary measure: what outcome is closest to that mechanism?
• Guardrails: what must not worsen, such as margin, accessibility, support load, or return rate?

Use a qualitative prototype when the question is comprehension or interaction. Use controlled experiments when traffic, implementation, and decision value justify them. Use before-and-after operational review when randomisation is infeasible, while naming the weaker causal inference.

Diagnose attribution before trusting the acquisition answer

Campaign reports allocate observed events according to platform rules. They do not automatically reveal what would have happened without the spend. A customer may have encountered several channels, already intended to buy, or converted after an operational change.

For the traffic decision, compare:

• platform-attributed revenue;
• analytics with a documented attribution model;
• new versus returning customer mix;
• contribution after discounts, media, fulfilment, and returns;
• geographic or audience holdouts where feasible;
• longer-term repeat and support outcomes.

The aim is not to wait for perfect causality. It is to prevent a precise-looking ROAS from overruling obvious journey evidence or poor economics. State which conclusion the evidence supports: tracking is functioning, the campaign reached suitable people, observed revenue was associated with exposure, or an experiment estimated incremental effect. Those are different claims.

Know when more traffic is the right move

Acquisition can be the constraint. A clear, usable, trustworthy journey with reliable fulfilment may simply reach too few suitable customers. The diagnostic should be capable of reaching that conclusion rather than treating internal repair as virtuous by default.

Increase spend when:

• the proposition is understood by the intended audience;
• customers can find and evaluate the appropriate offer;
• transaction blockers and serious accessibility defects are controlled;
• fulfilment and recovery match the promise;
• unit economics survive realistic media and return costs;
• measurement is adequate for the size of the decision;
• the operation can absorb the expected demand.

Increase in stages. Predefine a threshold that would pause or reverse the change, and watch downstream measures rather than only click and checkout events. New traffic can differ from the audience used to diagnose the original journey.

A seven-day diagnostic before the next campaign

Day one: choose one campaign-to-recovery journey and write the customer decision at each gate. Day two: inspect the experience across mobile, desktop, keyboard, slow connection, and relevant account states. Day three: combine analytics, search, errors, support, fulfilment, returns, and margin. Day four: observe several suitable customers attempting the task. Day five: identify the first high-consequence constraint and its likely mechanism. Day six: design the smallest repair and its evidence. Day seven: assign the owner, release condition, and date for review.

This does not replace deeper research. It prevents a media deadline from becoming the only force setting priorities.

Traffic is a multiplier. Diagnose what it will multiply: a coherent promise and dependable journey, or uncertainty and avoidable failure. Then place the next euro where it removes the binding constraint rather than where the dashboard makes action easiest.