Traffic but No Sales? An Ecommerce Conversion Checklist

CONVERSION DIAGNOSIS WORKSHEET
Scope: store/market/device/period/timezone (anonymous):
Decision owner / analyst / implementation owner / deadline:
Counting unit: sessions OR users (never mix):
Entry event / cart event / checkout event / successful purchase event:
Closed sequence / indirect or direct steps / time limit:
Consent, internal traffic, bot and refund policies:
Export time / evidence location / late-data policy:
Counts: entered / cart / checkout / purchase:
Backend reconciliation exceptions:
Top exit transition / observed symptom / evidence:
Alternative explanations (traffic, stock, price, measurement):
Customer observation: task / device / exact obstacle:
One proposed intervention / mechanism / affected audience:
Assumed recovered-dropoff fraction / justification:
Current downstream completion / evidence gap:
Extra converting units scenario / contribution / implementation cost:
Do not add separate stage scenarios.
Primary experiment outcome / guardrails / sample and duration plan:
Decision: MEASUREMENT_FIX / MORE_RESEARCH / CONTROLLED_TEST
Owner / next action / stopping condition / review date:
No lift or incremental revenue is established by this worksheet.

Start with a decision, not a redesign

Use this checklist when an ecommerce or consumer-brand site attracts visitors but produces fewer purchases than expected. The job is to decide what to investigate and test next, not to blame a page from one percentage. Bring a closed-funnel export, device and channel breakdowns, backend purchase evidence and a small set of consented usability observations. A founder may hold several roles, but an analyst must own the definitions and an implementation owner must own the fix.

Do not use this method to infer why an individual abandoned, compare incompatible platforms, or forecast revenue from a handful of sessions. If buyers normally return days later, a same-session funnel describes only one part of their journey. Define a user-based observation window separately instead of silently treating return visits as failures.

Five gates before spending on a fix

1. Validate the population. Record the counting unit, dates, timezone, identity and exclusions. Confirm every later count is a subset of the earlier step completed in order. If counts come from open funnels or independent event totals, stop and rebuild.
2. Locate the transition. Calculate adjacent losses and conditional rates. Separate mobile and desktop, then major channels where the counts support useful inspection. A large loss is an investigation address, not proof of broken design.
3. Observe the task. Try the relevant journey on a real narrow screen, without making a live payment. Look for hidden delivery costs, ambiguous variants, invalid address feedback or payment errors. Use consented replays with sensitive fields masked; never infer motivation from a cursor alone.
4. Model one change. Use a stated fraction of recoverable drop-offs and retain downstream completion as an explicit assumption. Compare plausible ranges and implementation cost. Do not sum scenarios for different steps.
5. Design evidence. Write a falsifiable mechanism, a purchase-level outcome and guardrails for refunds, contribution and errors. Agree an experiment plan before looking at results. If traffic is too low, prioritize direct defect evidence and acknowledge causal uncertainty.

Worked example: the largest gap is not the best project

In a synthetic closed session funnel, 10,000 enter, 1,000 add to cart, 600 reach checkout and 300 purchase. Assume 10% of a transition's drop-offs could advance. Visit-to-cart yields 900 extra cart sessions; at the existing 30% cart-to-purchase rate that implies 270 extra purchasing sessions. Cart-to-checkout yields 40 extra checkouts and 20 extra purchasing sessions. Checkout-to-purchase yields 30 extra purchasing sessions directly.

At 30 currency units of contribution per purchasing session, those independent scenarios are 8,100, 600 and 900 before implementation cost. The first number is not evidence that a homepage redesign will produce 8,100. If a payment error is reproducible and the earlier gap reflects visitors browsing normally, fixing checkout can still be the better project. The scenario ranks arithmetic sensitivity, not likelihood of success.

Turn evidence into a small, owned test

The analyst signs off the data contract. The researcher supplies the observed obstacle and alternative explanations. Product or engineering estimates cost and verifies accessibility. The growth owner chooses one test, defines the audience and keeps the change log. Finance validates the contribution boundary. Record a review date and an explicit stop condition such as increased checkout errors; do not improvise a statistical stopping rule after launch.

Keep measurement repairs distinct from customer-facing experiments. Correcting duplicate purchase events can reduce the reported conversion rate while improving truth. Record that change before comparing periods. A successful UI release means the intended interaction works; it does not yet mean more incremental sales.

When to hold the decision

Hold when exports mix users and sessions, events arrive late, consent coverage changes, stockouts distort the cohort, or a zero denominator prevents downstream estimation. Hold a broad rollout when customer evidence contradicts the proposed mechanism. Do not interpret the 10% example as an industry benchmark or assume every recovered visitor has the same intent as an existing buyer. Those are the hypotheses to challenge.

Sources and limitations

Primary documentation checked October 18, 2026: Shopify behavior reports explains session-based open and closed funnels; Google Analytics funnel exploration explains user-based sequence counting. Publication dates are unknown for these living documents. These systems do not automatically share a denominator. The worksheet and scenario method are original editorial synthesis, not vendor-approved benchmarks. Examples are synthetic, not client results.