The Checkout & Payments Playbook — The Last Meter of Conversion

Every playbook in this series feeds a funnel that ends at one step: payment. Yet in most organizations, checkout analytics is orphaned — product owns the UI, a payments team owns the PSP integration, finance owns the fees, and nobody owns the composite question: of the customers who tried to give us money, how many succeeded, and what did each failure cost? This playbook is that composite view.

1. The Most Expensive Funnel Step

Industry-wide, 60–70% of carts never become orders. Some of that is window shopping — but the share that fails after checkout begins is intent lost to friction and failure, the most expensive loss in the funnel because every upstream cost has already been paid. In emerging markets the stakes double: online payment success rates of 85–95% mean 1 in 10–20 willing payers bounce at the money step, and COD — often 40–70% of orders in MENA and South Asia — brings its own failure mode (RTO) that fails after fulfillment costs are spent.

The framing for leadership: a 1-point improvement in payment success rate is roughly a 1-point GMV increase, at zero marketing cost, permanently. Price that number for your business and this playbook funds itself.

2. Instrumenting the Checkout Funnel

You cannot fix what you log as one event. The minimum event schema:

checkout_started
  → address_completed        # drop here = address friction / delivery coverage
  → shipping_selected        # drop here = delivery cost/speed shock
  → payment_method_selected  # mix shifts are a leading indicator
  → payment_attempted        # the denominator for success rate
  → payment_succeeded | payment_failed(reason_code)
  → order_confirmedevent schema

Read it with three cuts, in this order:

• Step conversion × device: checkout friction is 2–3x worse on mobile web; an aggregate hides it (the mix discipline from the GMV runbook applies here verbatim).
• Step conversion × basket profile: where do high-value baskets stall? Drop-off at shipping selection on heavy items is a delivery-pricing problem, not a UX problem.
• New vs returning: returning customers with saved details should approach ~90%+ checkout completion; if they don't, something specific broke (expired cards, address changes, re-authentication friction).

3. Payment Success Rate — The Metric With Ten Causes

Success rate = successful payments ÷ payment attempts — but the aggregate is nearly useless. The diagnostic cuts:

Two structural upgrades that consistently pay: network tokenization (tokenized credentials out-approve manually entered PANs) and smart routing / cascading between PSPs — retrying a "do not honor" through a second acquirer recovers a meaningful slice of declines. If you process at scale through one PSP with no fallback, that's both a success-rate cost and a single point of failure; the case for a second integration writes itself from your own decline data.

4. Failed-Payment Recovery

A failed payment is a hot lead, not a lost order — the customer was literally mid-purchase:

• In-session: instant, specific feedback ("this card was declined — try another method") with the method sheet re-opened and alternatives pre-ranked. Vague "payment failed" errors are conversion arson.
• Post-session: the abandoned-payment message (push/WhatsApp/email within the hour, cart intact, one-tap resume). Measure it with a holdout like any CRM campaign (promo playbook rules apply — don't attach a discount by default, or you'll teach customers that failing a payment earns 10% off).
• Retry hygiene for stored-credential payments: declined subscription/saved-card retries succeed at meaningfully different rates by timing (payday effects are real). Schedule retries, cap attempts, and never hammer an issuer — excessive retries degrade your merchant standing and future approval rates.

5. COD Economics — The Full Cost of Cash

COD is simultaneously a conversion asset (it unlocks customers who won't or can't pay online) and the most expensive payment method you offer. The full unit economics, which almost nobody assembles in one place:

Put a number on it: cost per delivered COD order vs cost per delivered prepaid order, all four components included. The gap — typically several percent of order value — is your budget for converting customers to prepaid, which is the next section.

6. Converting COD to Prepaid — With Honest Math

The lever stack, cheapest first:

1. Default and ordering: prepaid methods listed first, COD available but not default, for customers with good payment history. Defaults move share meaningfully at zero cost.
2. Trust signals: COD preference is substantially a trust problem — refund-speed guarantees displayed at the method step, delivery-experience ratings. Cheap, compounding.
3. Targeted prepaid incentives: a small discount or free-shipping upgrade for paying online, offered only where it changes behavior. The uplift logic from the promo playbook applies exactly: paying a 2% incentive to a customer who would have prepaid anyway is pure burn. Score COD-preference, target the marginal segment, measure with holdouts.
4. Risk-tiered COD access: from the RTO model — high-risk orders get partial-deposit COD or prepaid-only; good customers keep frictionless COD. This is a margin decision dressed as a payments policy: you're pricing the option to refuse.

The honest math for the incentive case: incentive cost × takers vs (COD cost gap + avoided-RTO margin) × incremental conversions only. Run it with a holdout; the published range — every 10% of COD converted to prepaid cutting overall RTO 2–4 points — is a reasonable prior, but your mix decides.

7. The Risk-vs-Conversion Trade-Off

Fraud rules and payment friction trade against conversion, and the two teams optimizing each side rarely share a metric. The unifying view (same cost logic as the fraud section of the ML playbook):

Total cost = fraud losses + (false-positive blocks × order margin × LTV multiplier) + friction-driven abandonment — minimize the sum, not any single term.

• Measure the insult rate (legitimate customers blocked or stepped-up), not just the fraud rate. A rule that stops 50k of fraud by blocking 500k of good orders is a loss wearing a win's clothes.
• Apply 3DS/step-up selectively by risk score, not blanket — full-population 3DS in markets where it's optional trades a few basis points of fraud for points of conversion.
• Review the trade-off jointly, monthly: payments, fraud, and commercial in one room with one P&L view. The optimal operating point moves with mix, season, and fraud waves.

8. BNPL — Reading the Claims Skeptically

BNPL providers claim 20–40% AOV lifts. Before repeating that in a business case, apply the series' standing discipline: selection is not causation. Customers choosing BNPL for big baskets would often have bought anyway with a card — the honest questions are: how much volume is incremental (new customers or bigger baskets that wouldn't exist), does the provider fee (typically several % of transaction) beat the margin on that incremental volume alone, and what do approval rates do to your funnel at high amounts? Where BNPL clearly earns its fee: high-AOV categories with affordability constraints. Where it usually doesn't: low-AOV baskets that would have converted anyway. Segment-level rollout with holdout regions/categories answers it in a quarter.

9. The Dashboard & Operating Rhythm

Payment success degradations are the purest example of the "alert on the leaves" principle from the GMV runbook: a 2-point issuer problem at 11am is a paged incident; discovered in Friday's GMV review, it's a week of lost revenue and an archaeology project.

Sources & Further Reading:
Baymard: Cart Abandonment Research • RTO in COD E-Commerce • Stripe: Optimizing Authorization Rates