Where you are
The flow simulator (b4) treats checkout as one dial; the PDP session (b5) covered the decision room before it. This session zooms into the machine's last gate: the 16,200 sessions that begin checkout and the 9,720 that finish. In b7 you use this per-step view as the finest layer of the diagnosis drill-down.
Inside the last gate: three steps 8 min live
GA4's checkout journey report opens the gate up as a closed funnel: begin_checkout → add_shipping_info → add_payment_info → purchase, with abandonment shown below each step. Mango Lane's month: 16,200 → 14,580 (90%) → 12,150 (83.3%) → 9,720 (80%). Multiply the chain: 0.90 x 0.833 x 0.80 = 0.60 - exactly the 60% completion the metrics course reported as one number.
LiveCart abandonment vs checkout abandonment - stop mixing them4 min▶
Baymard's famous 70.22% is CART abandonment - everyone who added to cart and did not buy - and roughly 42-43% of those people were just browsing with no intent to buy that day. Mostly normal shopper behavior. The micro-funnel above measures checkout abandonment: people who started entering shipping details and quit. That 40% loss (16,200 → 9,720) is the fixable slice - these shoppers proved intent by starting to pay.
- Cart abandonment - big, partly normal, watch the trend not the level.
- Checkout abandonment - smaller, mostly actionable, diagnose per step.
Name each step's killers 10 min live
Baymard Institute's checkout-abandonment reasons (2025 survey wave, just-browsing excluded) are not one list - they land on specific steps. Mapped:
| Step | Baymard reason | Share citing it |
|---|---|---|
| Shipping step | Extra costs too high (shipping, fees, tax) | 40% |
| Delivery was too slow | 20% | |
| Could not see total order cost upfront | 12% | |
| Payment step | Did not trust the site with card information | 19% |
| Card was declined | 10% | |
| Not enough payment methods | 9% | |
| Whole flow | Site wanted an account created | 18% |
| Checkout too long or complicated | 17% | |
| Website errors or crashes | 17% | |
| Returns policy unsatisfactory | 13% |
LiveForm-field forensics4 min▶
Baymard's 2024 benchmark: the average US checkout runs 5.1 steps and shows 11.3 form fields - while an ideal checkout needs only about 8 fields. Their key finding: field count predicts friction better than step count - a 5-step flow with 8 fields beats a 3-step flow with 14. The "too long or complicated" reason (17%) is usually a field problem wearing a step costume.
Self-studyFriction vs reassurance - the measured trade3 min read▶
Shorter is not automatically better. The card-trust reason (19%) means some ADDITIONS help: security badges near the card field, a visible returns link (13% cite returns policy), an order summary that keeps the total honest (12% cite hidden totals). The discipline: every added element must answer a named Baymard reason, and every removal must target the field count. "Cleaner" is not a reason; "attacks the 18% account-wall loss" is.
Also policy beats pixels at the top of the table: extra costs (40%) and slow delivery (20%) are business decisions - free-shipping thresholds, carrier choices - not UI work. Half of checkout optimization is a pricing meeting.
Locate and price a drop ★ 15 min · everyone works
Three drills with full arithmetic - the per-step chain is small enough to run on paper.
Verify the chain. 16,200 x 0.90 = 14,580. 14,580 x 0.8333 = 12,150. 12,150 x 0.80 = 9,720. And 0.90 x 0.8333 x 0.80 = 0.60 - the chain reproduces the metrics course's completion exactly. Cross-footing is the habit; a chain that does not multiply back is a tracking bug, not a business event.
Locate a drop. This week completion falls to 54%. The per-step read: 16,200 → 14,580 (90%, held) → 12,150 (83.3%, held) → 8,748. The last step fell from 80% to 8,748 ÷ 12,150 = 72%. The leak has an address: the payment step. Suspects, straight off the Baymard map: declined cards (a provider issue?), a payment-method outage, or a trust regression (new layout burying the security cues). Nobody redesigns the shipping form today.
Price a fix. Wallets lift the payment step from 80% to 86%: orders = 16,200 x 0.90 x 0.8333 x 0.86 = 10,449. That is +729 orders, +7.5% - the same rung you saw in b4's ladder, now derived from inside the gate. At $68 AOV: about +$49,600 GMV per month on current traffic - the one-line business case.
Why per-step beats per-gate in a war room. "Completion fell" starts a blame meeting between UX, payments, and shipping. "The payment step fell from 80% to 72% at 2pm Tuesday, shipping step flat" ends it - payments owns the next hour, everyone else goes back to work. The micro-funnel is an argument-compression device.
Try it yourself - this week ◐ 20-30 min total
- Pull your per-step checkout funnel (GA4's checkout journey report or your tool's equivalent). Verify the steps multiply back to your reported completion rate.
- Map each step's biggest Baymard reason for YOUR store - which of the top three (extra costs 40%, delivery 20%, trust 19%) applies, and at which step?
- Do your own checkout on your phone; count the form fields shown. Compare against 11.3 average / ~8 ideal.
- Price one fix: pick a step, a plausible delta, and compute the monthly order and GMV value.
Sources this session draws on
The checkout is the best-researched screen in ecommerce; this session leans on the primary work. This page draws on:
Three questions before you go 🎯 ◐ 90 seconds
1 · Completion falls from 60% to 54%; shipping and payment-entry steps held at 90% and 83.3%. Where is the leak?
The chain multiplies: if two steps held, the third must account for the drop. 0.54 ÷ (0.90 x 0.8333) = 0.72. Per-step data turns "completion fell" into an address.
2 · A card-decline spike at the payment provider hits which reason and step?
Declines bite between entering payment details and completing purchase. A sudden fall isolated to that step with UX unchanged points at the provider before the pixels.
3 · Baymard's field research says the better friction predictor is...
Fields, not steps, are where shoppers give up. "Too long or complicated" (17%) is usually excess fields; wallets attack the count and the trust reason at once.