The average e-commerce checkout in Baymard Institute's 2024 review runs 5.1 steps and contains 11.3 form fields, against the eight fields most sites actually need. That gap is real. It is also the reason most form work starts in the wrong place: someone counts fields, deletes three, ships it, and the conversion rate does not move.
Field count is a symptom of form friction, not the diagnosis. This is a scoring rubric for anyone who owns a checkout, a booking flow, an appointment form or a public-service application and needs to know which part to fix first. Fifteen criteria, three points each, forty-five available. You finish with a score, a ranked fix list, and a threshold telling you whether to optimise the form you have or rebuild its field inventory.
Why "just cut fields" is the wrong first move
The number everybody quotes, that cutting a form from 11 fields to 4 lifted conversion 120%, comes from a 2008 case study by a Chicago web agency. Read the PDF and the sample is 184 form views producing 10 submissions against 219 views producing 26, on a B2B "Contact Us" form, across two non-adjacent windows nearly two decades ago. Ten conversions against twenty-six is an anecdote. It was never an e-commerce result and it cannot carry the weight the industry puts on it.
The Expedia story, an optional "Company" field costing roughly $12 million a year, traces to a 2010 conference talk reported by a publication that no longer exists, so every surviving write-up is second-hand. Treat it as a story about ambiguous labelling, not a coefficient.
Controlled tests cut both ways. CXL documents a nine-field form that lost 14% when three fields were removed, then gained 19.2% when all nine were kept and the label copy was rewritten. The field that matters is rarely the one easiest to delete.
The rubric: 15 criteria, 45 points
Score each criterion 0 to 3, on what the form actually does on a real device rather than what the design file says.
| Score | Meaning |
|---|---|
| 0 | Absent, or present in a form that actively harms the user |
| 1 | Present but inconsistent across fields or steps |
| 2 | Correct on the happy path |
| 3 | Correct on every path, including error states, B2B orders and non-domestic addresses |
Group A — Field inventory (12 points)
| # | Criterion | What a 3 looks like |
|---|---|---|
| 1 | Named downstream use | Every field maps to a system, legal obligation or fulfilment step you can name. "Marketing might want it" scores 0. |
| 2 | Deferred collection | Anything not required to take the order is asked after payment confirmation, not before. |
| 3 | Conditional display | Tax-ID, company and region-specific fields appear only when the user's selections trigger them. |
| 4 | No redundant entry | Nothing is requested twice across the flow, per WCAG 2.2 SC 3.3.7. Billing address defaults to shipping. |
Group B — Input mechanics (12 points)
| # | Criterion | What a 3 looks like |
|---|---|---|
| 5 | Autofill tokens | Every field carries a valid WHATWG autofill token: given-name, address-line1, postal-code, tel, cc-number. This is also the sufficient technique for SC 1.3.5. |
| 6 | Keyboard match | Correct type and inputmode so numeric fields open a numeric keypad on mobile. |
| 7 | Format tolerance | Postcodes, phone and card numbers and tax IDs accept spaces, punctuation and mixed case, then normalise server-side. This is explicit in the GOV.UK address pattern. |
| 8 | Paste survives | Nothing blocks paste. One-time codes carry autocomplete="one-time-code"; password managers are not defeated by custom inputs. |
Group C — Labels and guidance (9 points)
| # | Criterion | What a 3 looks like |
|---|---|---|
| 9 | Persistent labels | Visible labels that survive focus and input. Placeholder-only fields score 0 against SC 3.3.2. |
| 10 | Just-in-time help | Ambiguous fields such as "Address line 2", a tax identifier or a security code carry inline explanation, not a tooltip to hunt for. |
| 11 | Accessible name matches | The programmatic label contains the visible label text, and every input is properly associated with it. |
Group D — Errors and recovery (6 points)
| # | Criterion | What a 3 looks like |
|---|---|---|
| 12 | Errors that repair | Errors name the problem and the fix, per SC 3.3.3, with an error summary at the top that takes focus and links to each failing field. |
| 13 | State survives failure | Nothing is lost on validation error, back navigation, session timeout or a declined payment. |
Group E — Identity gates (6 points)
| # | Criterion | What a 3 looks like |
|---|---|---|
| 14 | Guest path | Purchase completes without an account. Account creation is offered on the confirmation screen using data already given. Forced registration is the reason 18% of abandoners give for leaving. |
| 15 | No cognitive test | Authentication does not require remembering or solving anything without an alternative, per SC 3.3.8. |
Reading the score
Two rules. First, the band:
- 38–45. The form is not your bottleneck. Go upstream: extra costs are cited by 40% of abandoners and slow delivery by 20%, against 17% for checkout complexity.
- 28–37. Fix every 0 and 1 in group order. Expect measurable but bounded movement.
- 18–27. Rebuild the field inventory before A/B testing anything cosmetic.
- Below 18. Do not test. Rebuild, then set a new baseline.
Second, the override: any Group D or E criterion scoring 0 goes first, whatever the total says. A form that loses data on a declined payment is not a candidate for optimisation.
Worked example
A composite pattern common in Turkish e-commerce; the scores are illustrative, not a measured client audit. A three-step checkout asks for name, surname, email, phone, address, district, city, postcode, national ID and an account password, then repeats name and address on the billing step.
Group A scores 4 of 12: unconditional ID field (1), billing repeats shipping (0), nothing deferred (1), every field nameable (2). Group B scores 5 — no autofill tokens on the address block, postcode rejects spaces. Group C scores 4 on placeholder-only labels. Group D scores 2: inline errors, but the password clears on failure. Group E scores 1: registration is mandatory. Total 16 of 45 with two Group D/E zeros, so the instruction is not "remove three fields" but "make registration optional and stop clearing state".
[INTERNAL DATA NEEDED: distribution of first-audit scores across Switas checkout audits, and the median score change after remediation.]
Three fields that need local judgement
National ID in Turkey. Turkish sellers need identification data for e-Arşiv invoicing, so many checkouts make the T.C. Kimlik Numarası mandatory for everyone. It does not have to be. Under the 2026 rules from Tax Procedure Law General Communiqué No. 588, invoices to buyers who are not registered taxpayers may carry the placeholder value 11111111111, and the seller is not responsible for verifying customer-supplied identity data. A mandatory eleven-digit field on every order is a product decision, not a legal requirement. Criterion 3 exists for exactly this.
Address structure. Turkish addresses run province, district, neighbourhood, street, building, apartment — six levels against the four address-level tokens the HTML spec provides. Pick a mapping, document it and keep it stable, because changing it later silently breaks every returning customer's autofill. GOV.UK's rule applies: use separate fields only when you know which countries you serve and can find one format covering them all.
Company and VAT fields. B2B identifiers belong behind a "this is a business order" toggle. Shown to every buyer, they are the Expedia problem in new clothes.
Where this rubric breaks down
It scores form mechanics, and those are a minority of checkout abandonment. Baymard puts the documented average abandonment rate at 70.22% across 50 studies, with extra costs, delivery speed and payment trust all ranking above checkout complexity. A 45/45 form on a site with surprise shipping costs still loses the order.
It is a heuristic instrument, not evidence: it tells you where friction is likely, not what each fix returns. Use it to build a ranked hypothesis list, then size and test the top items properly. It does not cover payment method mix, delivery options or trust signals. For EU sellers, e-commerce services sit inside the scope of the European Accessibility Act, so the WCAG criteria above are a compliance floor.
It also assumes a transactional form. Lead-generation forms trade volume for lead quality deliberately, and a high score there can mean you are collecting too little.
FAQ
How long does scoring a form take?
Two to three hours for a single flow, testing on a real phone and a screen reader and including one deliberate failed payment. Most of that goes into Group D, because nobody has looked at the recovery states.
Who should do the scoring?
One person scores, a second reviews every 0 and 1. Single-scorer audits drift toward 2s, and the form's designer should not score it alone.
Is fewer fields ever the right answer?
Often, but as a consequence of criteria 1 to 3 rather than a target. No named downstream use, delete it. A use that applies only to some orders, make it conditional.
Can I skip the accessibility criteria if my market is outside the EU?
You can, and you lose the conversion benefit. Autofill tokens, persistent labels and pasteable one-time codes were usability fixes before they were compliance items.
What if my form scores well but conversion is still poor?
That is the 38–45 band, and it is useful: the problem is upstream. Look at cost transparency, delivery promise and payment options before touching the interface again.
Should I weight the criteria?
Not on the first pass. Equal weights keep scoring honest and let the Group D/E override do the prioritisation. Once you have outcome data from several flows, weight against your own results.
Does this work for booking and appointment forms?
Yes. Booking, clinic appointment and public-service application forms fail the same criteria, usually on Groups C and D. Only criterion 14 needs rewording, since some services legitimately require an identified account.
How often should we re-score?
After any release that touches fields, otherwise twice a year. Scores decay quietly, usually through conditional fields becoming unconditional.
Next step
Score your own checkout against the fifteen criteria this week and fix the Group D and E zeros first. If you would rather have it audited independently, with findings ranked and sized, that is what our UX audit engagement delivers.
Sources
- Baymard Institute — Checkout Optimization: Minimize Form Fields (26 June 2024)
- Baymard Institute — Cart Abandonment Rate statistics (updated 22 September 2025)
- Imaginary Landscape — Contact form case study (PDF, 2008)
- CXL — Should You Really Reduce Form Fields?
- Usability Counts — report of the Expedia form field talk (2010)
- W3C — Understanding SC 3.3.7 Redundant Entry
- W3C — Understanding SC 1.3.5 Identify Input Purpose
- W3C — Understanding SC 3.3.2 Labels or Instructions
- W3C — Understanding SC 3.3.3 Error Suggestion
- W3C — Understanding SC 3.3.8 Accessible Authentication (Minimum)
- WHATWG HTML Standard — Autofill field names
- GOV.UK Design System — Addresses pattern
- GOV.UK Design System — Error summary component
- European Commission — European Accessibility Act
- MuhasebeTR — 2026 invoicing limits under Tax Procedure Law General Communiqué No. 588
