What matters in automated form filling?

We’re surveying developers and product folks on automated form filling beyond browser autofill—think multi-step flows, client-side validation, and SPA/iframe quirks. We’ve run a small study across 120 public forms (auth, checkout, job apps, SaaS signups) to identify pain points and success criteria, and we’d like your input on what really matters.

What we tested (lightweight, not a full product):

- Field detection: labels/for, aria-* attributes, nearest text nodes, and simple layout proximity (bbox distance, alignment).

- Basic value policies: skip sensitive fields by default (SSN/passports), prefer vault-sourced data for PII, and respect “required” heuristics.

- Flow handling: wait for async validators, detect SPA route changes, and retry on soft failures (error text appears).

- Validation priors: regex + checksum for email/phone/postal/IBAN; minimal locale rules.

Early metrics (small sample, indicative only):

- Field association accuracy: ~86% on simple forms; drops to ~63% with shadow DOM or custom inputs.

- End-to-end completion: ~58% for single-page flows; ~34% for multi-step with conditional fields.

- False positives (filled the wrong field): ~7% overall; concentrated in tables/repeaters.