How to evaluate a PDF form-filling API: 9 criteria that actually matter
Every PDF API demo looks the same: a clean sample form, a JSON payload, a filled PDF in under a second. The products differ in the parts the demo skips — what happens with a 40-field government form, a revised template, or month three of production traffic. (We make SimplyFill, so read with that in mind — but every criterion here is checkable yourself on any vendor, including us.)
Here are the nine criteria that separate the options, with the question that exposes each one.
1. Who does the field mapping?
The single biggest cost difference between products is who figures out that
f1_07[0] means date of birth. There are three answers on the market: you
do it in a visual editor, clicking field by field; AI proposes it from
the form's visible labels and you review; or you do it in code, maintaining
a lookup table. For one small form they're equivalent. For many forms, or
40-field institutional ones, the setup model dominates the total cost — the
arithmetic is in
how a PDF mapping API cuts weeks off document automation.
Ask: "Upload a real 40-field form in the trial. How long until every field has a name your code can use — and how much of that was your time?"
2. AcroForm and XFA, really
Most vendors say they support "fillable PDFs," meaning AcroForm. A large share of government and institutional forms are XFA — a different technology that much PDF tooling can't fill at all. If your roadmap touches government paperwork, this is a binary gate: the vendor either handles XFA (for example by flattening it to AcroForm at upload) or your project stalls in week three.
Ask: "Here's an XFA form. Upload it. What happens?"
3. Output integrity
A filled PDF that re-renders wrong is worse than no PDF: unflattened fields get mangled by e-sign tools, fonts substitute differently across viewers, regulated forms must stay visually identical to the official edition. Look for flattened, sign-ready output by default, and verify it survives a round trip through your actual e-sign vendor and a phone browser.
Ask: "Is output flattened by default? Show me the same output in Acrobat, Preview, and mobile Safari."
4. What happens when the form is revised?
Forms change on the issuer's schedule, and revisions can silently renumber internal fields — the failure mode is wrong data in the wrong box, not an error message. The product question is whether the mapping layer is versioned and re-derivable: can you upload the new edition, see a reviewable diff against your existing field names, and promote it deliberately?
Ask: "Walk me through a form-revision update. Where would a silent mis-mapping be caught?"
5. Environments and promotion
A template edit should not be able to change production output the moment someone saves it. Mature pipelines need dev/staging/production separation for templates and mappings, with explicit promotion — the same discipline you apply to code.
Ask: "Can a teammate edit a template without affecting production? What's the promotion step?"
6. One document or a packet?
Real events rarely produce one PDF. An onboarding, application, or claim typically fans out into several documents from the same payload. If the API only fills one template per call, that orchestration — N calls, partial-failure handling, grouping — becomes your code.
Ask: "One payload needs to produce five filled PDFs as a unit. How many API calls, and what happens if the third one fails?"
7. Sync, async, and webhooks
Single documents should return in around a second, synchronously. Bulk runs shouldn't hold an HTTP connection for minutes — you want async jobs with webhooks on completion, and those webhooks should be signed (HMAC) so your receiver can verify them.
Ask: "Generate 10,000 documents. What does my integration look like, and how are completion webhooks authenticated?"
8. Compliance posture
If documents carry PII or PHI, the vendor's controls become part of your audit story: SOC 2 Type II (ask for the report, not the badge), a BAA if HIPAA applies, data residency options if you serve the EU, and clear retention controls on generated files.
Ask: "Can I see the SOC 2 report under NDA? Do you sign BAAs, and what are the data-residency options?"
9. A pricing model that fits your volume curve
Per-document pricing is friendly at low volume and punishing at scale; plan tiers are predictable but can carry unused headroom early. Don't evaluate the sticker price — evaluate your curve: pilot volume, expected production volume, and the 10× case if the workflow succeeds. As of June 2026, the market spans pay-as-you-go (around $0.05–$0.25 per document at various tiers) and subscription plans; we compared two specific vendors' models in SimplyFill vs DocSpring and SimplyFill vs Anvil.
Ask: "Price my realistic year-two month, not my pilot."
Run the evaluation in an afternoon
The criteria sort into a usable order: gates first (2, 3, 8 — format support, output integrity, compliance: a miss is disqualifying), then cost drivers (1, 4, 9 — mapping, revisions, pricing: this is where products actually differ), then production fit (5, 6, 7). Most vendors, SimplyFill included, have a free tier or trial — so the honest evaluation is empirical: pick the two or three ugliest PDFs in your workflow and run them through each candidate's real setup flow. The marketing pages all say the same thing; the upload screens don't.
SimplyFill's quickstart is the five-minute version of that test, and the criteria above are exactly the ones we'd ask you to score us on.