Introducing Smart Fields: signatures that place themselves

Preparing a document for signature has always been the least interesting part of using SignTheDoc. You upload a contract you have sent a hundred times, then spend five minutes dragging little blue boxes onto signature lines that were already right there on the page. It works, but nobody enjoys it.

Smart Fields fixes that. Upload a document and our engine reads the page, finds every signature line, initial box, and date field, and places the correct field type on each one — assigned to the right signer, sized to fit, ready to send. For a standard contract, document prep now takes about four seconds.

What it actually detects

Smart Fields looks for the visual and textual patterns that humans use to mark where something goes. In practice that means:

• Signature lines — a horizontal rule with "Signature", "Signed by", or a name beneath it
• Initial boxes — small bracketed or underscored regions, usually in page margins or footers
• Date fields — "Date", "Dated this ___ day of", and common date formats left blank
• Printed-name fields — a line paired with "Print name" or "Name (printed)"
• Title and company fields — the lines that almost always accompany a countersignature block
• Checkbox groups — initial-here-to-acknowledge patterns common in disclosures

Each detected field comes back with a confidence score. Anything above our threshold is placed automatically. Anything below it is surfaced as a suggestion with a dotted outline, so you can accept it with one click or ignore it.

Assigning fields to the right signer

Detecting a signature line is only half the problem. A two-party agreement has two signature blocks, and putting both on the same recipient produces a document nobody can actually complete.

Smart Fields reads the surrounding text to work out who each block belongs to. Words like "Landlord", "Tenant", "Company", "Contractor", "Buyer", and "Seller" map to signer roles. When you add recipients, we match them to the roles we found and colour-code the fields accordingly. If we guess wrong, changing the assignment is a dropdown.

How we built it

The first version was a pure computer-vision model. It found lines beautifully and understood nothing. It would confidently place a signature field on the underline of a section heading, because visually that is exactly what a signature line looks like.

What actually works is a combination. We extract the text layer and its positions, run layout analysis to find candidate regions, then use the text around each candidate to classify it. A horizontal rule with "Authorized Signature" underneath is a signature field. The same rule under "Section 4 — Termination" is a heading underline. The text is what disambiguates.

For scanned documents with no text layer, we run OCR first and accept lower confidence. Scanned contracts get suggestions rather than automatic placement, because a misread on a scan is much more likely than on a native PDF.

What it will not do

We would rather be honest about the edges than have you discover them on a live send.

• Heavily designed documents — if your contract is laid out like a brochure, detection accuracy drops considerably
• Handwritten forms — a scanned page of handwriting is not something we attempt to parse
• Languages beyond our current set — English, Spanish, French, and German are supported today
• Deciding signing order — we place fields; sequencing recipients is still your call

In every one of those cases you get the same manual placement you have always had. Smart Fields is additive; it never blocks the old path.

How we handle being wrong

A field placed incorrectly is worse than a field not placed at all, because it is easy to miss during a quick review. That shaped several design decisions.

Automatically placed fields are visually distinct until you confirm them — a subtle marker in the corner that disappears once you interact with the document. We also show a count in the toolbar: "12 fields placed automatically". You cannot send without the preparation screen having been opened at least once.

We considered requiring explicit confirmation of every detected field and decided against it. Confirming twelve fields individually is barely faster than placing them, and confirmation dialogs that appear every time get clicked through without being read. A clear visual state plus a mandatory review step gets most of the benefit without the ritual.

Privacy of the detection process

Detection runs inside our own infrastructure. Document contents are not sent to any third-party service, and nothing from your documents is used to train the model.

The model was trained on a corpus of publicly available contract templates and synthetic documents we generated. When we improve it, we do so from that corpus and from aggregate signals — how often suggestions were accepted or rejected, without the content that produced them. We think this is the only defensible position for a company holding other people's agreements.

Turning it on

Smart Fields is on by default for every plan, including Free. When you upload a document you will see detected fields appear with a brief highlight. If you would rather place everything yourself, there is a toggle in the preparation toolbar, and the setting sticks per account.

Across our beta, median preparation time for a standard two-party contract fell from four minutes and ten seconds to eleven seconds. The remaining eleven seconds are mostly people double-checking our work, which we consider entirely reasonable.