Templates vs AI capture

Scanix Desktop gives you two ways to pull data off a document, and the fastest path to a reliable, high-volume workflow is knowing which one to reach for. A template captures data from a known layout with pinpoint precision. AI capture reads documents whose layout you can't predict, with almost no setup. This page builds the mental model behind both, then shows how to combine them so each does what it's best at.

Two ways to read a document

Both approaches do the same job — turn a scanned page into named field values you can verify and export — but they get there differently, and they shine in different situations.

A template is a saved recipe for one document type. You draw extraction zones over a sample page (the spots where the values live), tell Scanix what each one is — a date, a currency amount, a barcode — and it reads exactly those regions on every document you run through. Because you've told it precisely where to look, a template is fast, deterministic, and cheap to run: the same input always produces the same output, with no cloud call required.

AI capture takes the opposite stance. Instead of "read this rectangle," you hand the AI a whole page and a goal — "this is an invoice, find the fields an invoice has" — and a model reads the document's text to classify it and return structured fields. You don't draw zones or measure coordinates. That flexibility is the whole point: AI capture copes with layouts you've never seen and variation you couldn't anchor for.

When a template wins

Reach for a template-only approach when the layout is fixed and repeatable and you'll run it at volume.

A template is the right tool when:

• The layout is consistent. The invoice number is always in the top-right; the passport's machine-readable zone is always the bottom two lines. When values sit in predictable spots, a zone nails them every time.
• You need exact, repeatable results. Zone OCR is deterministic — identical pages give identical output — which is what auditors, downstream systems, and compliance reviews want.
• You run the same document type at high volume. The one-time cost of drawing zones pays for itself across thousands of documents, and there's no per-document AI call to think about.
• The data must never leave the machine. Zone OCR runs entirely on your device, so it's a natural fit for sensitive records.
• The fields are positional, not interpretive. A barcode, an MRZ band, a fixed label-and-value pair — these are about where the data is, which is exactly what zones express.

Templates also bring tools that keep them accurate as real-world scans drift. Anchoring locks a zone to a visible reference — a Text label, a Barcode, a QR code, or a Page edge — so the right value is captured even when a scan is shifted, skewed, or rotated, or when a layout varies slightly between document variants. You build all of this in the Custom Builder (the Designer). Learn the full toolset in Templates.

When AI capture wins

Reach for AI capture when the layout is varied or unknown, or when drawing and maintaining zones would cost more than it's worth.

AI capture is the right tool when:

• Layouts vary. Every supplier's invoice looks different; you can't draw one set of zones that fits them all. The AI reads each document on its own terms.
• You can't predict what's coming. A stack mixes invoices, receipts, and delivery notes with no separator sheets. The AI classifies each page and decides where one logical document ends and the next begins.
• Setup time matters more than per-document precision. A Smart Template builds a working template from a single sample in a few clicks — no manual zone-drawing — so you're processing in minutes, not after an afternoon in the Designer.
• The fields are interpretive. "Which line is the grand total?" or "who is the supplier?" are reasoning questions, not coordinate lookups, and a model handles them well.

Under the hood, AI capture is organised around AI Services — pre-built extraction schemas for a document class such as Invoices AI, Receipts AI, Identity Documents AI, or Bank Statements AI. A service tells the AI exactly which fields to pull and normalises the result to canonical forms (dates to YYYY-MM-DD, money to an amount plus a currency code, and so on). You pick a service, and in the Viewer the primary button becomes Analyse with AI, which runs OCR and the service together and fills your mapped fields. The full catalogue and setup live in the AI Services catalogue.

Combine them — the best of both

You rarely have to choose one or the other. The most capable Scanix workflows layer templates and AI together, and the product is built for exactly that.

Smart Templates — AI builds, the template runs

A Smart Template uses AI to do the setup work, then hands you a normal template to do the running. The wizard reads one sample page, classifies it, and proposes the zones, index fields, splitting strategy, outputs, and — when it's confident — an anchor. You review the suggestions and save. From then on you have a fast, deterministic template you can refine by hand in the Designer; the AI's job was getting you there without drawing a single zone. It's the shortest path from "I have a stack of these" to a working template — see Smart Templates.

AI in Job Groups — classify, route, then capture

A Job Group is the container for a batch that mixes document types. AI classification and splitting reads a single stack with no separator sheets, decides where each logical document begins, and routes each segment to the right member template — which then captures with its own zones (and its own AI, if you set one). This is templates and AI in the same pipeline: AI handles the sorting that zones can't, and each template handles the precise capture it's built for. A Job Group only earns its keep when batches genuinely mix types; for a single-document-type batch, a plain single-template job is faster. Learn more in Job Groups.

AI fallback — zones first, AI for the gaps

You can also keep a template's zones as the primary capture and let AI fall back for anything they miss. Map most fields to their OCR zone, and point the trickier or more variable ones at an AI: Service value instead. When you've turned on Fall back to plain OCR when no service matches in Settings → AI Services, the pipeline degrades gracefully — if no AI service applies, you still get the zone-based OCR result rather than nothing. This gives you a template's speed and determinism where the layout is stable, and the AI's flexibility precisely where it isn't.

A short decision guide

When you're not sure which way to go, work down this list — the first match is usually your answer.

Next steps

Related

• Templates
• AI Services catalogue
• Smart Templates
• Job Groups