Docyield's document classifier looks at a PDF or image and tells you what it is — an invoice, a receipt, a bank statement, a contract, an ID, and more — together with a broader category, the document's language, a confidence score, and a one-line summary of its contents. Instead of opening every file to sort it by hand, you upload the document and get back a quick, structured verdict you can route, file, or act on.
Classification is the step that usually comes before extraction. When documents arrive in a mixed stream — email attachments, a scanner output folder, an upload inbox — something has to decide which is which before the right parser can run. This tool fills that gap: it identifies the document type so the rest of your workflow knows what to do with each file, whether that is sending invoices to accounts payable or contracts to legal. Get the sort right at the front, and every step after it becomes simpler and more accurate.
What a document classifier does
The classifier reads the document, weighs the evidence on the page — its layout, headings, key phrases, and the kind of data it contains — and assigns the single best-fitting type from a defined list. It also returns a wider category such as financial, identity, logistics, HR, or legal, detects the primary language, and writes a short summary so a person can confirm the call at a glance.
It does not try to pull out every field; that is a job for a dedicated parser. The classifier answers a narrower, earlier question — "what is this?" — and answers it in a structured way. That distinction matters: a fast, reliable label up front lets you send each document to the correct downstream tool instead of running everything through one ill-fitting extractor. Skipping that step is a common reason extraction pipelines produce poor results, because a parser tuned for invoices will struggle if you feed it a contract or an ID card by mistake.
Why automated classification beats manual sorting
Sorting documents by hand is slow, dull, and surprisingly error-prone — it is easy to misfile a credit card statement as a bank statement, or a packing list as an invoice, when you are working through a large pile. Those mistakes then cascade, because the wrong document reaches the wrong process and has to be unpicked later.
An automated classifier applies the same judgement to every file and returns a consistent, machine-readable label plus a confidence score, so you can branch your workflow on the result. High-confidence documents flow straight through; the small number that score low can be set aside for a person to check. That is far more scalable than eyeballing each file, and it makes the sorting step auditable — you have a record of how each document was classified and how sure the system was, which matters when someone later asks why a file went where it did.
Who uses a document classifier
- Operations teams triaging a mixed inbox of uploads before sending each file to the right parser.
- Document-management and archiving projects that need to tag and file documents by type at scale.
- Developers building an intake pipeline that routes invoices, IDs, and contracts to different services.
- Back-office teams pre-sorting scanned mail so each department receives only what is relevant.
- Anyone who needs to know what a stack of files contains without opening every one.
Confidence scores and review
Every classification comes with a confidence value between 0 and 1, and that number is the point. It lets you set a threshold: route anything above it automatically and queue anything below it for a human to confirm. Treating the score as a signal rather than ignoring it is what keeps an automated pipeline honest.
No classifier is correct on every document, and some files are genuinely ambiguous — a hybrid document, a poor scan, or something that does not fit any standard type. When nothing fits well, the classifier returns "other" and explains what it saw in the summary rather than forcing a misleading label. Pairing that behaviour with a confidence threshold means the few hard cases surface for review instead of slipping through mislabelled.
What makes some documents hard to classify
Clear, complete documents classify easily; the difficulty rises with poor inputs. Low-resolution scans, photos taken at an angle, heavy glare, and faint print all reduce how much the classifier can read, so a sharper image leads to a more confident, more accurate call.
Content can be tricky too. Multi-part documents — a cover note stapled to an invoice, or a bundle of several forms scanned as one file — blur the boundaries between types. Documents in languages other than English are supported, and the detected language is returned as its own field, but unusual or mixed-language layouts can still lower confidence, which is exactly when the review threshold earns its place. The summary field helps here as well: even when the type is uncertain, a one-line description of what the document appears to contain often gives a reviewer enough to make the call in seconds rather than opening the file in full.
Output formats and scaling to the API
Each classification is available as JSON, CSV, Excel (XLSX), or XML from the same result. JSON is the natural choice for routing logic in code, since your pipeline can read the type and confidence straight from named fields; CSV and Excel suit anyone logging classifications in a spreadsheet, one document per row; XML fits older systems that expect markup. Because the formats are serializations of the same result, switching between them is instant and free.
When you need to classify documents continuously — sorting an incoming stream before extraction — the Docyield API and batch dashboard handle volume and return the same schema, including the confidence score, with webhook delivery so your pipeline can branch automatically the moment a result is ready. The structure stays identical between the free tool and the API, so a routing rule you test here against a single file carries over unchanged to production. That makes the classifier a natural front door to the rest of Docyield: classify first, then hand each document to the matching parser.
What the document classifier returns
Each document is described against a fixed schema, so your routing logic can rely on the same fields every time.
- Document type
- The single best-fitting type — for example invoice, receipt, bank statement, contract, passport, ID card, or form — or "other" when none fit well.
- Category
- A broader grouping such as financial, identity, logistics, HR, or legal.
- Language
- The primary language detected in the document.
- Confidence
- How sure the classifier is, expressed from 0 to 1, so you can set a review threshold.
- Summary
- A one-sentence description of the document's contents to help confirm the classification at a glance.
How to classify a document
- 1Upload your document — drop a PDF, PNG, JPG, or WEBP onto the box above, or click to choose a file.
- 2Wait a few seconds while Docyield reads the document and weighs the evidence.
- 3Review the returned type, category, language, confidence, and summary.
- 4Pick the output format you need from the tabs — JSON, CSV, Excel, or XML.
- 5Copy the result or download the file, and use the type and confidence to route the document in your workflow.
Frequently asked questions
Processing documents at scale?
Batch upload, an extraction API, and webhooks for 100+ documents a month.