AI Agents & AutomationFree worksheetPDF, Excel, Word export

Document Classification Taxonomy Worksheet

Before a document AI tool can route a PDF anywhere, someone has to define the list of document types and what happens to each. This worksheet is that catalog: one row per type, with the channel it arrives by, how it will be classified and extracted, where it lands in the ERP and what review effort it will create.

Free, optional. Reopen it on any device.

Download:

Document type	Category	Arrives via	Extraction method	Key fields	Lands in	Retention	Documents per month	Share needing review (%)	Minutes per review (min)	Review hours per month
										53 hrs
										10 hrs
										24 hrs
										9 hrs
										7 hrs

Total documents per month

5,780

Total review hours per month

102 hrs

Review minutes per document, all types

1.0606

Retention periods are legal and contractual questions, so confirm them with your records policy and customers rather than relying on defaults. Start with the five or six document types that make up most of your volume.

Get a taxonomy review from your own inbox sample

Send an anonymised week of document filenames and types and we will suggest what to automate first. Optional follow-up.

Optional. The tool and every export work without this. No spam.

Was this tool useful?

How to use the worksheet

Pull a week of documents from your shared inbox, scanner folder and portals, and group them by type. Add a row for each, noting the channel, the key fields you need, and the ERP object that should receive the data. Choose an extraction method per type, then estimate the share that will need review and the minutes per review. The computed review hours show where the human workload will sit, so you can decide which types to automate first and which to leave manual.

Automate by volume and rule clarity, not by document importance
Keep a catch-all type for unknown documents so nothing disappears
Version the list as you add types

Designing the taxonomy

Keep it flat and operational. A taxonomy of two levels, category and type, is easier to maintain than a deep tree. Name types by what the business does with them, such as supplier invoice or goods receipt support, not by file format. Each type needs one owner, one destination and a defined fallback when classification is uncertain. Overlapping types such as invoice and credit note should be separate because the posting logic differs, even if the layouts look alike.

Sequencing the rollout

Start with high-volume, structured types where errors are cheap to fix. Add certificates and quality records afterward, with higher review rates and stricter confidence bars. Leave rare, complex documents manual until you have the volume to justify tuning. After go live, track classification accuracy separately from extraction accuracy, since a misclassified document fails before any field is read.

Measure misrouted documents per week
Review the unknown bucket monthly for new types
Retire types that no longer arrive

Frequently Asked Questions

How many document types should I start with?

Five to ten, covering most of your volume. A long tail of rare types adds tuning effort for little saving. Add types when the unknown bucket shows enough volume to pay for them.

What is the difference between classification and extraction?

Classification decides what kind of document it is and where it goes. Extraction reads the fields from it. Both can fail separately, so measure them separately and route low-confidence results to review.

Do I need retention periods in the taxonomy?

It is useful, because the document type usually determines how long you must keep the record. Confirm the periods with your records policy, tax rules and customer contracts, since they differ by country and industry.

Can an LLM replace a taxonomy?

An LLM can classify and extract flexibly, but you still need the taxonomy to define the destinations, owners and rules. The model reads the document; the taxonomy tells the business what to do with it.

If you want an inbox classifier that routes each document type to the right ERP workflow on your own servers, we can design the taxonomy with you and run it privately, with a person confirming anything uncertain.

Talk to an Expert Explore More Free Tools