Hours come back to your team
Work that is repetitive, rule-based and high-volume is where automation pays, and it is found by measuring the process you run today.
AI document classification
An intake folder can contain several jobs: an application to review, a contract to file, or a form whose fields need extracting. Norml builds AI document classification for that first decision. We define the document classes with your team, test representative files and route the results to named queues. Files that do not meet the agreed routing checks stay available for a person to inspect.
A few clients we’ve worked with
What we do
Classify incoming documents into agreed categories, route them to the right queue, and retain uncertain files for review before extraction or processing.
What an automated step is actually worth.
The expensive mistake is automating the wrong step convincingly. The value is real, and it comes from measuring the process before building anything.
Growth stops adding headcount
When the routine path runs itself, the next increase in volume does not need another person watching a queue.
The output can be checked
An automated step nobody can verify is a liability, so the review point is part of the design rather than something bolted on later.
Classification before document processing
The starting scope is a known set of document families and destinations. We retain the original file and make the routing decision traceable.
- Solution
- Automations
- Tools
- Azure AI Document Intelligence, n8n, Airtable
- Trigger
- A document enters the agreed intake with a source ID and access to the pages needed for classification.
- Outcome
- A document category, relevant page ranges and routing decision are recorded. Eligible files reach the selected queue or extractor; uncertain cases enter manual review.
Categories your team can apply consistently
We define each class with examples and exclusions, including similar-looking documents that need different handling. Categories correspond to a business destination or next step. We test how the model handles unfamiliar material rather than assuming every unknown file will identify itself as unknown.
Page ranges for mixed document packets
When an upload contains several documents, we assess whether classification should operate on the whole file, individual pages or detected document sections. Splitting behavior is configured deliberately and tested against packet examples. The original upload remains available while routing records point to the relevant pages.
Routing with an exception queue
Airtable can record the source, predicted class, checks and destination. Agreed thresholds and rule conflicts send a file to review. The processing record identifies which handoff completed, so a retry does not repeatedly notify an owner or submit the same pages to an extractor.
Client reviews
Read our reviews on Clutch ↗“Norml Studio's professionalism and responsiveness were impressive.”
“They were proactive in suggesting improvements we hadn't considered.”
“They delivered what they said they would, on time.”
“They focused on how to make our daily operations easier, not just on building features for the sake of it.”
“They understood that the app needed to be practical for someone sitting in a car, not just look good on a desktop.”
“Their level of dedication to the success of the project and GoPetZen's business as a whole is unique.”
What we need to define the classes
Examples for every category
Supply representative samples, confusing pairs and documents outside the intended classes. A new category needs a definition and evaluation examples before it becomes a routing destination.
A destination and permission map
Identify which queue, team or extractor receives each class. Classification must not grant someone access to a source document they are not permitted to read.
Rules for uncertainty and correction
Choose who reviews unclear files and how corrections are recorded. Agree when a file must remain manual, how long it is retained and who can approve a change to the class definitions.
How we test document routing
Define and evaluate the classes
We compare model output with checked examples and inspect confusing categories. This establishes routing rules and review conditions based on your document set, without promising a universal accuracy figure.
Connect intake to the selected queues
We build the classification call, page-range handling and routing ledger. Each destination receives the agreed file reference and context, with permission checks and a recorded handoff result.
Test packets, unknowns and retries
We test mixed uploads, blank pages, unfamiliar types and duplicate delivery events. Your team reviews corrections and learns how to pause routing when a new document format starts arriving.
The people who do this work.
No account layer between you and them. These are the Norml specialists this work runs through.
Max Tymoshyn
Founder & Architect
Artur Tsitou
DevOps & Backend Developer
Petr Savenok
Full-Stack Developer
Let's chat
“I've spent 10 years working across development and marketing. I started as a developer, but design, SEO and advertising were already part of the job. That taught me to treat a website as one system: the business, the content, the design and the code have to work together. At Norml, I lead the architecture behind our work and stay involved wherever strategy, design, development and search meet.”
Max TymoshynFounder & Architect
Get startedRelated services.
Document classification questions
Does classification also extract document fields?
Classification identifies the document type and can select an appropriate extraction workflow. Reading names, dates, totals or other fields is a separate scope with its own validation. We can connect the stages when both are needed.
Will an unfamiliar document always go to an Other category?
No. A classifier can assign an unfamiliar document to a known class. We test out-of-scope examples and use agreed confidence checks, routing rules and manual review to reduce that risk. An Other label alone is not a guarantee.
Can it separate several documents inside one PDF?
The selected classification model can return page ranges, but the splitting mode must match your packets. We test multi-page documents and boundaries before using those ranges for downstream processing. Producing separate files is an additional agreed action.
Can we change categories after launch?
Yes, through a controlled update. We add representative examples, revise definitions and retest existing categories before changing routing. Renaming a label without updating destinations and evaluation cases can send documents to the wrong owner.
Bring anonymized examples, your current categories and the destinations each type needs. We’ll identify a classification scope that your team can review and maintain.