Autonomous processing of sick-leave notes and medical documents
Dozens to hundreds of HR documents land on payroll every month — sick-leave notes, outpatient records, medical board decisions. Kognify reads them, validates them, links each to the right employee, and feeds finished data straight into payroll. A proven platform, a clear SLA, a configuration tuned for medical documentation — and our team owns setup, onboarding and AI tuning end to end.
Sick-leave notes eat into the HR tempo of a large company.
A large company processes hundreds of sick-leave notes every month. Each is a paper or scanned document that has to be read, linked to a specific employee, validated, and entered into payroll before the pay cycle. Manual handling isn't just slow — it's fragile: a wrong date, a missed note or an unrecognized diagnosis, and the monthly payroll comes back for correction.
From keying in documents to reviewing exceptions.
The same documents, handled two ways. Kognify moves your HR team off the keyboard and onto the decisions that actually need judgement.
The shift: from an HR team that keys in documents to an HR team that only reviews the exceptions and makes decisions.
Four types of medical documents — one service.
Each type gets its own configuration: which fields to extract, which rules to apply, and which system to send the structured result to. New types are added through the UI, not through an IT project.
Sick-leave notes
Standardized NHIF / Ministry of Health forms — the core volume.
Outpatient records
Documents from examinations and consultations — more variable in format, where Kognify's logical flexibility comes into play.
Medical board (LKK / TELK) decisions
Expert decisions that need specific payroll treatment — extended absences, reassignment, changed work capacity.
Other HR-relevant documents
Maternity, childcare and driver medical certificates — relevant for companies with driving and operational staff.
Special-category data, handled accordingly: medical documents fall under GDPR Art. 9. Processing follows your signed DPA exactly — encrypted at rest and in transit, and never used to train AI models.
Five steps — from the document arriving to the payroll record.
The same proven Kognify model, tuned for the HR context. The steps are almost invisible to the user — you see a finished result and an audit trail, not intermediate states.
📥 Document intake
Arrives through three channels: manual upload via the Kognify UI, a dedicated email address just for notes, or an integration with the system where employees already submit documents.
🏷️ Type recognition
Automatically determines whether the document is a sick-leave note, an outpatient record, a medical board decision or something else — which sets the configuration to apply.
🔎 Extraction & validation
Extracts every relevant field, then runs logical checks: is the national ID valid, does the period make sense, does it overlap with another note, is the physician known.
🧾 Reconcile to the employee
National ID and names are checked against your employee register and linked to the right record. Anything that doesn't match cleanly goes to a human, not into payroll.
📤 Deliver to payroll/HRIS
The structured result is posted via direct integration, scheduled CSV export or JSON — with a link to the source document and a full audit trail on every field.
Three AI steps, with cross-verification.
Sick-leave notes aren't ordinary documents — they combine free text, marks and checkboxes where position carries meaning. So we don't rely on a single AI pass, but on a three-stage pipeline in which each step checks the previous ones with a different context.
01 · Preparation & context
Turning the file into structurable data. Document-type recognition with custom semi-agentic AI, and a preliminary filter of the relevant nomenclatures for the type.
- file → structured data
- document-type recognition
- nomenclature pre-filtering
02 · Extraction with vision AI
Review and extraction of the data requested for the given document type, matched against the pre-filtered nomenclatures.
- visionAI · primary pass
- type-specific extraction
- structured result
03 · Verification & cross-check
A second vision pass, enriched with the main result and the pre-processing data — compares and verifies across all three sources before the final result.
- visionAI · verification pass
- 3-way cross-verification
- final confidence score
Why three steps and not one? A single AI pass — however good the model — has one opinion about a document, and even reasoning models vary. With three specialized steps, each checks the previous with a different context. That's the difference between "probably correct" and "soundly verified".
What makes the processing reliable in production.
The difference between a demo and a production system is in the cases that aren't ideal. These are the mechanisms that make it ready for your company's real traffic.
Logical verification
Not just OCR. It checks whether the extracted data makes sense — do periods overlap, is the national ID valid, does the diagnosis match the ICD-10 catalog.
Self-assessment of the result
Each document gets a status — green (ready for payroll), yellow (minor uncertainty), red (needs human review). ~99% of cases take the green path.
Reconciliation with your systems
Integration with the employee register for identification by national ID and name, with the option of extra checks against the absence calendar.
Readiness for peak periods
Month-end, flu season, the days after holidays — load spikes. The platform scales automatically, without the HR team noticing.
Full audit trail
For every document: when it arrived, what was extracted, which validations passed, who changed what and when — ready for HR and Labour Inspectorate checks.
No-code configuration
A new document type or rule is added through the UI as plain-language instructions — not as an IT project.
Special-category data, under EU controls.
Built on Kognify Universal Content Handling, in a secure environment with a dedicated tenant per client — the same posture that carries banking-grade audit certifications.
The implementation, onboarding and AI optimization — handled by the Kognify team.
Kognify takes on the full technical implementation and AI optimization until the target quality is reached. This isn't standard industry practice — it's part of our vision for AI, where providers take responsibility for the result so AI can move toward autonomy.
Implementation
Tenant configuration, defining the document types, setting up extraction and validation rules, and integration with your systems (payroll / HRIS / employee register).
Team onboarding
Training the HR/payroll users on the UI, how to review exceptions, and how to add new rules — with documentation and Q&A sessions.
AI optimization
Iterative improvement of extraction quality on your specific documents and formats, until the target success rate is reached. No timer, no separate phases.
From signed contract to production in 3 to 4 weeks.
A conservative estimate — with standard documents and access to the employee register, it's often shorter. The Frictionless Start period can be extended at no extra commitment if more tuning time is needed.
Week 1 · Setup & document analysis
Activate the tenant, receive samples of your real document types, begin configuration, and define which integrations are needed to your systems.
Week 2 · POC on real documents
A first round of extraction on your real (anonymized, if needed) documents. We review results together, note specifics, and apply corrections — a working POC you can test by the end of the week.
Week 3 · Integrations & fine-tuning
Connect to payroll and the employee register, set up intake channels, train the HR team, and fine-tune on edge cases — non-standard formats, handwriting, poor scans.
Week 4 · Production launch
Switch to production with close monitoring for the first few days, then move to the usual SLA model once stabilized.
Why right now is the moment to automate.
Zero risk at the start — the implementation, onboarding and AI optimization are on us. You see a working POC on your real documents by week two, and move to production only once the quality is proven.
The economics of volume
A few hundred notes a month is exactly the volume at which automation pays back fastest — roughly 0.5–1 FTE of manual effort, before mid-cycle corrections.
A regulatory moment
Labour Inspectorate and Revenue Agency checks keep getting more thorough. Structured, auditable processing with a full trail reduces the risk of penalties.
Technology maturity
Kognify Universal Content Handling already runs in production for other clients in similar scenarios — not an experiment; the results are validated.
A clear contract, a valid DPA, premium support — no hidden risks and no long learning curve for the HR/payroll team.
Not a demo — production at some of the largest companies in Bulgaria.
Kognify runs in production for enterprise clients in Bulgaria, Slovenia and Australia — with a flawless security record and bank-grade audits. The platform behind this service already processes documents every day for companies whose names you know.
As one of Bulgaria's leading courier companies, we process thousands of documents every month. With Kognify, in the very first month of deployment we achieved over 98% successfully processed documents — even while the system was still being tuned. It independently identifies discrepancies and escalates only genuine exceptions. That gives us confidence without losing control.
Ready to start this very month.
A signed contract — and week 1 starts right away. Send us the document types you process and we'll outline what implementation looks like on your end. Free onboarding. 3–4 weeks to production.