Please note: The algorithm descriptions in English have been automatically translated. Errors may have been introduced in this process. For the original descriptions, go to the Dutch version of the Algorithm Register.
NACE SBI Classifier
- Publication category
- Impactful algorithms
- Impact assessment
- DPIA, ...
- Status
- In development
General information
Theme
Begin date
End date
Contact information
Link to publication website
Link to source registration
Responsible use
Goal and impact
In 2025, there will be a major revision in the Standard Business Classification (SBI) imposed from Eurostat, the European statistics office. Every organisation gets one or more SBI codes when it is registered in the trade register, or when changes to activities take place.
The NACE 2025 revision is based on the 2008 classification. This revision means that by September 2025, KVK will start updating all existing 5.6 million SBI codes on a total of 3.75 million entries:
- 4.6 million "mutations" where the new code is the same as the old code (sometimes with an extra digit or letter) or 2 existing SBI codes merge into 1 SBI code
- 1 million "characterisations" where 1 old code will be split into 2 or more new codes
The NACE SBI Classifier supports this revision via several sub-processes.
- The first step is the removal of traceable data within the KVK network, which in subsequent steps, could potentially lead to bias or misuse.
- In the next step, by means of an LLM (ChatGPT 4o), based on a combination of inputs (prompt), a proposal for new SBI code according to 2025 standard for registered enterprises will be generated. This uses CBS' switching table and explanations of SBI coding from CBS.
- Outcomes of both steps are spot-checked manually and manually corrected by staff, the 2nd step using a motivation generated by LLM.
After revision, the SBI code can still be changed by entrepreneurs themselves via existing channels (Online Change/Mijn KVK). The entrepreneur will be informed about this through various channels.
Considerations
- Actors: Registered entrepreneurs, Chamber of Commerce employees
- Interests: For the entrepreneurs, their interests are better served by this solution as the conversion can be carried out more effectively and he/she will be better assisted in taking compulsory services through existing channels. The interests and frameworks of the two groups are laid down in the Commercial Register Act. In addition, this application also affects the interests of the KVK employees, in the sense that the alternative of manually typing over a million SBI codes would hamper the performance of the other declarations laid down in the Trade Register Act.
Human intervention
- Entrepreneur has registered or modified company according to existing methodology with intervention of Front Office employee or Back Office employee
- Model is trained and tested based on input by quality staff, for application to production data.
- Revision proposals on production data based on algorithms are spot-checked by employees.
- Entrepreneur can change enrolment/registration after revision with subsequent verification by Back Office employee
Risk management
The risk of processing personal data has been minimised by applying anonymisation (removing data that can be traced back to individuals) for the purpose of the "AI process".
DPIA checklist has been run through, which showed that we do not need to draw up a DPIA. The checklist refers to KVK's Cloud, AVG and AI and algorithm policies.
Legal basis
- Section 19 Trade Register Act 2007
- Section 13 Trade Register Act 2007:
Links to legal bases
- Artikel 19 Handelsregisterwet 2007: https://wetten.overheid.nl/BWBR0021777/2024-06-19/#Hoofdstuk3
- Artikel 13 Handelsregisterwet 2007: https://wetten.overheid.nl/BWBR0021777/2024-06-19/#Hoofdstuk2
Link to Processing Index
Elaboration on impact assessments
- DPIA = Data Protection Impact Analysis
- BIA = Business Impact Analysis
- The BIV score in the BIA is about Availability, Integrity and Security. Here 1 is the lowest and 3 the highest.
- During the impact tests we looked at: Internal AI policy based on EU AI Act, Internal Cloud policy based on Government-wide Cloud Policy 2022, Internal Privacy Policy based on General Data Protection Regulation (AVG)
Impact assessment
- DPIA checklist: Prescan van DPIA gedaan, daar is uitgekomen dat er geen DPIA benodigd is
- BIA checklist: Prescan van BIA gedaan, daar is uitgekomen dat er geen BIA benodigd is (BIV score: 1-2-1)
- Intern KVK AI beleid: Op basis van de EU AI Act
Operations
Data
- See below
Links to data sources
- CBS SBI codelijst: https://sbi.cbs.nl/
- Spacy: https://www.spacy.io/
- Handelsregister: https://www.kvk.nl/zoeken
Technical design
- Extracted company registration data from trade register.
- Transit through anonymisation tool and check.
- Individual enterprise data is attached to prompt with circuit diagram, typing information and activity descriptions.
- ChatGPT 4o provides suggestion for SBI code to be entered as per 2025 revision with rationale for choice.
- Outcomes are validated by quality experts before revision will be implemented in trade register.
External provider
Link to code base
Similar algorithm description
- CBBS is used in determining the disability rate.Last change on 17th of February 2025, at 14:12 (CET) | Publication Standard 1.0
- Publication category
- Other algorithms
- Impact assessment
- DPIA
- Status
- In use