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.

AiBezwaar (Taxation)

AiBezwaar supports the processing of objections by reducing the administrative burden. Incoming e-mails are analysed automatically, identifying relevant data such as the submitter, main parties involved and reasons for objection. The algorithm then suggests a text to answer the objection.

Last change on 6th of June 2025, at 12:21 (CET) | Publication Standard 1.0
Publication category
Other algorithms
Impact assessment
Field not filled in.
Status
In use

General information

Theme

Organisation and business operations

Begin date

Field not filled in.

Contact information

gemeente@aaenhunze.nl

Responsible use

Goal and impact

The aim is more efficient handling of objections, reduction of administrative burden and shorter processing time.


AiBezwaarwaar:

  • Identifies the applicant and main parties involved.
  • Identifies the WOZ object of objection.
  • Identifies whether the objection is a pro-forma objection.
  • Identifies the grounds of the objection (grievances).
  • Identifies whether a hearing is requested.
  • Prepares a draft response based on previously given answers to certain grievances.


Based on this, a case is immediately created in the objections module.

When not all required data can be recognised automatically, the e-mail is flagged for manual processing by an employee. Here, the employee can quickly and efficiently complete the missing information and register the objection correctly.


Considerations

It is about more efficient handling, with the resident getting a quicker response.

Human intervention

AiBezwaar is mainly aimed at booking objections to ease the administrative burden. The official should always assess the suggested text and draft his/her own objection letter. For this purpose, he/she can use the suggested texts, modify them, generate new suggestions or draft a new text himself/herself. When a suggested text is modified or newly created, it is included in the suggestion options at the next objection analysed. A case can never be completed without an employee himself indicating whether an objection is founded or unfounded. This is not something that the algorithm automatically fills in the application. Without this judgment, an objection cannot be finalised either.

Risk management

The data is processed on a Microsoft server within the European Economic Area.


Microsoft stresses that customer data will not be used for training the AI model. The Azure OpenAI service has mechanisms to detect and prevent abuse, including content filtering and abuse monitoring. If potential abuse is detected in input data, it can be reviewed by a large language model or, if necessary, by human reviewers. This helps enforce compliance with the code of conduct and other applicable product terms and conditions.


Under the EU Digital Services Act, Microsoft is required to take action against illegal content, such as hate speech, terrorism and misinformation. To check for this, in a standard situation Microsoft retains the prompts and generated responses that users send via the Azure OpenAI service for up to 30 days. If the user processes sensitive or confidential data and wants to modify or disable this default retention, Microsoft offers the option to apply for restricted access. Customers meeting certain criteria, such as PinkRoccade, can use this to adjust or disable the default misuse monitoring, which PinkRoccade has indeed had done. This means that human assessment is not performed. Because automated assessment still takes place, the prompts and generated responses should still be stored temporarily. Microsoft states that this review is done in real time. Data is deleted after running this scan.


PinkRoccade has an ISAE 3402 type2 certification. They regularly check Microsoft's compliance.

Microsoft has ISO 27001, SOC2 and SOC3 certification. Microsoft has prepared a Data Protection Addendum in response to the AVG, describing how they handle customer data. This provides users like PinkRoccade with guidance on how to check whether data is secure.


Microsoft also has itself audited several times a year by independent auditors, for compliance with these rules. These audits are not optional; if Microsoft does not comply, there can be consequences. In addition, Microsoft cooperates with regulators and has a strong focus on transparency. Through the Microsoft Trust Center, you can see which compliance requirements they meet.

Legal basis

It concerns the municipality's task of tax collection. See WOZ Act, Municipal Act, Environmental Taxes Act

Operations

Data

Assessment number, BSN, NAW, WOZ number, applicant, authorised representative, grievances, reasons for objection

Technical design

AiBezwaar is an optional module in the iBezwaar application. AiBezwaar consists of PinkRoccade's own algorithm that interacts with OpenAI's large language model GPT 4o. It connects to GPT 4o via Microsoft's Azure OpenAI service.


The proprietary algorithm delivers a prompt to GPT 4o. The prompt consists of a copy of the objection; the request to identify the necessary data, such as WOZ number; and a set of possible grievances and objection reasons. GPT 4o then looks for a match: what grievances and objection reasons are in the objection and do they match options from the set. Once that match is made, another interaction with its own algorithm takes place. Here, for each of the obsolete objections from a database of historical answers, the best matching answer is found and shown as a suggestion in AiBezwaar.


When certain data cannot be found in the objection, the other data is used to automatically find the missing data in the tax application. For example, using the assessment number + NAW to find out what the WOZ object number is.


The algorithm has a learning capability. When a text suggested by the algortime is modified by the clerk or the clerk writes a text himself, it is included in the text options suggested by the algorithm in the next objection analysed.

External provider

PinkRoccade

Similar algorithm descriptions

  • This algorithm helps e-mail users determine when an e-mail contains sensitive information and should be sent securely.

    Last change on 5th of September 2024, at 11:29 (CET) | Publication Standard 1.0
    Publication category
    Other algorithms
    Impact assessment
    Field not filled in.
    Status
    In use
  • This algorithm helps e-mail users determine when an e-mail contains sensitive information and should be sent securely.

    Last change on 15th of October 2024, at 8:14 (CET) | Publication Standard 1.0
    Publication category
    Other algorithms
    Impact assessment
    Field not filled in.
    Status
    In use
  • This algorithm helps e-mail users determine when an e-mail contains sensitive information and should be sent securely.

    Last change on 12th of February 2025, at 9:35 (CET) | Publication Standard 1.0
    Publication category
    Other algorithms
    Impact assessment
    Field not filled in.
    Status
    In use
  • This algorithm helps e-mail users determine when an e-mail contains sensitive information and should be sent securely.

    Last change on 7th of August 2024, at 9:43 (CET) | Publication Standard 1.0
    Publication category
    Other algorithms
    Impact assessment
    Field not filled in.
    Status
    In use
  • This algorithm helps e-mail users determine when an e-mail contains sensitive information and should be sent securely.

    Last change on 12th of September 2024, at 7:35 (CET) | Publication Standard 1.0
    Publication category
    Other algorithms
    Impact assessment
    Field not filled in.
    Status
    In use