Heidi: The Logging System Behind the Benefits Website

Summary

From mid-2013, the Belastingdienst/Toeslagen operated Heidi, a logging system that collected portal data from the benefits website. The system tracked user behavior, visits, clicks, form interactions, for approximately four years until its log files were deleted or transferred in September 2017. Heidi was deployed in the same period as Deloitte’s risk classification models that used nationality as fixed source data, forming part of a pattern of digital surveillance on citizens applying for benefits.

Context

What was Heidi?

According to parliamentary documents (Kamervragen kst-31066-1330), Heidi was a logging system deployed by the Belastingdienst/Toeslagen from mid-2013. It collected portal data, information about how users interacted with the benefits website, including navigation patterns, form completion behavior and click data.

The broader surveillance context

Heidi did not operate in isolation. In the same period:

• Deloitte’s risk models (from June 2013): SAS-based classification models for childcare and housing allowances using “BVR Nationaliteit” as fixed source data, trained on only 200 examples.
• RAM (1998-2018): The Risk Assessment Model profiling virtually all taxpayers, including scraped internet data from 2015.
• FSV (2007-2020): The Fraud Signalling Facility registering ~180,000 citizens as fraud suspects.
• SyRI’s predecessor “Black Box”: Operated for years without a separate legal basis until Wet SUWI art. 64-65 took effect on 1 January 2014.

Note on “Heide2”

The verified system name is “Heidi” as documented in parliamentary questions. The term “Heide2” has not been confirmed in official sources and may refer to an internal name, a successor system or a misunderstanding. This distinction is important for accuracy.

What happened

1. Mid-2013: Heidi is deployed on the benefits website, logging portal data from visitors.
2. 2013-2017: For approximately four years, user behavior data is collected from citizens interacting with the benefits website.
3. September 2017: Heidi log files are deleted or transferred. The reason and destination are not publicly documented.
4. 2018 onwards: No public disclosure of Heidi’s existence or scope until parliamentary questions.

The timing is significant. Heidi was deployed in the same period as the Deloitte risk models and the intensification of fraud detection under the Ministerial Commission on Fraud Approach (2013-2015).

Evidence

Analysis

Legal concerns

• AVG/GDPR art. 5(1)(b) (purpose limitation): Portal data collected for service delivery may have been repurposed for risk profiling without a separate legal basis.
• AVG/GDPR art. 13/14 (information obligation): Citizens visiting the benefits website were likely not informed that their behavior was being logged and potentially used for fraud detection.
• AVG/GDPR art. 22 (automated decision-making): If Heidi data fed into risk scores that influenced benefit decisions without human intervention, this constitutes automated decision-making with legal effects.
• ECHR art. 8 (private life): Collection of behavioral data from government website visitors.

Unanswered questions

1. What specific data points did Heidi collect? (clicks, mouse movements, time on page, form field interactions?)
2. Was Heidi data linked to individual BSNs or used for risk scoring?
3. Was Heidi data fed into Deloitte’s risk models or other decision systems?
4. Why were the logs deleted in September 2017, was this related to the upcoming GDPR?
5. Did Heidi have a DPIA (Data Protection Impact Assessment)?
6. Is there a “Heide2” successor system currently operational?

Frequently asked questions

What was the Heidi system?

Heidi was a logging system that the Belastingdienst/Toeslagen deployed on the benefits website from mid-2013. It recorded portal data from visitors: page views, navigation patterns, click behaviour and form interactions of citizens applying for or managing benefits. The name is confirmed in parliamentary questions, document kst-31066-1330.

How long did Heidi collect data on the benefits website?

For approximately four years, from mid-2013 until the summer of 2017, the system registered the digital behaviour of website visitors. In September 2017 the log files were deleted or transferred. The reason and destination are not publicly documented, which hampers retrospective investigation into how the data was used.

Which Belastingdienst systems ran parallel to Heidi?

In the same period, Deloitte built risk classification models using “BVR Nationaliteit” as fixed source data, while the “Black Box” system, SyRI’s predecessor, was operational. WRR Working Paper 021 from April 2016 also describes the BI&A unit: around 40 young academics processing data from 11 million citizens for fraud detection.

How many algorithms are listed in the Belastingdienst algorithm register?

The register lists 68 registered algorithms, of which 65 are classified as impactful and more than 50 are deemed discriminatory, according to the April 2026 register. The Autoriteit Persoonsgegevens concluded in July 2025 that the Belastingdienst did not take privacy law seriously.

Read also

• The Butcher as Gatekeeper — why the cabinet keeps the prosecution service out
• The Accountability Matrix — who knew what about the data vault, and who is still there
• The full data vault timeline — the full timeline, 2019-2026
• 64 million files out of sight — the discovery and the nine-month delay

Sources

• Kamervragen kst-31066-1330, Toeslagenaffaire, vergaderjaar 2021-2022
• PwC, Onderzoek effecten FSV MKB, 17 maart 2022
• KPMG, Rapport onderzoek RAM, 25 februari 2025
• Autoriteit Persoonsgegevens, Onderzoek RAM en vergelijkbare systemen, 9 juli 2025
• WRR Working Paper 021, april 2016 (ISBN 978-94-90186-30-2)
• Belastingdienst Algoritmeregister, april 2026