Definition: Technical and Organizational Measures (TOM)
Technical and organizational measures (TOM) are the concrete security and governance safeguards, such as encryption, access control, pseudonymization, and staff training, that a data controller or processor must implement under Article 32 of the GDPR to protect personal data against unauthorized access, loss, or misuse.
Core characteristics of TOM
TOM combine technical safeguards built into IT systems with organizational rules that govern how people and processes handle data. Both layers must match the risk of the processing activity, not a generic checklist.
- Risk-based selection tied to the sensitivity and volume of processed data
- A mix of technical controls (systems, encryption, backups) and organizational controls (policies, training, access rights)
- Documented in writing, most commonly as a TOM-Anhang attached to a Data Processing Agreement
- Reviewed and updated whenever processing purposes, vendors, or systems change
TOM vs. Data Processing Agreement
A Data Processing Agreement is the contract that defines the legal relationship between a controller and a processor, including scope, duration, and liability. TOM are the technical annex to that contract: a table-based description of exactly which safeguards the processor runs in practice. The DPA states that data must be protected; the TOM document states how, system by system.
Importance of TOM in enterprise AI
TOM have become a central audit point as companies connect AI tools to customer and employee data. According to the CMS GDPR Enforcement Tracker, insufficient technical and organizational measures were behind close to 29 percent of all GDPR fines issued in 2025, with the category’s fine volume exceeding 40 million euros that year. For any company rolling out AI agents, chatbots, or automation, TOM define whether that deployment can pass a GDPR audit.
Methods and procedures for TOM
Building a TOM document follows a structured, repeatable process rather than a one-off checklist exercise.
Risk-based measure selection
Article 32 does not list fixed measures; it requires controllers to weigh the state of the art, implementation cost, and the risk to individuals. A DPIA often precedes this step for higher-risk processing, since it identifies which risks the TOM must actually address.
- Classify data by sensitivity (personal, special category, trade secret)
- Map which systems and third parties process that data
- Match each identified risk to a specific technical or organizational control
Technical measures
Technical measures live inside the IT stack: encryption, pseudonymization, role-based access control, multi-factor authentication, and automated logging. These controls are testable and are the ones external auditors verify first.
Organizational measures
Organizational measures govern people and process: a documented data protection policy, mandatory staff training, an incident response procedure, and clear subprocessor contracts. These matter as much as technical controls, since most incidents originate from human error, not a broken system.
Important KPIs for TOM
TOM are tracked through both compliance and operational metrics.
Operational security metrics
- Access review frequency: quarterly or more often
- Mean time to detect a security incident: under 24 hours
- Staff data protection training completion: 100% annually
- Encryption coverage across systems handling personal data: 100%
Strategic compliance metrics
Beyond individual controls, companies track how current their TOM documentation is against their actual system landscape. Bitkom’s 2025 survey on data protection in business found that the technical implementation of DSGVO requirements remains one of the two largest effort blocks German companies report, alongside internal process documentation.
Quality and audit metrics
A well-maintained TOM document should pass a vendor security questionnaire without follow-up questions and should be revisited every time a new subprocessor, cloud region, or AI tool enters the processing chain.
Risk factors and controls for TOM
Weak TOM documentation is one of the most common gaps supervisory authorities find.
Outdated documentation
A TOM-Anhang written once during onboarding and never revisited is the most common audit finding, because companies add SaaS tools and AI vendors faster than they update their records.
- New cloud services added without a TOM review
- AI tools connected to production data without updated access logging
- Subprocessor chains that grow without new assessments
Gap between documented and actual measures
A document that lists controls the company does not actually run is treated as a compliance failure, not a technicality. Authorities increasingly ask for evidence, not just a description.
AI-specific coverage gaps
Many TOM documents predate the company’s AI tools and skip model access, prompt logging, or output retention. Extending the TOM to AI usage is a deliberate part of an AI compliance program, not an afterthought.
Practical example
A 45-person tax advisory firm in Hamburg needed a TOM-Anhang before signing with a new AI-based document processing vendor. The firm’s Data Protection Officer had only a two-year-old template listing a since-retired VPN setup, with no mention of the new cloud backup provider. Working with the vendor’s security team, the firm rebuilt its TOM, mapped every subprocessor, and tied access rights to individual client mandates instead of blanket partner access.
- Encrypted client file storage with per-mandate access scoping
- Quarterly access reviews tied to the firm’s HR offboarding checklist
- A documented 72-hour breach notification procedure
- A subprocessor register reviewed before each new tool onboarding
Current developments and effects
TOM requirements are expanding as AI tools multiply inside German companies.
AI tool sprawl widens TOM scope
Employees connecting AI assistants to email, CRM, and shared drives create processing activities the original TOM never anticipated.
- TOM documents increasingly include a dedicated AI systems section
- Vendor security questionnaires now ask directly about model training use of customer data
- Logging requirements extend to prompts and generated outputs, not just stored files
Standardization through recognized frameworks
Certifications such as ISO 27001 and attestations like BSI C5 are becoming the shortcut German vendors use to prove their TOM to business customers without a bespoke questionnaire each time.
Rising enforcement pressure
Supervisory authorities across the EU are issuing more fines specifically tied to Article 32 shortfalls, pushing TOM from a back-office document into a board-level compliance item.
Conclusion
Technical and organizational measures translate the abstract promise of data protection into a concrete, auditable list of safeguards. They sit at the intersection of IT security and compliance documentation, and they only hold up when they describe systems as they actually run, not as they ran during onboarding. As AI tools extend into more business processes, the TOM document becomes one of the fastest ways to see whether a company’s data protection practice is current or inherited. Treating it as a living document, reviewed with every new vendor or AI deployment, is what keeps it useful rather than decorative.
Frequently Asked Questions
What is the difference between TOM and a Data Processing Agreement?
A Data Processing Agreement defines the legal terms of a processing relationship, while TOM describe the specific safeguards the processor actually runs. TOM are usually attached to the DPA as a separate annex, often called a TOM-Anhang in German contracts.
Does Article 32 GDPR require specific measures?
No. Article 32 requires a risk-based approach, weighing the state of the art, implementation cost, and the risk to individuals. This is why TOM documents vary significantly between a small advisory firm and a manufacturer.
Is a TOM-Anhang mandatory for small and medium businesses?
Yes. Any company acting as a controller or processor under GDPR must maintain TOM documentation regardless of size, though scope and complexity should scale with the actual risk and volume of data processed.
How does TOM documentation relate to AI tool adoption?
Connecting AI agents to customer or employee data creates a new processing activity the existing TOM should cover, including access logging, retention for prompts and outputs, and assessment of the AI vendor itself.
What happens if our TOM documentation is outdated during an audit?
Authorities treat a mismatch between documented and actual measures as a compliance failure, not a formality. Outdated TOM documentation was a factor in close to 29 percent of GDPR fines issued in 2025.
Do we need external help to build or update our TOM documentation?
Most Mittelstand companies work with their Data Protection Officer or an external consultant on the initial TOM document, then update it internally whenever a new vendor or AI tool joins the processing chain.