Draft Annex 22: which performance metrics do you need to assess an AI model?
A model can make many correct decisions while missing the errors with the most serious consequences. Translate those consequences into performance metrics and criteria set before testing. The worked example draws on the 2025 draft Annex 22. Its numerical threshold is an example chosen for this exercise.
Which metric demonstrates that an AI model is fit for its intended use?
No single metric establishes suitability. Choose metrics that reflect the consequences of incorrect decisions, intended use and relevant subgroups. Set acceptance criteria before the decisive test and calculate metrics from traceable individual decisions. Draft Annex 22 lists possible metrics. It does not prescribe one universal minimum accuracy for every model.
First, define which error has which consequence for the process.
Evaluate overall performance and relevant subgroups separately.
Define acceptance criteria prior to testing and consider the uncertainty of the sample.
1. Establish the draft's status and the boundaries of intended use
On 7 October 2026, Annex 22 will continue to be treated as a draft consultation. The consultation ended in 2025, which did not result in an automatic entry into force. The following proposals are intended to prepare a concrete use of human medicine GMP in addition to the existing applicable requirements.
The draft concerns static, deterministic models in critical GMP applications. Turn its list of possible metrics into a test plan for your own use. First define the decision expected of the model and how a person or another process step contributes to it.
This article elaborates on performance assessment. The independence of the test dataset is addressed separately, as a properly calculated metric can still be misleading if based on known or inappropriate data.
Take a specific decision: An image is classified as "defect" or "no defect". A missed defect and a falsely rejected good part have different implications. The department must explain which error is relevant to product quality, further processing, and available inspection capacity.
Define, in addition, what "positive" means in your invoice. In the following example, "positive" means "defect detected". Without this definition, sensitivity and specificity can be correctly calculated but incorrectly interpreted in a professional context.
Define which reference determines the actual state. Uncertain or disputed reference evaluations constitute a separate quality issue. They should not be adjusted solely to align with the model outcome, merely to improve the key performance indicator appearance.
3. Select additional metrics instead of a total rate
The draft lists measures including sensitivity, specificity, accuracy, precision and F1. Choose measures suited to your decision and define their numerators and denominators. Explain which type of error each metric reveals.
The following table uses standard definitions for binary classification. TP denotes detected defects, FN denotes missed defects, FP denotes false positives (defective parts incorrectly flagged), and TN denotes correctly accepted good parts. The suitability of these definitions for other applications remains to be assessed on a case-by-case basis.
Work aid: Link metric and statement to accountability
Metric
Calculation
Process question
Sensitivity
TP ÷ (TP + FN)
What proportion of actual defects is detected?
Specificity
TN ÷ (TN + FP)
What proportion of the good batches is correctly not flagged?
Precision
TP ÷ (TP + FP)
What proportion of reported issues are actual defects?
Accuracy
(TP + TN) ÷ Total number
What is the proportion of all correct decisions?
F1
2 × TP ÷ (2 × TP + FP + FN)
How are precision and sensitivity linked in this summary?
4. Sample Calculation: 240 test cases fully evaluated
For the example, 240 independent test cases are assumed: 40 actual defects and 200 non-defective cases. Prior to testing, only two criteria apply to this example: sensitivity of at least 95 percent and specificity of at least 97 percent. The technical appropriateness and statistical validity of this assumption are not established.
The model detects 38 defects and overlooks 2; it also flags 8 good parts; 192 good parts are correctly identified. Thus, 230 out of 240 decisions are correct, amounting to approximately 95.83 percent. Despite this, one of the two assumed criteria is missed.
The matrix preserves the absolute values. Reporting only the overall percentage masks the differential impact of errors. Furthermore, the fulfillment of an observed percentage value would still not constitute sufficient evidence for a safe population-based conclusion.
Example: Results matrix with pre-defined target values
Size
Calculation and value
Assessment in this example
Defects
TP 38 + FN 2 = 40
2 evaluate each overlooked defect individually
Conforming units
TN 192 + FP 8 = 200
8 defective batches
Sensitivity
38 ÷ 40 = 95 %
95% confidence limit achieved
Specificity
192 ÷ 200 = 96 %
Accepted limit 97% failed
Precision / F1
38 ÷ 46 ≈ 82.61 %; 76 ÷ 86 ≈ 88.37 %
Supplementary description, no own acceptance criteria
Overall decision
230 ÷ 240 ≈ 95.83% Accuracy
No acceptance based on overall percentage; specificity criterion not met
A strong overall performance can mask a weak subgroup. In the example calculation, 30 defects could belong to a common defect type and 10 to a rare defect type. If all 30 common defects are detected, but only 8 of the rare defects, the overall sensitivity remains at 38 out of 40, or 95 percent; the rare defect group achieves 80 percent.
The draft describes predefined subgroups and suitable criteria. You need not test every possible combination to the same depth. Justify which differences matter to the intended use, such as material, site, device or failure type.
Avoid retrospective exclusion of only well-performing groups. A weakness identified during testing may serve as a basis for a new hypothesis and further investigation. It should not disappear due to a revised group definition in the final report.
6. Separate decision threshold from acceptance criterion
The decision threshold controls at what model value a case is flagged for investigation. The acceptance limit defines the measured performance level that should be met for the intended use. These are distinct specifications. Optimizing the decision threshold based on review of the final test results affects the independence of the performance evaluation.
Also assess the handling of unresolved or technically defective cases. Such cases must not be silently excluded from consideration. Priorly define how they are classified, investigated, and taken into account in the overall decision.
When a boundary is exceeded, a justified decision follows: investigate root causes, restrict the scope of application, evaluate additional controls, or revise the model. Simply adjusting the boundary post hoc to match the result does not provide credible evidence of suitability.
7. Preparation for a credible performance qualification
Subject Matter Experts justify the consequences of errors and acceptable performance. Data specialists define calculation methods and uncertainty, while QA verifies the traceability of the decision. The roles require identical definitions; a purely technical performance export does not substitute for this alignment.
Define Intended Use, Positive Class, and Reference Assessment.
Identify relevant failure consequences and subgroups.
Define numerator, denominator, and handling of undecided cases.
Justify criteria prior to the critical test from a technical standpoint.
Evaluate the independence, representativeness, and scope of the test data.
Keep absolute case numbers, individual findings and computational path.
Consider failed criteria and remaining uncertainty in the decision.
The conclusion should specify for which intended use any particular statement is made. Identify the version that was reviewed, the key criteria, significant errors, and limitations. A model may be appropriate for a limited scope of application without necessarily demonstrating its suitability for the entire originally intended use.
Link the assessment to subsequent operations. Which metric is monitored further, which subgroup requires attention, and what signal triggers a renewed evaluation? Thus, test results serve as the starting point for controlled use rather than a single success metric.
Examine critical errors behind the overall performance rate
High overall accuracy does not establish that a model suits your process. With process owners, we distinguish missed defects, unnecessary rejections and relevant subgroups. Data owners explain how the test set was assembled and which cases it lacks. A narrowly defined use may need a different combination of metrics from multiple sites or product variants. We agree acceptance criteria before the test, then examine whether the results support the intended use or require a narrower scope.
Frequently asked questions about AI performance metrics and acceptance criteria
Does Annex 22 require a specific minimum accuracy?+
The proposed draft does not establish a universal minimum threshold. It outlines the selection of appropriate metrics and pre-defined criteria tailored to the specific application. The numbers presented in this article's example are internal assumptions and must not be construed as official regulatory limits.
Why can 95.83% accuracy in the example of this article not be sufficient?+
Because the previously accepted specificity criterion of 97 percent is not met: 192 out of 200 batches achieve only 96 percent. The overall high rate does not justify this finding. Furthermore, the appropriateness of the criteria themselves must be substantiated.
Is sensitivity always more important than precision?+
This depends on the consequences of errors and additional controls. Undetected defects and further deviations may give rise to different risks. Prioritization should emerge from the specific process, not from a general ranking of key performance indicators.
What does it mean to have a small subgroup with poor performance?+
It may point to a specific usage limit, but with a small sample size it provides a potentially unreliable estimate. Assess error type, data volume, and impact jointly. A good average should not conceal the weakness.
Can we select a different threshold after the test?+
A change is possible but must be evaluated as a development or change decision. The performance observed in the previously assessed test batch cannot automatically constitute an independent conclusion. The further validation evidence should be planned accordingly.
Determine in advance how technical failures or undecided cases are recorded and evaluated. A silent exclusion can embellish the performance. Whether manual processing is permissible and sufficient depends on the defined process.
Do 240 cases provide sufficient statistical reliability?+
No. The volume merely illustrates the invoice. For a reliable conclusion, selection criteria, subgroups, defect frequencies, and required statistical confidence must be applied. An observed percentage does not guarantee future performance.
Which uses are supported by evidence, which criteria have been missed, and what limitations or additional controls are required. A single success rate is insufficient for this assessment. The decision should identify the associated effort and responsibility for outstanding items.
Source date: 7 October 2026. Annex 22 remains a consultation draft. All example quantities and limit values are assumptions, not regulatory requirements or claimed test results. No statistical sampling advice is provided for a specific product.
Translate performance metrics into traceable decisions
We support you in consolidating root cause analyses, metrics, and acceptance criteria for your intended use and in clearly documenting the limitations of the evidence.
Bring your AI system into GxP use with traceable evidence and keep it under control in operation. Once the system and intended GxP use are defined, we validate according to risk, addressing data, supplier, tests, release and change control.
VAT identification number pursuant to § 27a German VAT Act (UStG): DE328451232
Consumer dispute resolution (VSBG)
In accordance with § 36 VSBG: we are neither willing nor obliged to participate in dispute resolution proceedings before a consumer arbitration board.
Liability for content
We are responsible for our own content on these pages under the general laws. Under Articles 4 to 8 of Regulation (EU) 2022/2065 (Digital Services Act, DSA), as a service provider we are not obliged to monitor transmitted or stored third-party information or to investigate circumstances that indicate unlawful activity.
Note: The §§ 7–10 of the previous Telemedia Act (TMG) on the liability for digital services have not been transferred to the German Digital Services Act (DDG). Liability for digital services is now primarily regulated by Regulation (EU) 2022/2065 (DSA), in particular Articles 4–8 — referenced above. The §§ 7–10 DDG cover other subjects (limited responsibility under DSA Articles 4–8 + Wi-Fi access, blocking claims under copyright law, lists of audiovisual-media-service and video-sharing-platform providers, and information-request rights of the competent authorities under state law).
Copyright
The content and works on these pages created by the site operator are subject to German copyright law. Reproduction, processing, distribution and any kind of exploitation outside the limits of copyright require the written consent of the respective author.
Privacy Policy
1. At a glance
The following information gives a simple overview of what happens to your personal data when you visit this website. Personal data is any data that can be used to identify you personally.
2. Controller
The controller responsible for data processing on this website pursuant to Art. 4 (7) GDPR is:
Daniel Herrmann Consulting Daniel Herrmann (sole proprietor) Enzweilerweg 3a · 66709 Weiskirchen · Germany Phone: +49 170 7878065 Email: info@daniel-herrmann.io
A data protection officer has not been appointed; the controller is your point of contact for all data protection matters.
3. Server logs
Our website is hosted by Kinsta Inc., 8605 Santa Monica Blvd #92581, West Hollywood, CA 90069, USA. When you visit, Kinsta processes your IP address, the time of the request, the page requested, the referring page and browser information. This helps keep the site secure and diagnose faults (Article 6(1)(f) GDPR). The site is hosted in Frankfurt, Germany. Kinsta keeps server logs for up to four days. It also analyses access log data, including IP addresses, in the US to operate and secure its services. It keeps this analytics data for up to 30 days. Kinsta’s data processing addendum includes the EU standard contractual clauses.
3a. Cookie-free page request counts
We count page requests using a small image served by this website. Each request increments a counter for the UTC day and page path. The counter stores no IP addresses, user agents, referrers, query parameters or visitor identifiers, sets no cookies and reads no browser storage. Known automated requests and prefetches are filtered using request headers without storing those headers. Counts include repeated loads and requests without statistics consent; they cannot identify individual visitors or measure their time on the site. Aggregates are removed after 90 days. Technical hosting logs are separate, as described above. The purpose is to understand the use of our public pages; the legal basis for the technical request processing is our legitimate interest under Art. 6(1)(f) GDPR in a data-minimising usage overview.
4. Contact
If you contact us via form, email or phone, we store the information you provide (name, email, telephone number, company, message content) to process the enquiry and for follow-up questions. Legal basis: Art. 6 (1) (b) GDPR (pre-contractual measures) and Art. 6 (1) (f) GDPR (legitimate interest in answering enquiries). Retention: deletion as soon as the enquiry has been finally processed, at the latest after 12 months, unless statutory retention obligations apply.
Our English project enquiry form emails your details to our team. It asks about your role, company size, the support you need and your timeframe. We do not send these enquiries to Calendly.
5. Practical guides and worksheets
For our other practical guides (e.g. Strategy Guide, Compliance-Gap Analysis, Execution Playbook), we ask for your name and email address. Your company is optional. The data is stored solely to provide the requested material and for documentation of consent. Legal basis: Art. 6 (1) (a) GDPR (consent). Retention: until your withdrawal, at most 24 months. The newsletter subscription is offered as a separate, opt-in checkbox on the lead-magnet form (no coupling — you can request the material without subscribing). No marketing emails are sent without separate consent.
For the GAMP 5 practice package, name, company, business email address and telephone number are required. We use these details to process your business enquiry, make the short telephone call you requested and personally send the documents afterwards. There is no automatic download in this process. The information under Contact applies to handling this enquiry. A newsletter subscription remains separate and optional.
The same manual enquiry process applies to the German worksheet packages on Annex 22, CSV project planning, CSA/CSV, Annex 11 and ALCOA+. Name, company, business email and telephone number are required for the requested short call and personal delivery of the worksheets. The information under Contact applies. Submitting the form does not verify your affiliation with a company or your telephone number, does not provide an automatic PDF download and does not subscribe you to a newsletter. This worksheet notice was added on 26 October 2026.
5a. Newsletter (Double-Opt-In)
If you sign up for our newsletter, we use the double-opt-in procedure: after entering your email and name you receive a confirmation email with a verification link. Only after you click the link do we activate the subscription and document your consent. Data: email, name, consent timestamp, confirmation IP. Purpose: sending our monthly newsletter with practice insights, plus documentation of the consent process (Art. 7 (1) GDPR). Legal basis: Art. 6 (1) (a) GDPR (consent), § 7 (2) Nr. 2 UWG. Retention: for as long as your subscription is active; consent logs retained for 3 years after the end of the subscription as evidence in the event of complaints. Withdrawal: every newsletter contains a one-click unsubscribe link in the footer. Alternatively a short email to info@daniel-herrmann.io is sufficient. Processor: the newsletter is sent via our hosting provider's SMTP infrastructure; we currently do not use a third-party email-marketing tool. As soon as we transition to a dedicated newsletter-tool (e.g. Brevo), we will update this section and conclude an additional data-processing agreement.
5c. Link-click tracking in our emails
To understand which content in our lead-magnet emails and newsletter issues is actually useful, we route outbound links through an internal redirect endpoint (`/t/`). Each link in each send gets an aggregate click counter — but we do NOT store an identifier of the individual recipient, NO IP address and NO user agent. The data we keep is comparable to anonymous server logs (e.g.: ‘link X in send Y was clicked 42 times in total’). It does not allow us to infer who clicked. Therefore no separate consent or cookie banner is required (recital 26 GDPR — anonymous data, no personal-data processing per Art. 4 (1) GDPR). Legal basis for the redirect itself: Art. 6 (1) (f) GDPR (legitimate interest in measuring content effectiveness for our own communications). Legal pages (Impressum, Datenschutz) and the unsubscribe link are NEVER routed through the redirect, so the click behaviour on legally relevant links is never measured.
5b. Retention overview & justifications
A consolidated view of all retention periods with the rationale behind each one — the principle: minimal storage, clear purpose, documented basis.
Data categoryRetentionJustification
Server logsUp to 4 days for server logs and up to 30 days for analytics dataSite operation, security and troubleshooting. See section 3.
Contact enquiries12 monthsB2B sales cycle in pharma typically runs 6–9 months. 12 months covers follow-ups without unnecessary stockpiling. Statutory retention (e.g. § 257 HGB) only applies once a contract is concluded.
Lead magnetsup to 24 monthsUntil withdrawal of consent; 24-month cap covers documentation of consent (Art. 7 GDPR) and re-engagement cycle. Withdrawal at any time, deletion within 7 days of request.
Newsletter subscriptionactive subscription + 3 yrs consent logEmail + name only as long as the subscription is active. Consent log (timestamp + IP) retained 3 years to cover statutory limitation period (§ 195 BGB) for complaints under § 7 UWG.
Cookies (incl. analytics)30 min – 14 monthsPer-tool detailed in 6.9. GA4 capped at 14 months (the GA4 minimum, shorter possible only via property reset); all other tools below or equal to industry standard.
Cookiebot consent12 monthsMaximum window the EDPB considers reasonable for repeat-consent requests. After 12 months a fresh banner appears.
6. Cookies, analytics & tracking
This website uses analytics and tracking tools that go beyond pure reach measurement and create usage profiles. Personal data may be processed (in particular IP address, device and browser information, behaviour data) and transferred to third countries (including the USA). Legal basis is your consent pursuant to § 25 (1) TDDDG and Art. 6 (1) (a) GDPR. The consent is granted on first visit via our cookie banner and can be withdrawn at any time with effect for the future — see section 6.10 below. Withdrawal is as simple as granting consent (Art. 7 (3) GDPR); the lawfulness of processing carried out on the basis of consent prior to withdrawal remains unaffected.
6.0 Consent management (Cookiebot)
Provider: Cybot A/S, Havnegade 39, 1058 Copenhagen, Denmark (a Usercentrics company) — EU-based. Purpose: Cookiebot is the consent management platform (CMP) we use to document your consent in line with § 25 TDDDG and Art. 7 GDPR and to block analytics / tracking tools until you opt in. Data: Cookiebot stores a ‘CookieConsent’ cookie containing your consent state (categories, timestamp, anonymous identifier) and a server-side consent log. Retention: up to 12 months from the moment consent is given; renewed on each new visit after expiry. Legal basis: the processing of consent itself rests on Art. 6 (1) (c) GDPR (legal obligation to document consent). Third-country transfer: primary processing in the EU; sub-processors may include service providers outside the EEA under EU Standard Contractual Clauses. Data processing agreement: concluded with Cybot A/S. More info:Cookiebot Privacy Policy.
6.1 Google Analytics 4
Provider: Google Ireland Limited, Gordon House, Barrow Street, Dublin 4, Ireland (parent company: Google LLC, USA). Purpose: Tracking tool that goes beyond pure reach measurement. Google uses the collected data for the operation of GA4 and partly for its own purposes (no commissioned processing in the narrow sense for these purposes). Data: pseudonymous identifiers (client ID), IP address (shortened on EU server, see below), page views, dwell time, scroll depth, click events, device and browser information, approximate location (country / region from IP). Retention: up to 14 months for event-level data, then automatic deletion. IP shortening: The IP address is shortened on an EU server before being forwarded to the USA (‘_anonymizeIp’ equivalent in GA4 is active by default). Third-country transfer: data is processed on servers in the USA. Safeguards: EU Standard Contractual Clauses and the EU-US Data Privacy Framework (Adequacy Decision of 10 July 2023). Note that the supervisory authorities point out that the legal certainty gained may only be temporary; previous adequacy regimes (Safe Harbor, Privacy Shield) were invalidated by the CJEU. Data processing agreement: concluded with Google Ireland Ltd. More info:Google Privacy Policy, Browser opt-out.
6.2 Microsoft Clarity
Provider: Microsoft Ireland Operations Ltd., One Microsoft Place, South County Business Park, Leopardstown, Dublin 18, Ireland (parent: Microsoft Corporation, USA). Purpose: Session recordings and heatmaps to improve usability. Note: session recordings are a particularly intensive form of processing — they can capture sensitive content even though input fields are masked by default. We have configured the strictest masking level (‘Strict’) for all form fields. Data: mouse movements, clicks, scroll behaviour, page views, device and browser information, shortened IP address, country. Retention: up to 12 months from the last recorded session. Third-country transfer: Microsoft is a US group; data may be processed on servers in the USA. Safeguards: EU Standard Contractual Clauses and the EU-US Data Privacy Framework. The reservations noted above for GA4 apply analogously. Data processing agreement: Microsoft Online Services DPA concluded. More info:Microsoft Privacy Statement.
6.3 Leadfeeder (Dealfront)
Provider: Dealfront Germany GmbH (Leadfeeder), Markgrafenstraße 36, 10117 Berlin, Germany — EU-based provider. Purpose: B2B identification of visiting companies based on commercially licensed IP-address databases (e.g. RIPE, ARIN, public corporate IP ranges). We use this to inform our sales outreach. The aim is identification of the company, not of an individual user. Important note from a data protection perspective: IP addresses can constitute personal data, particularly when combined with other data. We therefore treat Leadfeeder processing as relevant under GDPR. Data: IP address, page views, timestamp, dwell time, referrer. Data sources: Dealfront enriches IP data with publicly available company information and licensed B2B databases. Retention: up to 12 months on visit level; aggregated reports may be retained longer. Third-country transfer: Dealfront operates primarily on EU infrastructure; sub-processors may include US service providers under EU Standard Contractual Clauses. Data processing agreement: concluded with Dealfront Germany GmbH. More info:Dealfront Privacy Policy.
6.4 Google Ads
Provider: Google Ireland Limited, Gordon House, Barrow Street, Dublin 4, Ireland (parent: Google LLC, USA). Purpose: We use Google Ads conversion tracking and (where activated) remarketing to measure the success of advertising campaigns and to address visitors with relevant ads. Data: pseudonymous click ID (gclid), conversion event, conversion timestamp, device and browser information. Cookies used include `_gcl_au` (conversion linker) and `test_cookie` (doubleclick.net technical test cookie). Retention: `_gcl_au` up to 90 days, conversion logs in the Google Ads account according to Google retention settings. Third-country transfer: data is processed on Google servers including the USA. Safeguards: EU Standard Contractual Clauses and the EU-US Data Privacy Framework. Data processing agreement: concluded with Google Ireland Ltd. More info:Google Privacy Policy, Ad personalisation settings.
6.5 LinkedIn Insight Tag
Provider: LinkedIn Ireland Unlimited Company, Wilton Plaza, Wilton Place, Dublin 2, Ireland (parent: LinkedIn Corporation, USA). Purpose: The LinkedIn Insight Tag enables campaign measurement, audience analytics and (where activated) retargeting for LinkedIn ad campaigns. Data: IP address (truncated), timestamp, page URL, device characteristics, LinkedIn member ID where the visitor has been logged in to LinkedIn. Cookies used include `bcookie`, `lidc`, `bscookie`. Retention: up to 6 months for direct identifiers; aggregated campaign reports may be retained longer. Third-country transfer: data is processed on LinkedIn servers including the USA. Safeguards: EU Standard Contractual Clauses and the EU-US Data Privacy Framework. Data processing agreement: concluded with LinkedIn Ireland Unlimited Company. More info:LinkedIn Privacy Policy, LinkedIn opt-out.
6.5a Meta Pixel (Facebook Pixel)
Provider: Meta Platforms Ireland Ltd., Merrion Road, Dublin 4, D04 X2K5, Ireland (parent: Meta Platforms, Inc., 1601 Willow Road, Menlo Park, CA 94025, USA). Purpose: The Meta Pixel (also ‘Facebook Pixel’) enables reach measurement, conversion tracking and (where activated) retargeting for advertising campaigns on Facebook and Instagram. It lets us measure the effectiveness of our ads and show relevant advertising to visitors. Data: pseudonymous identifiers, IP address, device and browser information, referrer URL, triggered events (e.g. page view). Cookies used include `_fbp` (first-party cookie used to recognise the browser) and `_fbc` (click identifier derived from the `fbclid` URL parameter). If the visitor was logged in to Facebook / Instagram at the time of the visit, Meta can attribute the visit to the respective account. Joint controllership: for the collection and transmission of the data to Meta there is joint controllership within the meaning of Art. 26 GDPR between us and Meta; we have concluded Meta’s controller addendum for this. Meta is solely responsible for any subsequent processing of the data for its own purposes. Retention: `_fbp` up to 90 days; on the event level according to the retention settings in the Meta Events Manager. Third-country transfer: data is processed on Meta servers including the USA. Safeguards: EU Standard Contractual Clauses and the EU-US Data Privacy Framework (Adequacy Decision of 10 July 2023). The reservations noted above for GA4 apply analogously. Legal basis: your consent pursuant to § 25 (1) TDDDG and Art. 6 (1) (a) GDPR. The pixel is only loaded after you have given consent via our Cookiebot banner. More info:Meta Privacy Policy, Meta ad settings.
6.6 Google Tag Manager
Provider: Google Ireland Limited (see 6.1). Purpose: Google Tag Manager (GTM) is a tag-management system that loads other tracking tags (e.g. Google Analytics, Google Ads, LinkedIn Insight Tag) on this website. GTM itself does not set any cookies and does not collect personal data per se; it only orchestrates the tags loaded after consent. Data: technical request data (IP address, timestamp, user agent) is briefly seen by Google’s GTM-loader server. No persistent identifier is set by GTM itself. Third-country transfer: the GTM loader is hosted on Google infrastructure including the USA. Safeguards as in 6.1. Note: we have configured GTM so that no measurement tag fires before you have given consent via our Cookiebot banner. More info:Google Privacy Policy.
6.7 Cloudflare (CDN & bot management)
Provider: Cloudflare, Inc., 101 Townsend Street, San Francisco, CA 94107, USA, with EU representative Cloudflare Germany GmbH, Rosental 7, c/o Mindspace, 80331 Munich. Purpose: Cloudflare is used by our hosting provider as a Content Delivery Network (CDN) and security layer. The ‘__cf_bm’ cookie is set by Cloudflare’s bot-management to distinguish humans from automated traffic and to mitigate DDoS attacks and content scraping in real time. Why we consider this technically necessary: Without bot-mitigation our website would be vulnerable to credential-stuffing, scraping and DDoS attacks that compromise availability, integrity and the security of the personal data we process (e.g. lead-magnet form submissions, newsletter signups). Bot-mitigation is therefore a technical security measure within the meaning of Art. 32 GDPR (security of processing). Cloudflare itself classifies ‘__cf_bm’ as ‘strictly necessary’ (Cloudflare cookie documentation). Replacing Cloudflare bot-management would weaken the security of personal data on this site. Data: IP address, request headers, technical fingerprint of the request. No persistent identifier across sites. Retention: ‘__cf_bm’ cookie expires after at most 30 minutes of inactivity. Aggregated security logs at Cloudflare are retained briefly under Cloudflare’s standard retention. Legal basis: § 25 (2) Nr. 2 TDDDG (technically required to operate the requested service) and Art. 6 (1) (f) GDPR (overriding legitimate interest in a secure, available website). Third-country transfer: Cloudflare is a US group; data may be processed on servers in the USA. Safeguards: EU Standard Contractual Clauses and the EU-US Data Privacy Framework (Adequacy Decision of 10 July 2023). The EU General Court dismissed the challenge to the adequacy decision on 3 September 2025 (T-553/23). An appeal was filed (C-703/25 P). Data processing agreement: concluded as part of the hosting contract. More info:Cloudflare Privacy Policy.
6.8 Google Search Console
Provider: Google Ireland Limited (see 6.1). Purpose: Google Search Console (GSC) lets us monitor how the site performs in Google’s search results — indexing status, search queries that lead to the site, click-through rates, technical crawl errors. Cookies / tracking on visitors: GSC itself sets no cookies on visitors of this website and runs no client-side script. The site ownership is verified by a static HTML meta tag and / or a DNS TXT record. Data: aggregated search analytics from Google’s side (search queries, impressions, clicks); these are not directly linked to individual visitors. The data is made available to us only in aggregated form. Legal basis: Art. 6 (1) (f) GDPR — legitimate interest in monitoring the site’s search visibility. Third-country transfer: the aggregated reporting is generated on Google infrastructure including the USA. Safeguards as in 6.1. More info:Google Privacy Policy.
6.9 Cookie storage durations
Session cookies are deleted when the browser is closed. Permanent cookies have the following maximum lifetimes: Google Analytics 4 — up to 14 months (data retention shortened to 14 months in GA4 settings); Microsoft Clarity — up to 12 months; Google Ads (`_gcl_au`) — up to 90 days; Meta Pixel (`_fbp`) — up to 90 days; LinkedIn (`bcookie`) — up to 6 months; Cloudflare (`__cf_bm`) — at most 30 minutes of inactivity; Leadfeeder — typically session-based, not stored on the device; CookieConsent (Cookiebot) — up to 12 months from the moment consent is given.
6.10 Withdrawal of consent
You can withdraw your consent at any time with effect for the future — withdrawal is as easy as granting consent:
1. Open cookie settings: click the ‘Cookie settings’ link in the footer to re-open the banner and adjust your choice (this controls all of GA4, Clarity, Google Ads, LinkedIn, Meta Pixel and Leadfeeder). 2. By email: send a short message to info@daniel-herrmann.io — we will block the relevant tools for you. 3. Per tool: for Google Analytics the official browser opt-out is available; for Google Ads under Ad personalisation settings; for LinkedIn under LinkedIn opt-out; for the Meta Pixel under Meta ad settings. Microsoft Clarity and Leadfeeder are controlled exclusively via our cookie banner.
The lawfulness of processing carried out on the basis of consent prior to withdrawal remains unaffected (Art. 7 (3) GDPR).
6.11 Appointment booking with Calendly
We use Calendly LLC, 115 E Main St, Ste A1B, Buford, GA 30518, USA, for appointments on our German booking and compliance check pages. The embedded calendar loads when you explicitly allow preferences, statistics and marketing in our cookie banner. Your choice also applies when you return. You can change it at any time in Cookie settings. You can also arrange an appointment by email or follow the direct link to Calendly, where its own cookie settings apply.
When the calendar loads, Calendly and its service providers receive technical data such as your IP address and browser information. They may use cookies and other browser storage to run the calendar, protect it against abuse, analyse usage and support marketing. Your consent covers optional storage and the associated processing under section 25(1) TDDDG and Article 6(1)(a) GDPR. Under Calendly’s terms, we and Calendly each act as independent controllers for data collected through the embedded cookie technologies.
When you book, we process your name, email address, company, answers to the booking questions and chosen time. We use these details to prepare the meeting and send your Microsoft Teams invitation. The legal basis is Article 6(1)(b) GDPR for steps you request before entering into a contract, or Article 6(1)(f) GDPR for our legitimate interest in responding to a request on behalf of your company. We keep booking details for as long as needed to organise the meeting and handle follow-up correspondence. Statutory retention periods may also apply.
Calendly’s data processing addendum forms part of its customer terms. It covers the booking data Calendly processes on our behalf in the US. For EU-US transfers it provides for the EU-US Data Privacy Framework, with EU standard contractual clauses if that framework ceases to apply. See Calendly’s privacy notice for more information.
7. Web fonts
The fonts used on this website are served locally. Loading these fonts does not connect your browser to Google or any other font provider.
8. Technical and organisational measures (TOM)
To protect your data we implement appropriate technical and organisational measures pursuant to Art. 32 GDPR — among them: TLS encryption (HTTPS) for the entire site, encrypted database connections, server hosting in the EU (Kinsta / GCP Frankfurt), restricted admin access via individual accounts with strong passwords and 2FA, regular automatic backups, role-based access controls, a documented record of processing activities pursuant to Art. 30 GDPR, and data minimisation at the application level. We continuously review and update these measures.
9. Your rights
Under the GDPR you have the following rights:
Right of access (Art. 15 GDPR) — information about which of your personal data we process.
Right to rectification (Art. 16 GDPR) — correction of inaccurate data.
Right to erasure (Art. 17 GDPR) — deletion of your data where the legal conditions are met.
Right to restriction of processing (Art. 18 GDPR).
Right to data portability (Art. 20 GDPR) — receipt of your data in a structured, common, machine-readable format.
Right to object (Art. 21 GDPR) — to processing based on legitimate interest, including for direct marketing. See highlighted box below.
Right to withdraw consent (Art. 7 (3) GDPR) — at any time with effect for the future, see section 6.10.
Right to lodge a complaint (Art. 77 GDPR) — with a supervisory authority, in particular in the EU member state of your residence, your workplace or the location of the alleged infringement. Competent supervisory authority for our office location: Independent Data Protection Centre Saarland (Unabhängiges Datenschutzzentrum Saarland), Fritz-Dobisch-Straße 12, 66111 Saarbrücken.
Requests regarding these rights are to be directed to: info@daniel-herrmann.io. We will respond within the statutory period (usually one month).
10. Currency of this Privacy Policy
Last updated: 22 November 2026. We reserve the right to adapt this policy so that it always meets current legal requirements. The current version is always retrievable from this website.
Free material by email
Your material is on the way.
Enter your name and email, we send the material directly to your inbox, with a short personal note from Daniel.
Thanks, check your inbox.
We just sent the material to your email. If it doesn't arrive within a few minutes, please check your spam folder.
Once a month · CSV practice
CSV insights from real Pharma projects.
What you get next: the GAMP 5 2nd Edition impact, what the FDA CSA Final Guidance means for your validation strategy, and how the 8-week CSV sprint works. Unsubscribe anytime in one click.
Almost done.
Confirm your email via the link we just sent to your inbox. After confirmation we'll send the latest newsletter issue right away.
We use your data exclusively for the newsletter. No sharing with third parties. Privacy policy.