Data Processing Agreement (AVV)

1. Parties, scope and duration

This Data Processing Agreement (DPA / Auftragsverarbeitungsvertrag, AVV) forms part of the agreement between Baoks UG (haftungsbeschränkt), Kruppstraße 41, 72760 Reutlingen, Germany ("Provider"; privacy contact: info@baoks.de), and the business that accepts it in the checkout or in the app, or that is identified in a signed order ("Customer"). It applies to personal data that Provider processes on Customer's behalf in providing Holy Quack ("Customer Personal Data"). It remains in force for as long as Provider processes Customer Personal Data, including any return and deletion period.

Customer acts as controller, or as a processor authorized by its controller to engage Provider as a subprocessor. Provider acts as processor or subprocessor accordingly. Customer is responsible for the lawfulness of its instructions and for any notices, permissions and authority needed to supply the data. Provider remains responsible for its own legal duties. Account administration, sign-in and activity records, invoicing, contract records and Provider's own legal compliance are Provider's own controller activities and are described in the Privacy Policy. This DPA covers the content that users create or retrieve in the workspace, the Amazon Data and the related logs. Aggregate information that does not relate to an identifiable person is not personal data merely because it appears in a campaign report.

2. Instructions and confidentiality

Provider processes Customer Personal Data only on Customer's documented instructions, including with regard to transfers to third countries. Instructions are given by this DPA, the service configuration Customer chooses and written instructions from Customer's authorized contacts. If Union or Member State law requires other processing, Provider informs Customer before processing, unless that law prohibits such information on important grounds of public interest. Provider informs Customer without delay if it considers that an instruction infringes data protection law, and may suspend that instruction until the parties have resolved the matter.

Only persons who need access to provide the service may access Customer Personal Data. Provider commits them to confidentiality or ensures they are under an appropriate statutory obligation of confidentiality. Staff open a Customer workspace in the app only in time-limited, logged support sessions at Customer's request or where necessary to investigate a security incident or malfunction. Outside such sessions, only the managing director and named staff access Customer Personal Data, through personal accounts with multi-factor authentication, and only where necessary to operate, secure and troubleshoot the service and to check the quality of proposals and models for Customer. Additional work outside the agreed scope requires prior agreement, and any reasonable charges for it must not obstruct the assistance required by law.

3. Security and assistance

Provider implements the technical and organizational measures required by Art. 32 GDPR, taking into account the nature, scope, context and purposes of processing and the risks to individuals. Annex II describes these measures. Provider may adapt them to technical progress, provided the agreed level of protection is not reduced.

Taking into account the nature of processing and the information available to it, Provider assists Customer with data subject requests and with Customer's obligations under Articles 32 to 36 GDPR (security, breach notification, data protection impact assessments and prior consultation), and with equivalent obligations under applicable US state privacy laws. Provider forwards requests concerning Customer Personal Data to Customer without undue delay and does not respond to them itself unless Customer instructs it to or the law requires it. Provider provides information early enough for Customer to meet its deadlines.

4. Personal data breaches

Provider notifies Customer without undue delay, and no later than 48 hours, after becoming aware of a personal data breach affecting Customer Personal Data. Notice goes to the email address of the accepting owner or to a privacy contact Customer designates later. Notification does not wait for a completed investigation or for any third party's approval. Provider provides the available information on the nature of the breach, the categories and approximate number of data subjects and records concerned, the likely consequences, a contact point and the measures taken or proposed. Information that is not yet available is provided in stages without undue delay.

Provider takes reasonable steps to contain and remedy the breach, preserves relevant evidence and cooperates with Customer. Customer decides on notifications to authorities and data subjects, unless the law places a direct duty on Provider. A notification is not an admission of liability. Provider's notification duties towards Amazon apply in addition and do not delay notices required by law.

5. Subprocessors

Customer gives general written authorization for the subprocessors listed in Annex III. Provider informs Customer of any intended addition or replacement of a subprocessor, with its identity, purpose and processing locations, at least 30 days before it begins processing, by email to Customer's privacy contact. Updating a web page alone is not sufficient notice. Customer may object on reasonable data protection grounds within that period.

The parties then look for a reasonable alternative or mitigation. If none is found before processing would start, Customer may terminate the affected service without penalty and receives a pro-rata refund of prepaid fees for the unused period. Where Provider cannot exclude the new subprocessor for Customer alone, Customer may terminate the affected service with immediate effect. For subprocessors engaged by Provider's own subprocessors (for example the inference providers engaged by Vercel), Provider forwards the notices it receives without undue delay, and Customer has the same right to object and terminate. Provider imposes on each subprocessor, by written contract, data protection obligations that are substantially equivalent to those in this DPA, and remains liable to Customer for its performance in accordance with Art. 28(4) GDPR. The same restrictions apply down the processing chain.

6. Information, audits and inspections

Provider makes available all information necessary to demonstrate compliance with Art. 28 GDPR, and allows for and contributes to audits, including inspections, conducted by Customer or an independent auditor mandated by Customer. Existing relevant reports, including those of subprocessors, may be used first where they are sufficient. Provider does not currently hold ISO 27001, SOC 2 or comparable certifications.

Routine audits require reasonable notice and appropriate confidentiality, and must protect other customers and the security of the service. These conditions must not prevent an effective audit, restrict a competent authority or delay an urgent audit following an incident or concrete indications of non-compliance. Each party ordinarily bears its own costs. Any fee for additional assistance must be reasonable, agreed in advance and must not obstruct mandatory rights. Provider bears the cost of remedying its own non-compliance.

7. Return, deletion and retention

At the end of the service, Customer chooses either the return of Customer Personal Data in a common, usable format followed by deletion, or deletion alone. On request, Provider provides one export of all workspace data as CSV or JSON free of charge; Customer can make the request up to 30 days after the end of the service. Provider then deletes all Customer Personal Data and existing copies, unless Union or Member State law requires storage. Customer may give this instruction earlier where the service agreement permits. Provider confirms the deletion on request. Data retained by law is restricted, used for no other purpose and deleted when the retention period ends.

On termination or a deletion request, Amazon syncing and all other processing stop immediately, except what is strictly necessary for return, deletion and legally required retention. Amazon Data in active production systems is deleted without undue delay. Residual encrypted backup copies are deleted automatically within 30 days at the latest. If a backup is restored, Provider re-executes all deletions instructed since the backup was taken before returning the service to normal use, following its documented restore procedure.

Other Customer Personal Data is returned or deleted without undue delay, and no later than 30 days after Customer's end-of-service instruction, except where retention is required by law. Machine-learning models and other artifacts derived from Customer Personal Data are deleted, or retrained without it, within 30 days after Customer's end-of-service or deletion instruction. If Customer switches off cross-customer learning, Provider stops using Customer's contributed statistics and, within 30 days, retrains or retires the cross-customer models trained on them. Customer-specific models remain in use while the service continues, unless Customer opts out of machine-learning training; Provider then stops using them at once and deletes them, or retrains them without Customer's data, within 30 days. Retention is not extended merely because a backup or a subprocessor holds the data. Disconnecting Amazon access alone is not an instruction to delete historical data.

8. International transfers

Provider transfers Customer Personal Data to a third country, including by allowing remote access from a third country, only on Customer's documented instructions and on the basis of a valid mechanism under Chapter V GDPR. The primary database and its backups are stored in the EU (Ireland). Short-lived hosting logs, performance measurements and language-model requests are processed by the US subprocessors in Annex III, which may also access data from the USA. Annex III states the transfer mechanism for each of them. The AI assistant runs either in the application functions in the EU or on Provider's own hardware in Germany, where its working copies are held. Machine-learning models are trained on Provider's own encrypted hardware in Germany. Customer instructs Provider to make these transfers.

Where the EU Standard Contractual Clauses (Commission Implementing Decision (EU) 2021/914) are used, they are concluded in the appropriate module between the relevant exporter and importer, with their annexes completed, together with any necessary transfer impact assessment and supplementary measures. Provider reviews the safeguards regularly and informs Customer without undue delay if it can no longer comply with them. The affected transfer is then suspended until a lawful solution is in place. If none can be found, Customer may terminate the affected service and have its data returned or deleted. This DPA does not cover transfers governed by UK or Swiss law unless the parties agree a specific schedule for them.

9. United States state privacy addendum

To the extent an applicable US state privacy law governs Customer Personal Data, Customer is the business or controller and Provider acts only as its service provider, contractor or processor. Provider processes that information only for the limited and specific purposes described in Annex I, within the direct business relationship, and provides the level of privacy protection required by applicable law, including the CCPA as amended by the CPRA where it applies.

Provider does not sell or share Customer Personal Data as those terms are defined by the CCPA. It does not retain, use or disclose it for any purpose other than the specified business purposes, or outside the direct business relationship, except as the law expressly permits. It does not combine it with personal information received from another person or collected from its own interactions with individuals, except as the law expressly permits and consistent with Customer's instructions. Provider certifies that it understands and will comply with these restrictions.

Provider informs Customer if it determines that it can no longer meet these obligations. Customer may take reasonable and appropriate steps to verify compliant use and, on notice, to stop and remedy unauthorized use, including the assessments, audits and tests required by applicable law. The confidentiality, request assistance, deletion and return, security, breach and subprocessor provisions of this DPA apply equally to such processing.

Carrying out Customer's advertising operations does not permit Provider to use Customer Personal Data for its own cross-context behavioral advertising, sale or independent profiling, or to train third-party AI or language models. The per-customer model training, the comparison of bidding methods and the optional cross-customer learning described in Annex I are part of the business purposes described there. Any other new purpose requires a separate assessment and agreement. This addendum does not cover sector-specific US regimes, and the service is not intended for regulated health, financial or children's data.

10. Affiliates, agencies and enterprise agreements

An affiliate is covered only if a signed order or accession identifies it as a participating legal entity, with its address, role, authorized signatory and processing scope. The signing Customer must be authorized to act for any affiliate it represents. Shared branding, group ownership, a connected Amazon profile or technical workspace access alone does not bind an affiliate. Where Customer is an agency or otherwise acts as a processor, it warrants that its controller has authorized this processing chain, including Provider as a subprocessor, and passes on the controller's instructions.

Each participating controller keeps its statutory rights and responsibilities. The parties may coordinate routine instructions and audits through a lead contact without removing any individual's mandatory rights. An existing individually negotiated DPA or master agreement is not replaced by a user accepting these standard terms. Such customers use the enterprise contracting process at info@baoks.de. Negotiated service levels, liability terms, security schedules and affiliate scope belong in the signed order.

11. Customer responsibilities, liability and precedence

Customer is solely responsible for the lawfulness of the processing under Art. 6 GDPR, for informing data subjects and safeguarding their rights, for its authority to make the data and connected Amazon accounts available, and for its advertising decisions, budgets, bid limits, settings, automation scopes and approvals of proposals made by the service or the AI assistant, including the weekly review of changes described in section 7 of the Terms of Service.

Liability towards data subjects follows Art. 82 GDPR and is not limited by this DPA. Between the parties, Provider bears damage, costs and third-party claims only to the extent they result from Provider breaching the obligations of the GDPR specifically directed to processors or acting outside or contrary to Customer's lawful instructions (Art. 82(2) sentence 2 GDPR). Customer bears all other such damage, costs and claims. Customer indemnifies Provider against third-party claims, including claims by data subjects, Amazon and authorities (except administrative fines), and against reasonable costs of legal defence, to the extent they result from the unlawfulness of processing within Customer's responsibility, its instructions, settings or approvals, or the data it provides, and Provider is not responsible for them.

Otherwise, contractual liability under this DPA, including recourse between the parties, follows the limitations of the Terms of Service. Nothing in this DPA limits liability for intent, gross negligence, injury to life, body or health, fraud, an expressly assumed guarantee or mandatory product liability. Administrative fines under Art. 83 GDPR are borne by the party on which they are imposed, without recourse against the other party unless the fine results solely from the other party's intentional or grossly negligent breach of this DPA. This DPA does not waive data subject rights, regulatory powers or rights under applicable transfer clauses.

Mandatory law and executed transfer clauses prevail over this DPA. This DPA prevails over the Terms of Service in conflicts about Customer Personal Data. A data processing agreement signed with Provider prevails over this DPA. Changes to this DPA require agreement; changes of subprocessors follow section 5. A later change to this web page does not alter a previously accepted version.

Annex I: Description of processing

Subject matter and purposes: providing Holy Quack, namely connecting the Customer's Amazon Ads accounts, importing, storing and analyzing advertising data, calculating bids and rule-based actions, applying changes Customer instructs or authorizes, training, evaluating and running machine-learning models on Customer's advertising performance data only to set bids for Customer's own campaigns, continuously comparing Provider's bidding methods (rule-based methods and these models) by assigning parts of Customer's campaigns to different methods, to find the method that best reaches Customer's targets for its own campaigns, within Customer's budgets, bid limits and automation scope and Provider's safety limits, unless Customer has opted out of the comparison or of model training by email, providing dashboards and reports, providing Amazon Marketing Cloud and Amazon Attribution aggregate reporting and audience configuration, operating the AI assistant, providing user management, support, troubleshooting and security for these services, and, only after Amazon has expressly confirmed in writing that this specific use is permitted for Provider's approved integration and Customer has switched on cross-customer learning, anonymizing and aggregating numeric performance statistics, excluding search terms, keywords, campaign names, ASINs and other identifying content, and excluding Amazon Marketing Cloud and Amazon Attribution data. Cross-customer learning remains switched off and will not be offered or enabled unless and until Amazon has expressly confirmed in writing that this specific use is permitted for Provider's approved integration. Customer opt-in alone does not activate it. Where source data are Customer Personal Data, Provider first anonymizes them on Customer's documented instruction so that no natural person is identifiable by means reasonably likely to be used. Provider documents re-identification and model-extraction testing; data that do not meet this standard are excluded from cross-customer learning. The resulting statistics are used to improve Provider's bidding models and are never shown to other customers. The anonymization is carried out on Customer's instruction; after anonymization, Provider uses the statistics as its own controller.

Use of AI and machine learning: Provider uses AI in the service. Provider trains, evaluates and runs machine-learning models on Customer's data only to set bids for Customer's own campaigns. It does not combine Customer's data with other customers' data, except through the optional cross-customer learning described above, which is currently inactive, and it does not use Customer's data to train third-party AI or language models. Models are deleted, or retrained without Customer's data, within 30 days after termination or a deletion instruction (section 7). Training takes place on Provider's own encrypted hardware in Germany.

Nature of processing: collection, retrieval, organization, storage, analysis, use for training and evaluating machine-learning models, anonymization and aggregation, transmission to recipients Customer instructs (in particular Amazon), restriction, export and deletion.

Duration and frequency: for the term of the agreement and the return and deletion period. Scheduled and user-initiated processing according to the functions Customer enables.

Data subjects: Customer's authorized users, employees and contractors, contacts in the advertising accounts and support correspondents; Customer's owners or partners where Customer is a natural person or partnership. Other individuals, such as shoppers, only where their personal data happens to appear in text or metadata Customer supplies or Amazon reports.

Categories of data: names, business contact and role data; account and profile identifiers and authorization metadata; encrypted Amazon sign-in tokens; campaign, keyword, search-term and targeting configuration; performance reports; aggregate Amazon Marketing Cloud and Amazon Attribution metrics and audience configuration; messages to the AI assistant, its replies and account notes; support content; access and action logs. Campaign and aggregate metrics are often not personal data. This DPA applies to them where they relate to an identifiable person. Provider does not need shopper-level identifiers and does not receive them.

Special categories: special categories of data (Art. 9 GDPR), criminal-offence data, payment-card details, health data, children's data and sensitive personal information under US law are not intended inputs. Customer must not submit such data unless a lawful scope and safeguards have been expressly agreed. If such data is received by chance, it is restricted or removed on Customer's instructions. This exclusion does not reduce statutory protection.

Contacts: Provider: info@baoks.de. Customer: the accepting owner's email address, or a privacy contact Customer designates later.

Annex II: Technical and organizational measures

• Access control (premises and systems): the service runs in data centers of the hosting providers in Annex III, with their physical security. Administrative access to hosting, database and code repositories is restricted to named personnel with personal accounts and multi-factor authentication. The AI assistant's worker runs on company-managed hardware with full-disk encryption, access limited to the managing director, and credential files readable only by the operating account.
• Access control (data): server-side resolution of organization and membership for every request; tenant isolation through separate database schemas with validated schema names and tenant-scoped queries; role checks for privileged operations; least-privilege service credentials. The AI assistant reaches customer data only through tenant-checked tool functions of the application or the authenticated agent API with scoped keys; the external worker has no direct database access.
• Support access: staff open customer workspaces in the app only in time-limited, logged support sessions. Other access to Customer Personal Data is limited to the managing director and named staff with personal accounts and multi-factor authentication, for the purposes in section 2. Support sessions cannot accept contracts.
• Transfer control: TLS encryption for all connections between users, the application, subprocessors and the Amazon Ads API. Amazon refresh tokens are encrypted at rest with AES-256-GCM. The database is encrypted at rest by the hosting provider.
• Input control: change logs for advertising changes made by the service, audit logs for support sessions and privileged administrative actions, and approval records for changes the AI assistant proposes.
• Availability and recovery: daily database backups, retained for 7 days; a documented restore procedure that re-executes deletions instructed since the backup; rate limiting and monitoring to prevent abuse.
• Deletion: a documented deletion routine for Amazon Data across all tenant tables, active disconnection of Amazon access, automatic expiry of backups and of working copies on the AI assistant worker within 30 days.
• Machine learning: models that optimize a customer's campaigns are trained on that customer's data and, only where cross-customer learning applies, on the anonymized statistics described in Annex I. Provider records which bidding method managed which campaign and when, so that it can tell Customer on request, and records any opt-out from the comparison and from model training. After an opt-out, Customer's campaigns are managed only by rule-based methods and its data is not used for training; the comparison passes through the same safety limits for bid and budget changes (write guard) as all other changes. Cross-customer learning stays technically disabled until Amazon's written confirmation is on file. Statistics for it are built only from an allow-list of numeric fields, contain no free text and no identifiers of customers, products or campaigns, and are never taken from Amazon Marketing Cloud or Amazon Attribution data. Training and the method comparison run only for customers that have not opted out; Provider checks this technically before each training run and each comparison. Before statistics are used and before each model release, Provider tests and documents that (a) every aggregate combines data from at least 5 customers and no single customer contributes more than 50 % of it; (b) a re-identification test cannot link statistics to a customer, campaign or natural person with a success rate more than 5 percentage points above random guessing; and (c) a model-extraction and membership-inference test against the model reaches an attack AUC of no more than 0.55. Data or models that fail a test are excluded or not released. Training runs on Provider's own hardware in Germany with full-disk encryption, access by a single user account and no synchronization to cloud storage; training data is kept there only as long as needed for training and evaluation. Models derived from a customer's data are deleted or retrained within 30 days as set out in section 7.
• AI processing: language-model requests are sent exclusively through the Vercel AI Gateway to the providers listed in Annex III; the assistant has no direct connection to any model provider, under zero-data-retention terms: prompts and outputs are not stored after processing and are not used for training.
• Organization: confidentiality obligations for all persons with access; a documented incident response process, including notification of Amazon within 24 hours for incidents involving Amazon Data, reviewed at least every six months; review of these measures at least once a year.

Annex III: Subprocessors

• Vercel Inc., 440 N Barranca Ave #4133, Covina, CA 91723, USA. Hosting of the website and application, with application functions running in the EU (Dublin); server logs; performance measurement in the app; AI Gateway routing language-model requests for the AI assistant. All categories in Annex I passing through the application. Transfer mechanism: EU-U.S. Data Privacy Framework (Art. 45 GDPR) and EU Standard Contractual Clauses.
• Supabase, Inc., USA. Database, authentication and backups, with data stored in the EU (AWS region eu-west-1, Ireland). All categories in Annex I. Transfer mechanism for access from the USA: EU Standard Contractual Clauses (Art. 46(2)(c) GDPR).
• IONOS SE, Elgendorfer Str. 57, 56410 Montabaur, Germany. Sending account emails (confirmation, password reset, invitations). Names and email addresses of users. Processing in the EU.
• SpaceXAI LLC (formerly xAI), USA; Fireworks AI, USA; Baseten Labs Inc., USA. Each engaged by Vercel Inc. as a listed subprocessor for the AI Gateway. Language-model processing for the AI assistant under zero-data-retention terms; the provider depends on the language model selected for the workspace, and open-weight models (for example DeepSeek) run on Fireworks or Baseten infrastructure without any data reaching their developers. Messages to the assistant, its replies and the account data needed to answer. Transfer mechanism: onward transfer by Vercel Inc. under its Data Privacy Framework obligations and its written agreements with these providers.
• TypeSafe AI, Inc., 255 California Street, Suite 1300, San Francisco, CA 94117, USA, reached through the Vercel AI Gateway. Decision model Jev: assigns messages to the AI assistant to a model tier and classifies campaign names, keywords and search terms. Data: the message or advertising text, with email addresses and phone numbers removed. Zero data retention. Transfer mechanism: EU Standard Contractual Clauses (Module 3) under the TypeSafe AI data processing agreement.

Amazon Ads is Customer's authorized platform, the source of Amazon Data and a recipient Customer instructs. It is not Provider's subprocessor. Stripe (payments) and Google (optional sign-in) act as independent controllers for their own processing and do not receive Amazon Data. External tools that Customer connects through the agent API receive data on Customer's instruction and under Customer's responsibility.