AI Disclosure

Druma uses AI to assist with two kinds of work: (1) extracting structured data from documents and emails, and (2) translating short messages between drivers and planners. We do not use AI to make legal decisions, regulatory filings, or any decision that produces a significant effect on a person without human review.

• Smart Import — bulk onboarding of trucks, drivers, clients, and insurance from existing documents (PDF / images / spreadsheets). Output is shown in a review table; you confirm before anything is saved.
• Inbound order email — when a shipper emails a transport request to your Druma alias, AI extracts the lane, dates, and goods into a draft order. The order stays in draft until a planner accepts it.
• Inbound carrier and supplier invoices — when invoices arrive at the dedicated email alias, AI extracts amount, VAT, dates, and reference numbers. The result is queued for human approval before it touches the books.
• Driver-planner message translation — when a driver and planner have different default languages, messages are auto-translated so each side reads in their own language. The original is always shown alongside the translation.
• Ask Druma — an in-app assistant with two modes. By default it answers questions about how to use Druma from our product documentation. When an administrator enables the optional data assistant, it can also answer questions about your own operational records — orders, fleet status, finance, driver hours, and order notes — by querying your database on your behalf. It is strictly read-only and never modifies data. Every query runs under the asking user's own permissions, so the assistant can only surface data that user is already allowed to see in the app (a driver sees only their own trips; a user without financial access cannot retrieve prices or margins, and so on). Order notes and other free-text fields may contain personal data such as names, which is sent to the AI provider to compose an answer.

Druma's own AI features run on Google Vertex AI (now named Gemini Enterprise Agent Platform). The Ask Druma documentation chatbot can alternatively run on Anthropic's Claude, as described below:

• Google Gemini (via Vertex AI) — document extraction, Smart Import, CMR validation, tender import, the Ask Druma assistant (data mode), the client-portal assistant, driver voice-note transcription and KPI narratives. Requests are processed on Google Vertex AI's EU multi-region endpoint (aiplatform.eu.rep.googleapis.com); according to Google's terms, data is stored and ML processing runs in EU member states when the EU endpoint is used (Service Specific Terms, AI/ML Data Location), and Druma is configured to use the EU endpoint. This is Google's published commitment, not an audit by Druma, and it is a commitment about where data is stored and processed, not a promise that no data ever leaves the EU. Google affiliates, including Google LLC in the US, may in limited cases access data to maintain the service or resolve technical issues (Google's subprocessor list). Google Cloud EMEA Limited is in the EEA; for onward transfers Google relies on the EU-US Data Privacy Framework (Google LLC is certified) for the US, and on Standard Contractual Clauses between Google and its own subprocessors for other third countries (Google Cloud Data Processing Addendum, Appendix 3, section 4.1). The one exception is the optional, user-triggered web research on a company or site (Google Search grounding), which is currently switched off for all customers and, if ever enabled, is off by default for each company: Google's terms exclude grounded requests from the EU data-location commitment, so such a request may be processed outside the EU, and Google retains grounding queries for up to 3 days for debugging, which cannot be disabled. It sends public business identifiers (company or site name, country, city, VAT number and its validity flag, the legal name from the VIES check, the website address, and the public text of a site's own website) — no order, document or driver data, and no contact records from your Druma account; the public text of a company's own website and a sole trader's business name may incidentally contain personal data.
• Google Cloud Translation API — driver-planner message translation. Runs on Google Cloud infrastructure; Google states the text is held briefly in memory to perform the translation and is not used to train its translation features; this API is not sent to the Vertex AI EU multi-region endpoint, so translated content may be processed outside the EU.

Anthropic Claude (Haiku) — the Ask Druma documentation chatbot (the Help tab, which answers questions about how Druma works) runs on Claude by default and on Gemini when Claude is unavailable. Druma chooses the model, and no other Druma AI feature (the company-data assistant, document extraction, voice notes, narratives, translation) uses Claude. While the chatbot runs on Claude, the question, the earlier turns of the conversation and the retrieved public documentation are sent to Anthropic's API (no operational records are), which is operated in the United States and is not configured to use an EU endpoint; the transfer safeguard is described in our sub-processor list. Under Anthropic's commercial terms, API inputs and outputs are not used to train its models. If Claude is unavailable, the request is retried once on Gemini.

We do not use OpenAI or Meta for any core Druma AI feature, and no Anthropic model other than the documentation chatbot above. The one exception is the optional, opt-in Druma AIdescribed below, where the company itself — not Druma — chooses and pays for the provider.

Druma AI is a separate, optional feature that lets a company administrator connect the company's own API key for OpenAI, Anthropic (Claude), or Google Gemini, and chat with an AI agent that can read (and, with explicit confirmation, propose changes to) the company's operational data.

• Off by default. A company must explicitly enable it and connect a provider before any data is sent.
• Own provider, own terms. Once enabled, the prompts and operational data (orders, fleet, finance, driver hours, and similar records within that user's normal permissions) that Druma AI queries on the user's behalf are sent directly to the company's chosen provider — api.openai.com, api.anthropic.com, or Google's public Gemini API — and processed under that provider's own terms and data-processing agreement with the company, not under Druma's agreements with Google Vertex AI. If the key is a free-tier Google Gemini key, Google's own terms allow Google to use what is sent to improve its products (and human reviewers may read it); Google permits only paid keys for users in the EEA, Switzerland and the UK. Use a paid key. The company's own Google terms apply to that use.
• May leave the EU. Unlike Druma's own features, which call Google Vertex AI's EU endpoint, the public OpenAI and Anthropic APIs, and the public Gemini API, are not configured to use an EU endpoint. Data sent to Druma AI may be processed outside the EU/EEA.
• API key custody. The key itself is encrypted at rest in a managed vault and used only server-side to call the chosen provider on the company's behalf; Druma never displays it again after it is saved.
• Requires explicit consent. Before a company can save Druma AI credentials, an administrator must acknowledge this data-flow and record that acknowledgement.

If your company has not enabled Druma AI, none of your data is ever sent to OpenAI, Anthropic, or the public Gemini API.

Ask Druma has two tabs, and they keep conversations differently — on purpose.

• Help mode is not saved. Questions about how to use Druma are answered from our product documentation. The conversation lives only in your browser tab and is discarded when you close the panel or reload the page. Nothing is written to our database, and there is nothing to retrieve later — including by us.
• The one exception is escalation. If you click "Talk to a human", the help conversation up to that point is attached to the support conversation so the person helping you can see the context. That support conversation, transcript included, is deleted 90 days after its last activity.
• Assistant mode is saved, for you. The optional data assistant and Druma AI answer questions about your own operational records, and those conversations are saved so you can reopen them later, ask a follow-up, and see what you already asked. This is the tab your company pays for, and the history is part of what it pays for.

Who can read a saved conversation. Only the user who created it. Not colleagues, and not a company administrator. We took that position deliberately: what a planner asks the assistant reveals what they were worried about, and making that readable by their employer would be workplace monitoring rather than data ownership. An operator that needs oversight of assistant use has it in the audit log, which records which user ran which query and when — what was queried, not what was asked.

How long, and how to remove it. A saved conversation is deleted automatically 90 days after its last message. You can delete any conversation yourself at any time from the assistant panel. Saved conversations are included in a personal-data export, and are removed when your user account or the company account is deleted.

A saved conversation contains what you asked and what the assistant answered — which, in Assistant mode, restates records you were already entitled to see when you asked. It does not grant any new access: the assistant queries your database under your own permissions every time, and a saved answer is a record of that moment, not a live view.

Under Google Cloud's terms (Service Specific Terms, training restriction), Google does not use customer data to train or fine-tune AI/ML models without our permission, and does not store prompts longer than reasonably necessary to produce the output, subject to the abuse-monitoring and grounding exceptions below. Requests are processed on Google Vertex AI's EU multi-region endpoint (see Models and providers for the web-research exception).

What we can and cannot evidence about Google's retention. According to Google's abuse-monitoring documentation, if its automated safety classifiers detect suspicious activity, Google may log the flagged prompts, for up to 90 days and in the same EU region, solely to investigate abuse. Google also keeps an in-memory cache of inputs for up to 24 hours, in the same region, which can be disabled at project level. Google's documentation says accounts under a Google Cloud Master Agreement are exempt from abuse-monitoring prompt logging by default; we have not confirmed which Google terms govern Druma's account, so we assume the logging applies. Google Search grounding (currently switched off for all customers; optional and off by default per company if ever enabled) has its own retention: Google stores the derived search queries for up to 3 days and this cannot be switched off. We will update this section once the position is confirmed with Google.

Druma itself does not train AI models on your data. We do not sell or share your data with model providers for any purpose other than serving the requested inference.

The CMR reader does learn one thing from your documents. On your paper CMRs it measures where the printed captions and boxes sit on each form type, so that it finds them on the next form of that type, for every customer. That geometry is the only part shared between customers: positions relative to the printed frame, identifiers of printed captions from a closed list, and counts, medians and spreads over several documents. It holds no filled-in text, no names, plates or other values, and no company identifier. No model is trained on it. Your documents, the text read from them and your corrections stay in your account, for at most 12 months, or 30 days if nobody labelled or corrected the result. The terms are in section 4.1 of the Data Processing Agreement.

Driver performance scoring is not AI. Druma's driver scorecard is a fixed, published weighted average of five operational components — on-time performance (35%), fuel efficiency against a benchmark (20%), idling (15%), incidents (15%) and compliance (15%). It is ordinary arithmetic on records Druma already holds. No model is trained, no model is applied, and no data is sent to any AI provider to produce it. Components with no data are excluded and the remaining weights renormalise, rather than the driver being scored zero. Small samples are flagged as low-confidence. The score is nevertheless profiling in the GDPR sense and is described as such in our Privacy Notice.

Every AI output is treated as a draft. A user must review and confirm before:

• An imported record is saved (Smart Import).
• A draft order is converted to an active order.
• An extracted invoice is matched and posted.
• A translated message is sent (the original is sent regardless; translation is shown alongside).

For Ask Druma, the assistant can read but never write. If you ask it to take an action, it will give you the steps to take in the planner, not perform them itself.

• Message auto-translation — toggle in your Profile settings. Off by default for any user whose default language matches the counterparty.
• Smart Import — entirely optional; never required to use the rest of the platform.
• Ask Druma — only triggered when you open the assistant panel; nothing is sent until you ask a question. The optional data assistant mode, which queries your operational records, is off by default and must be switched on by a company administrator before any account data is sent to the AI. Help-mode chat is never saved; Assistant-mode conversations are saved for you and can be deleted at any time — see Conversation history.
• Druma AI (bring your own AI) — off by default. A company administrator must enable it, connect a provider API key, and acknowledge that data sent through Druma AI goes to that provider under the company's own terms and may leave the EU, before any data is sent.
• Inbound email AI — disabled by default. Enable per-feature in Settings → Integrations once you have configured your incoming alias.

The EU AI Act (Regulation (EU) 2024/1689) entered into force on 1 August 2024 and applies in stages:

• 2 February 2025 — the prohibited-practices rules (Chapter II) and the AI-literacy obligation (Art. 4) began to apply.
• 2 August 2025 — obligations for providers of general-purpose AI models, together with the governance, notified-body and penalty provisions.
• 2 August 2026 — the bulk of the Regulation, including the Article 50 transparency obligations and the Annex III high-risk regime.
• 2 August 2027 — the remaining provisions, including high-risk AI embedded in products regulated under the Union harmonisation legislation in Annex I, and the transition rules for GPAI models placed on the market before 2 August 2025.

Our classification. Druma is a deployer of third-party AI models, and for the Ask Druma and Druma AI chat surfaces also the provider of the AI system the user interacts with. Our AI uses fall within the limited-risk / transparency tier — assistive tools with mandatory human review. We do not operate a prohibited practice under Chapter II, and we do not operate a high-risk AI system under Annex III: we do not score creditworthiness, determine access to essential services, or use AI to recruit, allocate work, evaluate performance, or make any decision about a worker's employment relationship.

Note in particular that Druma's driver performance score is not an AI system. It is a fixed, published arithmetic formula computed from operational records — no model is trained, inferred from, or applied. It is nevertheless profiling under Article 4(4) GDPR, and is described as such in our Privacy Notice.

Article 50(1) — you are interacting with an AI system. Where an AI system is intended to interact directly with natural persons, those persons must be informed that they are interacting with an AI unless it is obvious from the circumstances. This obligation applies to Ask Druma and Druma AI, and we state it here rather than deferring it: when you use Ask Druma or Druma AI, you are talking to an AI system, not to a member of Druma's staff. The one exception is the live-support hand-off inside the same panel, which is labelled Live Support and connects you to a human. The panels themselves carry an AI indicator (an assistant icon and a provider badge such as "Powered by Gemini", "Druma AI · read-only", or "Using your provider key").

Article 50(2) and (4) — labelling of AI output. Druma does not generate synthetic image, audio, or video content. AI-extracted fields are labelled as AI output in the review screens before a human confirms them, and AI-translated messages are shown alongside the untranslated original.