Trust is the product.
Overwise sends emails under your name to your prospects. Every safety guardrail below is on by default — you can't even turn most of them off. Domain reputation, deliverability, and 'never send something we can't cite' aren't upsells; they're table stakes.
The 8 things we don't let you turn off.
The buyer reflex on outbound AI is "is this thing going to torch my domain?" The answer is no — because the safety guardrails below override your campaign settings, not the other way around.
Every AI-drafted message is checked against the lead's actual signals (LinkedIn, hiring page, recent funding). If a claim can't be cited, the message is discarded and the lead is held back, not sent. We never invent facts about your prospect.
For every send, you can see exactly which signals the agent used and which channel it picked. No black box; no 'the AI decided'. Trust comes from being able to inspect the decision.
First 7 days of any new mailbox: every send is queued for one-click approval. After 7 days of clean sends, you can switch to autopilot. Default-on safety beats default-on power.
Send-volume caps for the first 14 days regardless of campaign settings: 10 → 25 → 50 → 100/day. We protect your domain reputation more conservatively than you would; you can't even override it.
If your mailbox-wide bounce rate hits 8%, the campaign pauses automatically and you get a banner with the diagnosis (deliverability issue / list quality / DNS / OAuth). Re-resume manually after fixing.
Single sender enforces suppression — across all your campaigns and mailboxes. Negative replies, unsubs, and CAN-SPAM opt-outs feed back into one list, so a 'no thanks' from one campaign suppresses the lead in every other campaign you'll ever run.
Every message ends with a one-line disclosure that an AI assisted the drafting, in the user's local language where applicable. Compliant by default; you don't have to think about it.
Auto-injected on every send: physical mailing address (your business address), valid List-Unsubscribe header (one-click), valid List-Unsubscribe-Post body. We don't ask whether you want to comply.
Your data, your domain, your call.
Google or Microsoft OAuth only — we never store your password. Refresh tokens are AES-256 encrypted at rest with envelope encryption; multi-factor authentication is enforced for all administrative access and available for end-user accounts.
Lead data and outreach history live in MongoDB Atlas in EU-Central (Frankfurt). Brand-voice embeddings live in Qdrant Cloud (EU). Encryption at rest (AES-256), encryption in transit (TLS 1.3), private subnets — no public network exposure of stateful services.
You own your data. Export to CSV/JSON from any campaign. On account closure: 30 days read-only access for re-activation, then hard-delete from primary storage; encrypted backups purged within a further 60 days; mailbox refresh tokens deleted within 24 hours. Bookkeeping records retained 7 years (UGB §190).
Customer data is not used to train Anthropic, OpenAI or any Overwise model. We call Anthropic and OpenAI via their commercial APIs under executed DPAs; both providers' API terms exclude training on inputs/outputs. Brand-voice samples stay in your project's private vector index — never shared, never aggregated, never used for cross-tenant retrieval.
GDPR Art. 13/14 transparency by default (Privacy Policy + first-message source-of-data notice). CAN-SPAM compliant headers and unsubscribe. EU AI Act Art. 50 disclosure footer injected into every send. SOC 2 Type 1 targeted Q2 2026, Type 2 targeted Q4 2026. ISO 27001 roadmap on request. DPA available on every paid plan.
Every third party that touches your data.
The complete list, with the role each one plays under GDPR Art. 28 and the SCC module we use for transfers outside the EEA. This is the live mapping the Privacy Policy and the DPA point to. We give 30 days' written notice before adding or replacing any of them.
| Provider | Purpose | Location | Role / SCC module |
|---|---|---|---|
| Anthropic, PBC | LLM — drafting and reply classification | USA | Processor (Module 2) |
| OpenAI Ireland Ltd / OpenAI, L.L.C. | Embeddings and fallback LLM | IE / USA | Processor (Module 2) |
| MongoDB Ltd (IE) / MongoDB, Inc. | Primary database — EU-Central (Frankfurt) | IE / USA | Processor (Module 2) |
| Qdrant Solutions GmbH | Vector database (brand voice, contact similarity) | Germany | Processor (Module 2) |
| Stripe Payments Europe Ltd / Stripe, Inc. | Subscription billing — incl. fraud / AML | IE / USA | Independent Controller (Module 1) |
| Resend Inc. | Transactional email — primary | USA | Processor (Module 2) |
| ActiveCampaign LLC (Wildbit / Postmark) | Transactional email — failover | USA | Processor (Module 2) |
| Sideguide Technologies Inc. (d/b/a Firecrawl) | Website rendering for lead research | USA | Processor (Module 2) |
| Apify Technologies s.r.o. | Optional lead-discovery scraping | Czechia | Processor (Module 2) |
| Functional Software, Inc. (d/b/a Sentry) | Error tracking | USA | Processor (Module 2) |
| Plausible Insights OÜ | Cookieless analytics for overwise.com | Estonia | Processor (Art. 28) |
End-user-controlled integrations. These are not Overwise sub-processors — they are services you connect on your own behalf using your own credentials, governed by the providers' own terms and DPAs with you:
- Google (Gmail API + OAuth) — If you connect a Gmail mailbox
- Microsoft (Microsoft Graph + OAuth) — If you connect an Outlook mailbox
- Instantly — If you provide your own Instantly API key for reply webhooks
How we keep data in Frankfurt — and what happens when it has to leave.
Primary processing of customer data stays in the EU (Frankfurt). For transfers outside the EEA we layer the legal basis. The live mapping below is what the Privacy Policy §8 and the DPA §9 refer to; we re-verify each entity before publishing changes.
- EU-US Data Privacy Framework (DPF) — we verify each receiving entity at dataprivacyframework.gov/list before each sub-processor change.
- 2021 SCCs (Commission Decision 2021/914) — Module 2 (controller→processor) for our sub-processors acting as our processors; Module 3 (processor→processor) for chained transfers where Overwise acts as the customer's processor; Module 1 (controller→controller) where the receiving entity acts as an independent controller (Stripe for fraud / AML).
- UK Addendum (IDTA) for transfers involving UK personal data.
- Swiss SCC addendum for transfers involving Swiss personal data.
- Transfer Impact Assessment per receiving entity, following EDPB Recommendations 01/2020 — reviewed at least annually and on triggers (FISA §702 reauthorisation, Executive Order 14086, new SCC versions).
Trust questions, answered short.
The security and compliance questions every founder should ask before letting an AI send under their name.
Will Overwise email anyone without my approval?
Not on a fresh mailbox. The first 7 days are 'review-each-send' mode — every draft is queued for one-click approval. After a clean track record, you can switch to autopilot per-campaign. Default-on safety, opt-in autonomy.
Can the AI invent facts about my prospects?
No. The MessageVerifier is a cite-or-discard step — every claim in the draft must be supported by a signal we actually scraped (LinkedIn role, recent funding, hiring page). If it can't be cited, the message is discarded and the lead is held back. We don't send messages we can't back up.
What happens if my domain reputation is at risk?
Three guardrails fire automatically: (1) a 14-day warmup ramp caps daily send volume on new mailboxes regardless of your campaign settings; (2) bounce-rate auto-pause kicks in at 8% mailbox-wide; (3) SPF/DKIM/DMARC checks block sending from broken domains. You can't override these — they exist to protect you.
Does Overwise comply with GDPR?
Yes. EU data residency (Frankfurt), the Privacy Policy at /privacy covers Art. 13/14 in full, the DPA at /dpa is available on every paid plan, suppression is a single chokepoint across all of your campaigns and mailboxes, and a GDPR Art. 14 source-of-data notice plus the EU AI Act Art. 50 disclosure footer are injected into every message. We do not enrich beyond publicly observable business context — no private addresses, no Art. 9 indicators.
Is my data used to train AI models?
No. We call Anthropic and OpenAI via their commercial APIs under executed DPAs; both providers' API terms exclude training on inputs and outputs. Your sent-folder samples (used for brand-voice extraction) live only in your project's private vector index — never shared, never aggregated across customers.
How do I delete my data?
Cancel from the billing page. After cancellation, your data is read-only for 30 days (so you can re-activate without losing context), then hard-deleted from primary storage. Encrypted backups are purged within a further 60 days. Mailbox refresh tokens are revoked at the provider and removed from our database within 24 hours. Bookkeeping records are retained for seven years (UGB §190). Email [email protected] to expedite where no statutory retention applies.
Your first leads, in 5 minutes.
"Set it up in five minutes, no demo gate. First verified list before my coffee was cold — and it sounds exactly like me."