Read AI launches an email-based ‘digital twin’ to help you with schedules and answers

An email-based AI twin automates meeting scheduling, answers routine questions from internal knowledge sources, and relieves teams directly in the inbox. The value comes from not having to introduce new tools, keeping sensitive data protected, and making every action traceable. This is exactly the kind of solution I can and want to build for German companies.

What Ada shows and what German companies really need

TechCrunch reports on Read AI, which is introducing an email-based assistant called "Ada" as a "digital twin." You start by emailing ada@read.ai with the text "Get me started." After that, Ada can automatically propose free times in email threads, send new slots in response to counteroffers, handle out-of-office messages, and answer questions based on the company knowledge base, previous meetings, and even public web searches. According to Read AI, the contents of other meetings are never disclosed, only availability. Justin Farris, VP Product, emphasizes that Ada does not rely on MCPs but instead builds a knowledge graph from meeting data and connected services. CEO David Shim describes Ada like a new employee: the more context you connect, the more tasks it takes on. Ada is also supposed to act proactively, for example by initiating follow-ups after meetings.

That is the right approach: email first, context from real data, step-by-step automation. At the same time, I see different conditions in German companies:

• Data protection and GDPR: No standard web search of external sources, no unnecessary data egress. Data residency in the EU, ideally in Germany. Auditability for IT, data protection, and the works council.
• Microsoft reality: Exchange Online or on-prem, Microsoft 365, Teams, SharePoint, often also Confluence, Jira, SAP, DATEV. An assistant must speak this world seamlessly.
• Works council and approvals: Draft first, then human approval. Transparent logs. Switchable off per team and mailbox.
• Clear boundaries: Calendar queries only return free/busy, no meeting contents without explicit approval. No sending of sensitive information without consent.
• German as the primary language, consistent tone of voice, binding signatures, GoBD-compliant archiving.

In short: I like Ada’s direction. I would harden, integrate, and secure it for the German context.

My German email twin: Architecture and privacy by design

I build such solutions with a clear blueprint. Goal: "assistant in the inbox," minimally invasive, traceable, secure.

1. Activation in the mailbox instead of a new tool

• Trigger via CC: Users put, for example, twin@company.com in CC. The assistant detects intent, role, and language and responds directly in the thread.
• Alternative rules: Certain mailboxes (info@, jobs@) are managed permanently.
• Onboarding as simple as with Ada: One initial email with "Let’s get started" is enough, the assistant explains the next steps and obtains explicit approvals.

2. Scheduling without content leaks

• Calendar access only as free/busy via Microsoft Graph or Exchange Web Services. No disclosure of subject or attendees.
• Preferences per user: core working hours, travel times, focus blocks, buffers before and after meetings, preferred meeting lengths.
• Flow like Ada: It proposes times, responds to counteroffers with new options, and after confirmation sends an ICS invitation. Title and description remain neutral, for example "Sales coordination."
• Rules for external contacts: No access to internal rooms or resources without approval, clear fallbacks in case of conflicts.

3. Answers from the knowledge base, but controlled

• Retrieval-augmented generation: I index SharePoint, Confluence, OneDrive, Jira, optionally SAP knowledge. The raw data remains in the source system. Only vector references and the minimally required text snippets are processed.
• Strict source citation: Every answer cites the source so a human can verify it.
• No open web search by default. External sources are opt-in and are marked in the answer.
• Language and tone profiles per team. German by default, bilingual on request.
• Models: Azure OpenAI in EU regions or Aleph Alpha. On-prem models on request for particularly sensitive areas.

4. Governance, approvals, transparency

• Shadow mode: The assistant drafts only; nothing is sent without "OK." Approval directly in the email via "OK" or a clickable link.
• Roles: Who is allowed to do what. Example: An HR assistant may send OOO emails, never contract contents.
• Logging: Complete audit logs for IT, data protection, and the works council. Insight into who proposed what, when, approved, or rejected.
• Redaction: Automatic detection and redaction of sensitive content according to policy (e.g., health data) before a draft is created.
• Retention: Journaling and retention per GoBD, integration into existing archiving systems.

5. Proactive, but safe actions

• Like Ada: Follow-ups after meetings. I extract to-dos from Teams transcripts or notes and suggest concrete emails, including context and attachments.
• Synchronization with tools: Created follow-ups can automatically create tickets in Jira or tasks in Planner, cleanly orchestrated via n8n.
• Reminders of deadlines, resubmissions, escalations with clear rules.

6. Tech that fits German IT

• Orchestration with n8n for transparent workflows and approval loops.
• Services in TypeScript and Python. Vector search in PostgreSQL with pgvector or Elasticsearch.
• SSO via Azure AD, secrets in Vault, network isolation in a German cloud or on-prem. S/MIME and DKIM for signatures.
• Interfaces: Microsoft Graph, EWS, IMAP/SMTP, Confluence/Jira APIs, SharePoint Graph, SAP OData, DATEV partner APIs where available.

The result is an email twin that works like a trained colleague, but within boundaries that fit data protection, the works council, and compliance. I deliberately implement the "train like a new hire," as David Shim describes it, in a controlled way: drafts first, then with graduated permissions.

Concrete use cases from my practice and how I implement them

1. Sales: Scheduling without ping-pong

• Starting point: A field sales team schedules appointments with customers and partners every day. Previously 6 to 10 emails per appointment.
• Solution: Sales CCs twin@company.com and writes "Please set up a 45-minute meeting next week, mornings preferred." The assistant proposes three slots, honoring core hours and travel times. If the customer replies "Wednesday 3 pm doesn’t work, is Thursday morning possible?", the assistant provides two new options. After confirmation it sends a neutral ICS invitation, without exposing internal subject details, just as Ada promises.
• Result: Fewer emails, faster confirmations, no leakage of sensitive calendar details.

2. Shared inbox customer service: Faster first responses with source citation

• Starting point: A mid-sized manufacturer receives many questions about lead times, warranties, and configurations via service@.
• Solution: The email twin reads the request, looks up suitable answers in Confluence articles and SharePoint manuals, quotes relevant paragraphs, and drafts a reply in the correct tone. The agent approves with one click. If the question cannot be answered confidently, the assistant only creates a structured follow-up question or opens a Jira issue.
• Result: Noticeably faster response times and consistent quality. From my projects, 30 to 50 percent automatically prepared first responses are realistic when documentation is well maintained.

3. HR and OOO: Reliable out-of-office communication

• Starting point: Frequent inquiries during vacation periods, some in English, some in German.
• Solution: The assistant maintains out-of-office emails with context-dependent hints: "You can find our time-tracking FAQ here," "For urgent contract inquiries, please contact …". For subjects detected as "urgent" it prepares a forwarding draft to the deputy with key information, never automatically without approval.
• Result: Fewer interruptions during vacations, structured handovers, satisfied requesters.

4. Proactive follow-ups after meetings

• Starting point: Action items from Teams meetings get lost.
• Solution: The assistant searches approved transcripts, extracts to-dos, and drafts follow-ups with context. Example: "As discussed, I’m sending you the updated price list and propose next Tuesday at 10 am for the technical deep dive." Read AI has announced precisely this proactive logic for Ada. I implement it with clear approval.
• Result: More completed to-dos, less chasing.

This is how I roll out such projects:

• Week 1: Data protection impact assessment, scope, involve the works council.
• Week 2: Connectors to M365, Confluence, SharePoint. Initial indexes.
• Week 3: Calendar logic with free/busy, preferences.
• Week 4: Draft and approval workflows in n8n.
• Week 5: Pilot in shadow mode with 10 to 20 users, fine-tuning.
• Week 6: Training, enforce policies, rollout.

Professionally, I follow Read AI’s core ideas: email-first, answers from knowledge sources, proactive follow-ups, knowledge graph instead of loose integrations. Technically, I ensure everything runs GDPR-proof, auditable, and Microsoft-compatible. And by default I opt for "assistance rather than autopilot."

Conclusion: Email remains the main channel, and the AI twin finally makes it easy

Ada shows how much can be accomplished in email alone: fixing appointments, delivering answers, initiating follow-ups. For Germany, that also requires a privacy-hardened foundation, clear approvals, seamless Microsoft integration, and transparent logs. That is exactly what I am building: an email twin that provides tangible relief and adheres to your rules. If you want a realistic pilot in 4 to 6 weeks, get in touch.

Frequently asked questions

Does it run on-prem or in the cloud?

Both are possible. Many customers start in a German cloud VPC with Azure AD SSO. For particularly sensitive areas I plan on-prem deployments. Data stays in the EU, logs are fully exportable. I prefer to use models in EU regions, alternatively on-prem.

How do we ensure GDPR and works council visibility?

With privacy by design: free/busy instead of content access, shadow mode as the default, explicit approvals via click or "OK" reply, comprehensive audit logs, redaction of sensitive content, clear data flows. I involve data protection and the works council from week 1 and document policies and purposes.

Which systems can the assistant connect to?

Outlook and Exchange, Microsoft Teams, SharePoint, Confluence, Jira, OneDrive, IMAP mailboxes, optionally SAP OData and DATEV interfaces. Answers are based on your sources; external web search is only active on request and is indicated in every answer.