After a customer call there is often no time left for documentation. The call ends on the way to the next appointment, and “I’ll enter that later” turns into “I forgot”. That’s why we built a workflow that handles documentation right after the conversation – by voice, without typing.
How it works
- Trigger: After the call, I say “Record CRM” in the car via Android Auto or Apple CarPlay.
- Dictation: I record a short note on the conversation: the request, commitments and next steps. Only my own summary is recorded, not the phone call.
- Secure transport: The audio file travels through a VPN tunnel to our company-owned AI instance. The data never leaves our infrastructure.
- Transcription and analysis: Whisper converts speech to text. A language model structures the content and suggests what should be created from it.
- Draft: The result is stored as a draft, not written directly to the CRM.
- Read-back and correction: The draft is read back to me, and I correct it by voice (“change the deadline to Friday”, “the contact person is Mr. Müller”).
- Approval: Nothing is saved until I give the okay.
- Transfer to EspoCRM: The approved draft creates the matching records.
What gets created in EspoCRM
A single voice note can produce several objects:
- Tasks: Call back, prepare a quote or confirm an appointment – each with a due date and an owner.
- Cases: If the customer reports an issue, a case with description and priority is created right away.
- Time tracking: Time spent on the call and preparation is assigned to the customer or case, which simplifies billing.
- Knowledge articles: If a conversation solves a recurring problem, a draft for the knowledge base is created, so the know-how stays with the team.
Why it works
- Human in the loop: The AI makes suggestions, I decide. Nothing ends up in the CRM unchecked.
- Data protection: Transcription and analysis run in our own environment, reachable only via VPN. For customer data in small and mid-sized businesses, that is a clear advantage over cloud services.
- Hands-free: The entire process is voice-controlled, so safety and traffic rules don’t suffer.
- Consistency: Every note gets the same structure, and no commitment slips through the cracks.
Two practical notes
Don’t record the phone call itself. Without the other party’s consent, this is not permitted in Germany. Your own summary after the call is unproblematic.
Also: the approval step belongs firmly in the process. Whisper and language models make mistakes, especially with names and technical terms. The draft step catches them.
Conclusion
AI doesn’t have to be complicated to help in everyday work. Here it doesn’t replace employees – it removes the most tedious part of daily business: documentation after the call.
How do you document customer calls – and what would you automate? We build workflows like this for our clients as well. Learn more about our automation projects or get in touch with us directly.