EN 
FR 
Customer issue
Case study: Automating incoming email management with AI - 80 % time saved on administrative tasks
Customer issue
This company offers outsourced administrative services for professional federations, franchise networks and national associations.
On a day-to-day basis, it manages for its customers :
-
Event registration
-
Member reminders
-
Updating contact details
-
Billing and receipts
Every day, she receives dozens of emails with highly variable content: some very clear, others ambiguous or incomplete.
The team spent several hours a day to read, sort, answer and record requests.
They wanted save time without sacrificing response quality, while maintaining a structured history of all requests.
Context
All customer exchanges arrive in a shared Gmail inbox.
Important information was then manually copied into Airtable for follow-up.
But as the volume increased, this system became :
-
Slow
-
Source of errors
-
And dependent on the availability of a dedicated person
Identified challenges
-
Emails very heterogeneous, sometimes vague ("as seen together...")
-
Manual extraction of key information (names, addresses, status, amounts, dates, etc.)
-
Risk of answering the wrong question for lack of context
-
Clear traceability in Airtable is essential
Work carried out by ActivDev
What we have put in place
1. Automatic reading of new emails
As soon as a new message is received in the dedicated box, the Make scenario is triggered.
The content is analyzed in natural language:
-
Type of request detected automatically
-
Estimated level of urgency
-
Extracted intentions (modification, cancellation, info request, etc.)
2. Extraction and recording in Airtable
Relevant elements are automatically extracted and sent to Airtable :
-
Customer name
-
Member login
-
Subject of request
-
Key data (e.g. "would like to receive a new invoice", "has changed address")
-
Initial status ("to be processed", "responded", "pending")
-
Link to source email
3. Contextual search with Pinecone
Some emails do not contain no obvious keywordsbut follow on from an existing exchange (e.g. "yes, that's perfect, go ahead").
In this case :
-
Recent email history is vectorized in Pinecone
-
With each new message, a semantic search finds the 2-3 closest exchanges
-
AI can thus understand what the message refers toeven without an explicit context
This avoids answering inaccurately or asking for unnecessary details.
4. Automatic response generation
Depending on the request, the AI generates a response:
-
Claire
-
Courtesy
-
Adapted to the company's usual tone
A draft is automatically created in Gmail for validation (or immediate dispatch, as the case may be).
Results
✅ 80 % of emails processed without manual intervention
✅ Fewer errors and less cutting and pasting
✅ Clear, structured tracking in Airtable
✅ Faster responses, with the right tone and context
✅ The team focuses on complex or sensitive requests
Results
Sorting and answering e-mails from 30 to 5 minutes