An AI email agent for the inbox where your requests arrive
It reads each email in your shared inbox, checks that it belongs there, works out what kind of request it is, compares the attachments with your list of required documents and leaves a draft reply in the inbox itself: asking for what is missing, or answering the question. Someone on your team reviews it and sends it.
What happens in a shared inbox
If your requests land in an inbox like requests@ or info@, this will sound familiar. Some of what arrives does not belong there at all: an invoice, a question for another department, an email someone sent to the wrong address. Some of it is a real request, but it comes with half the documents.
Someone opens everything, decides what each email is, opens the attachments one by one, checks them against the list of what that kind of job needs and writes, again, the email asking for what is missing. The questions that keep coming back get answered by hand every time.
And when a request finally reaches the technician, it is sometimes still incomplete, or was never for you in the first place.
A person still reviews and sends every reply. What changes is where they start from.
Five steps, in the order the email takes
Each step starts from what your team already knows: the kinds of request you receive, the documents each one needs and the answers to the usual questions. Whatever is not written down yet, we write with you, in plain language.
-
Step 01
Decides whether the email belongs in this inbox
It reads the email and its attachments and checks whether they belong in this inbox. If not, it flags the email and says which inbox it thinks it is for, but it does not forward it: at first, a person makes that call.
A persondecides where the rest goes.
-
Step 02
Works out what kind of request it is
From your list of job types, it works out what is being asked: a new request, a question, a reply to an earlier request for documents. If it fits none of them, it flags it for your team.
-
Step 03
Checks the attachments against your list
For that type of request, it compares what arrived with your list of required documents: what is there, what is missing and what will not do, and why.
- “The equipment data sheet is missing.”
- “The attached certificate has expired.”
- “The serial number cannot be read in the photo.”
-
Step 04
Leaves a draft in the inbox
It writes the reply and leaves it as a draft in the inbox, in the same thread: one message asking for everything that is missing, or an answer to the question in your own wording. It gives no prices, since those go in the quote, and on technical questions it points the way without giving a verdict.
A personreviews it, edits it if needed and sends it.
-
Step 05
Logs everything
Every email leaves a record: what came in, what the agent proposed, what was finally sent and who approved it. If you work to ISO 9001 or go through audits, it shows who approved each reply.
Nothing leaves the inbox until someone on your team sends it.
The draft, as your team finds it
An example of what appears in the inbox when a new request arrives with three attachments. The document list is yours; the agent shows what it checked, what is missing and what it leaves for your team.
-
Seen in: attachment 1
Request form
ReceivedComplete and signed.
-
Seen in: attachment 2
Photos of the equipment
ReceivedThe plate and the serial number are readable.
-
Seen in: attachment 3
Manufacturer's certificate
Will not doIt has expired. The draft asks for a current one.
-
Per your list for this type
Equipment data sheet
MissingNot received. The draft asks for it.
-
Seen in: body of the email
“How much would it cost?”
For your teamNo price in the reply: the draft says it will come with the quote.
“Hello, and thank you for your request. Before we can start, we still need the equipment data sheet and a current manufacturer's certificate: the one you sent has expired. As soon as we have them, we will send you the quote.”
The draft does not send itself. It sits in the thread, in front of whoever runs the inbox, to review, change if needed and send.
The agent prepares and flags; a person reviews and sends.
Who does what: the agent prepares, your team decides
The agent prepares
- 01Reads every email that reaches the inbox, attachments included.
- 02Sorts it: whether it belongs here and what kind of request it is.
- 03Checks the documents against your list for that kind of job.
- 04Leaves a draft in the thread: what is missing, or the answer to the question.
- 05Logs everything in a record you can show an auditor.
The agent never
- ×Sends an email without a person's approval.
- ×Forwards misdirected email on its own: it flags it, and at first a person decides.
- ×Quotes prices: those go in the quote.
- ×Gives a final verdict on a technical question: it points the way, and a technician decides.
- ×Changes your rules: your team does.
For teams that get the same requests by email, with documents
It fits wherever a shared inbox receives repeat requests, each with its own list of documents: testing labs, engineering firms, administrative agencies, certification bodies, property managers and, more broadly, any business that processes applications or case files.
When requests arrive by email
Because that is how your clients write to you, or because, for traceability, documents have to come in through one fixed, documented channel. The agent works inside that inbox and opens no new channel.
When your customers message you on WhatsApp
The same agent, with the same rules, can work on WhatsApp: it answers, asks for the documents and hands your team whatever needs a person. See the WhatsApp customer service AI agent.
In your inbox, not in another tool
You do not switch email providers. We connect to the mailbox you already have, and the drafts appear where your team already works.
- The request types, document lists and reply wording are yours, written in plain language, and never shared with other clients.
- We document everything at handover, so your team can maintain it without us.
- Google Workspace · GmailThe Gmail API can create drafts in the mailbox (Google's documentation).
- Microsoft 365 · OutlookMicrosoft Graph can create drafts, replies included, saved in the Drafts folder (Microsoft's documentation).
- We start withOne inbox, the one that receives new requests. The others come later.
- What staysDrafts in the thread and a log of every email: what came in, what was proposed, what was sent and who approved it.
One inbox, measured on your own email
We do not connect every inbox at once. We start with one and agree up front what counts as success: whatever reaches the technician is either a correct “not for us” or a request with complete documentation, ready for work to begin.
- Pick one inboxUsually the one that receives new requests.
- Write your rulesRequest types, the documents each one needs and what the agent must not do.
- Test on closed emailsWe agree in writing what it should have done with each one.
- Pilot with reviewA person reviews every draft before it is sent.
- Measure, then extendIf it works, we add the next inbox. If not, you know before it grows.
Why this order: AI proof of concept vs production.
The same approach already runs on WhatsApp, at Fixlab
We do not have a published email case yet; this page explains how we approach it. The same approach (the client's criteria written down, the right documentation for each case and a person wherever judgment is needed) has run on WhatsApp since June 2026 at Fixlab, a vehicle certification lab in Spain.
Its agent answers from the lab's own criteria, collects the documentation, opens the reform request in Fixlab's platform through its API and brings in a person only for the cases that genuinely need one. Before launch we agreed a written acceptance test with Fixlab: 50 real scenarios, with zero critical failures and at least an 80 % pass rate as the bar.
Read the Fixlab case- 279conversations in the first three months
- 51 %closed without a person; most handovers are price questions, which the agent passes to the team by design
- 20reform requests opened end to end in Fixlab's platform, with documents and data
- 0critical failures across the 50 acceptance-test scenarios
AI email agent: common questions
Does it send emails on its own?
Does it work with Gmail and Outlook?
Where is the data processed? What about GDPR?
What if an email does not belong in that inbox?
What if it does not know the answer?
Can we change the rules ourselves?
Is there a record for audits?
Does it also work on WhatsApp?
How much does it cost?
Sources, checked 8 Oct 2026: Google, “Create and send draft emails” (Gmail API) and Microsoft, “Create message” (Microsoft Graph). Fixlab figures: Fixlab case study. The emails, documents and wording on this page are examples.