Solutions / Front desk
An inbox sorted and replies drafted by the time you sit down.
An agent that reads every incoming email, sorts it, and drafts the reply in your voice using your prices, policies and past answers.
What outside research says
14% more customer issues were resolved per hour when 5,179 support agents were given an AI assistant. Newer agents gained the most.
Source: NBER, 2023 (opens in a new tab)8.8 hours a week on email is what the heaviest quarter of email users spent, in Microsoft’s 2023 analysis of how people use its office apps.
Source: Microsoft, 2023 (opens in a new tab)
Findings from the research linked under each figure. They are not Yielde results.
The loop
How it works
From a full inbox to replies ready to send
Seven steps, from the first event to the result. Choose a step to read what happens in plain words, how it is built, and what keeps it right.
Step 1 of 7
An email lands
In plain words
A customer asks for a quote, a supplier sends a statement, someone forwards a complaint. It all lands in the same inbox while you are on site.
Safeguard It has no permission to delete email. The most it can do to a message is label it and file it.
Under the hood
The agent is built to watch your Gmail or Outlook mailboxes, shared ones included, through the mail provider’s own connection, with access you grant and can withdraw. It can read and label mail but has no permission to delete it. Each email is handled once: its message id, a unique reference every email carries, is recorded before work starts, so a retry after a dropped connection does not draft the same reply twice.
Works with
- Gmail
- Outlook
- Shared mailboxes
It sorts the email
In plain words
Before you open the inbox, every email is labelled: new enquiry, existing customer, supplier, account, complaint or junk. The ones that need you today sit at the top.
Safeguard Text inside an email is content, never a command. An email asking for something to be sent, paid or changed goes to a person.
Under the hood
The language model returns a fixed set of fields for each email: category, urgency, whether it is sensitive, and how sure it is. That output is checked against a schema before any label is applied. Complaints, money, legal letters, personal details and low confidence all mark an email sensitive. An email is text from a stranger, so anything in it that reads like an instruction to the agent is treated as content to sort, never as an order.
Works with
- Gmail
- Outlook
Sensitive or unsure: goes straight to step 05, You send, edit or bin it.
Junk or noise: goes straight to step 07, You get the day’s summary.
It reads the thread
In plain words
Like you would, it reads the earlier emails in the conversation and looks up your price list, your policies and how you answered the same question before.
Safeguard It reads only the mailboxes and documents you chose, and nothing at this step can change them.
Under the hood
The full thread is rebuilt from the reference headers that link each reply to the email before it, so a reply-all or a forwarded chain is read in order. A search over your own documents and your sent replies then finds the passages that answer this email, and each passage keeps a note of where it came from. On a shared mailbox, a thread someone on your team is already answering is recognised as theirs and left alone.
Works with
- Gmail
- Outlook
- Shared mailboxes
It drafts the reply
In plain words
A reply waits in your drafts, written the way you write and inside the right conversation. Beside it you see which price list or past email each fact came from.
Safeguard A fact it cannot point to a source for is left as a visible gap. The model never fills it in.
Under the hood
The draft is written from the passages found and a style brief taken from your own sent mail. Every price, date or policy in it is tied to a passage, and a fact with no source becomes a marked gap for you to fill. The draft is saved as a reply in the same thread, with the right headers, so it stays in the conversation. A sensitive email gets a short summary for you instead, unless you ask for drafts on that kind too.
Works with
- Gmail
- Outlook
You send, edit or bin it
In plain words
You read the draft, change a word if you need to and press send. For the dull ones, such as directions or opening hours, you can let it send by itself.
Safeguard Complaints, money, legal and personal matters always wait for a person, whatever is switched on. A request to change bank details is always flagged.
Under the hood
Sending is a separate permission, off by default. You can switch it on per category, for example directions, or a note that an enquiry has arrived. Even then a reply only sends on its own when it is in that category, every fact has a source and the model is sure. Otherwise it waits for you. A daily limit you set caps automatic sends, and each one is listed in your summary.
It learns from your edits
In plain words
When you fix a draft, it takes note. If you always soften the opening or add your cell number, the next drafts do it too.
Safeguard What it learns changes how it writes, never what it may send on its own. Only you can change that.
Under the hood
Each sent reply is compared with the draft it came from, and the difference is stored as an example. Edits that keep coming back are proposed as changes to the style brief and reviewed with you. The model itself is not retrained: the brief and the examples change, and each change is replayed against past emails before it is used.
You get the day’s summary
In plain words
Each day you get one short email: what needs your answer, what is drafted and ready, what went out by itself, and what was filed as junk.
Safeguard Every automatic send is listed in the summary with the rule that allowed it.
Under the hood
The summary is built from the log, not from memory: every email received, how it was sorted, what was drafted, what you sent or changed, and what went out on its own. Each line links to the email itself. A sensitive email that has waited longer than the time you set moves to the top, so it cannot sit unseen under a pile of newsletters.
Works with
- Gmail
- Outlook
What it knows
Your price list, policies, terms and the replies you have sent before, searched for each email with the source kept. A style brief taken from your own sent mail, and the categories and escalation rules agreed with you.
What it may do on its own
Read and label email, rebuild threads, write drafts into the right conversation and send the daily summary. Once you switch a category on, it may send routine replies in that category within your rules and daily limit.
What waits for your yes
By default, every reply. Always: complaints, anything about money, refunds or quotes, legal letters, personal details, a supplier asking to change bank details, and anything it is not sure about. It never deletes email.
How we test it
Past emails you choose are replayed through sorting and drafting and compared with what you actually did, awkward ones included: a phishing email, a long reply-all, a forwarded chain, an angry customer. It then runs draft-only, and you switch a category to send alone only once its drafts stop needing edits.
What we measure
Once it is live, these are tracked and reviewed with you.
- Emails sorted before you opened the inbox
- Drafts sent without changes
- Sensitive emails passed straight to a person
What you get
- Sorting and labels across shared inboxes
- Draft replies in your tone
- A daily summary of what needs you
- Escalation rules for anything sensitive
Works with
- Gmail
- Outlook
- Shared mailboxes
Who it is for
Owners and small teams who answer email after hours.
Built to order. We scope it with you on a call, price it before any work starts, and run it in draft mode until you trust it.
Related
- AI receptionist
Never miss a lead again.
Never Miss a Lead is one agent built to answer your calls, WhatsApp and email while you are on the job, take the details and line up the booking for your yes.
Waitlist open - Booking agent
Bookings made and kept, without the phone tag.
An agent that offers customers real open slots, books the one they pick, sends the reminders and fills cancellations from a waiting list.
Built to order - Support and knowledge chatbot
Customers get answers from your own material, without waiting for your team.
A chatbot that answers from your own manuals, policies and product pages, shows where each answer came from, and hands over to a person when it should.
Built to order