Email support for small and mid-sized businesses
Every customer email arrives with its reply already written.
SupportPilot reads along in your support mailbox, looks up the booking or order in your system, and lays the finished reply next to each mail. Your team checks and sends – the searching, looking up and typing has already happened. Only the last word stays where it belongs: nothing leaves the house without a human's approval.
Add a card · 7 days of drafts only, €0 · then from €99/month, cancel during the week at no cost · no DNS changes
Why this is not a chatbot ↓
“Hello, we will not get in before 10pm tonight. Is check-in still possible?”
reads · checks
“Dear Ms Meier, a check-in at 10pm is no problem at all. We look forward to seeing you.”
Your next Monday
The mailbox is full. The dread is gone.
Instead of a full inbox
A list of cases with finished drafts. Each one shows what it is built on: which house rule, which booking, how confident the system is. Your colleague reads top to bottom and approves.
Instead of “what was the rule again?”
The house rule is in the draft, with its origin. Even the new hire in month two answers the way you have answered for years – without interrupting anyone to ask.
Instead of “I'll do that tonight”
The mailbox is empty before you leave, because writing replies is no longer the bottleneck. What is genuinely difficult stays with you – and is marked as such.
From 800 mails a month: two to three working days instead of ten to seventeen
The same example as on the pricing page: 800 support mails a month, 6–10 minutes per reply today, 1–2 minutes with a finished draft · Blue: today · Amber: with SupportPilot
Example calculation · your real numbers show on the usage page after the trial week
You have probably tried this before
And it failed for good reasons
Most people looking for a fix for the support mailbox already have an attempt behind them. The four most common disappointments – and what SupportPilot does differently so they do not repeat:
The chatbot made something up and you had to apologise
That is why SupportPilot claims nothing it cannot back with evidence. Every draft carries its sources: house rule, booking record, order number. If the system cannot find the case, the draft politely asks – instead of guessing. And a second, independent review step scores every text before a human ever sees it.
The tool wanted clean data you simply do not have
SupportPilot starts with what exists: your website gets analysed, your house rules are uploaded as a plain text file – the way they sit in your wiki or in your head today. Connecting the booking or ERP system is a later, optional step, not a precondition for starting.
The rollout became an IT project that never finished
Here a program signs in to your existing mailbox – like Outlook on a new machine. No MX record, no SPF, no DNS change, no migration, no new software your team has to learn.
You corrected the same mistake fifty times
One correction is enough: remember: dogs are allowed on request in a ticket comment becomes a durable rule that stays findable in the knowledge base. What your team teaches once, it does not have to teach again.
Shadow mode
First it reads along. Then it writes along. It sends only once it has earned that.
Other systems go live and write to your customers from day one. SupportPilot starts in shadow mode: it sees every request, writes the reply it would send – and sends nothing. For seven days your staff compare it with what they would have written themselves.
So you are not buying a promise – you are watching a candidate on a trial day: on your own mail, before any money moves and before a single customer notices anything.
Autonomy is earned, not configured
Even after the trial week, SupportPilot sends nothing on its own. When you are ready, you enable autonomy one category at a time – and the system checks for itself whether that category has a track record through human hands.
| ✓ | Daily limit and an undo window before every send |
| ✓ | Automatic shutdown when corrections start piling up |
| ✕ | Money, complaints, legal: permanently with a human |
Out of the box: sending off
Seven days of reading along – and zero emails sent
Blue: drafts per day (illustrative) · Amber: auto-sent emails – technically locked during the trial
Shadow mode · draft counts illustrative · sending: product default "off"
The flow
Three steps your team feels every day
Mail becomes a case
Incoming email lands as a ticket in the right thread – including replies to old mails, including mails without a ticket number in the subject.
Draft from real data
The system pulls the booking, order or contract from your systems and writes a reply in your tone. What it cannot back with evidence, it does not claim.
Your staff decide
Read, adjust if needed, send with one click. Automatic sending only where you explicitly enable it per category – never for money or legal matters.
Your alternatives
You have four other options. All four cost you more.
Not because the other paths are bad – but because, for a full support mailbox in a small company, they carry the wrong price.
| Option | What it really costs you | With SupportPilot |
|---|---|---|
| Hire someone | A full salary, recruiting, training – and when that person leaves, the knowledge leaves with them. | A fraction of that per month. The knowledge stays in a knowledge base you own and can export as a file. |
| Website chatbot | Knows only what is on the website, guesses about bookings, and talks in public. A mistake is visible to everyone. | Works inside the mailbox you already use, on real data from your systems – and never speaks to a customer unreviewed. |
| A new helpdesk | Migration, training, a new interface for everyone. At the end the mail is sorted, but still not answered. | Sits behind your existing mailbox. Your staff keep working as they do, except the replies arrive already written. |
| ChatGPT by hand | Copy every mail, explain the context, paste the answer back. Without booking data, without a consistent tone, without a record. | Matching, lookup and tone happen automatically. Every decision is in the audit log. |
| Carry on as before | Evenings, weekends, slower replies – and the mistakes people make at 10pm. | Visible on your own mail within a week, at no risk: nothing is being sent. |
Easy start
Connected like a mail client, not like an IT project
SupportPilot signs in to your existing support mailbox – like Outlook on a new machine. No MX record, no SPF, no DNS changes, no migration.
What you need
| 1 | Credentials for your support mailbox (app password) |
| 2 | Your house rules as a text file – or we read them from your website |
| 3 | Read access to booking or order data (read only, never write) |
What you don't need
| ✕ | DNS or domain changes |
| ✕ | New software for your staff's daily work |
| ✕ | An IT project with a requirements document |
Control
Here, trust is a technical property
Every promise on this page has a place in the code that enforces it. None of it is a house rule a language model is kindly expected to follow.
Costs with a hard cap
You set a monthly limit. Once reached, drafts continue rule-based – no mail is lost and the bill stops growing.
Predictable
Manipulation attempts hit a wall
If a mail tries to reprogram the AI ("ignore your instructions…"), a code-level check detects it. Such cases always go to a staff member.
Checked in code
Internal stays internal
Internal notes and staff playbooks are separated from customer context at the database level. They cannot appear in any customer mail – technically, not organisationally.
Separated at DB level
Learning on the job
Your team coaches the system in passing
One comment in the ticket is enough: remember: dogs are allowed on request – and the rule applies to every future draft. Every learned rule appears in the knowledge base, waits for approval there, and stays findable.
After a few weeks, something exists that never existed before: the knowledge of your business, written down, searchable and exportable. Including on the day the colleague who kept it in her head moves on.
One remember: – and the same mistake does not come back
Corrections of the same mistake per week, schematic · Grey: chatbot or macro without a learning loop · Amber: SupportPilot – the correction becomes a rule in the knowledge base
Learning loop · schematic, not measured · learned rules wait for approval (governance gate)
Before you ask
The objections people don't like saying out loud
Our business is too specific for something like this.
Our data is a mess. Do we have to clean it up first?
My staff don't want to learn another tool.
What if it writes nonsense and I am liable for it?
We are too small for AI.
How do I know it will work for us?
Let it write on trial for a week
For 7 days, SupportPilot writes drafts for your real requests – nothing is sent, nothing is charged. Your staff compare with what they would have written themselves. On day eight you decide with numbers instead of gut feeling.
No DNS changes · nothing sent during the trial week · card on file, €0 until day 7 · cancel monthly · rather talk first?