From first conversation
to AI staff that work.
Six stages take you from an intake conversation to AI staff doing real work inside your business. You hold the gate at every one of them, and nothing reaches a customer until a person has said yes.
- Reply to Tom · website enquiryNeeds you
- Booking · Thursday inspectionNeeds you
- Follow-up · PriyaNeeds you
- Reply to Marcus · approved by SarahSent
- Approve: it goes out as written
- Edit, then approve
- Reject: nothing happens
Tell us about your business
The first stage is a conversation, not a form full of jargon. We ask what your week actually looks like: which tools you already pay for, where enquiries land, which jobs keep slipping past Friday and what an ordinary busy day does to your team. You finish with a written picture of the problem that you have read and corrected, and with nothing built, nothing connected and nothing owed.
Plain questions, no jargon
Questions about your business, not about software. If a word needs explaining, we explain it or we stop using it.
Your tools, written down
We note the inbox, calendar and CRM you already work in. Noting them is all that happens at this stage; nothing is connected yet.
You correct the record
The intake goes back to you before it becomes a plan. If we have misread the problem, this is the cheap place to find out.
- Where do enquiries come in?
- Which tools does the team live in?
- What keeps slipping to next week?
- What would a good week look like?
- Email inboxNoted
- Shared calendarNoted
- CRMNoted
Get a plan you approve
Your intake becomes a written plan: which AI staff we would build, what each one is allowed to do, which of your tools they would touch and a fixed price for the work. You read it, ask for changes, and ask again. The plan sits at “awaiting your approval” for as long as that takes, and no builder starts while it does. The stage ends when you sign off on a scope you recognise as your own business.
Scope in plain English
One page per AI staff member: the job it does, the tools it touches, and the actions it is never allowed to take on its own.
A fixed price, written down
The price is agreed before any work happens, so the cost of the build is a decision you make with the number in front of you.
Revise it as often as you like
Sending the plan back is expected, not an inconvenience. Approval is a deliberate act; silence is never taken as a yes.
- Scope · 2 AI staffDrafted
- Tools requested · inbox, calendarScoped
- PriceFixed
- BuildNot started
- Approve the plan as written
- Send it back with changes
- Walk away: nothing has been built
Our AI crew builds it
Five specialist AI builders take the approved plan apart and build it: the connections to your tools, the way your AI staff talk, the guard rails on what they may do, and the approvals queue every draft has to pass through. Our AI Director reviews each piece against the plan you signed and sends work back for rework until it is good enough to move on. You do not have to supervise any of this. The stage ends when the build is ready to be tested, not when it is ready to be used.
Five builders, one director
Each builder owns one part of the job, and none of them decides on its own that the work is finished.
Built to the plan you signed
The approved scope is what the review is scored against. Anything outside it comes back to you as a change, not as a surprise.
Reviewed before you see it
Weak work is rejected internally and rebuilt. What reaches you at the next stage has already survived a quality review.
- Integrations builder
- Conversation builder
- Guard-rails builder
- Approvals builder
- Reporting builderBuilding
- Matches the approved scopePass
- Every send routed to approvalsPass
- Booking wordingSent back
We test it in a sandbox
Before a single real customer is involved, your AI staff run on test data in a sandbox: sample enquiries, sample calls, sample bookings on a calendar nobody uses. We keep the log of everything they drafted, booked or got wrong, and you get the QA verdict exactly as it came out, including the parts that failed. If the verdict is not good enough, the stage repeats instead of moving on.
Dry runs on test data
Real-shaped work, fake people. No customer of yours is contacted and no live record is touched at this stage.
The evidence, not a summary
You read the actual drafts and the actual failures. It is a lot less flattering than a promise, and a lot more useful.
A verdict that can say no
A failed check sends the work back to the builders. Nothing is waved through because the stage was scheduled to end.
- Reply quality on sample enquiriesPass
- Test bookings landedPass
- Out-of-hours wordingFix first
- Sent to real customersNone
You approve go-live
Nothing switches itself on. We show you exactly what will be live, which accounts your AI staff can reach, and which actions stay closed to them, and then the switch waits for you. Sign off and your AI staff start work; ask for another round of testing and they stay in the sandbox until you are happy. The access they get is connected per client, scoped to the job in the plan, and revocable from your own account without asking us.
Your sign-off is the switch
There is no automatic launch date. Go-live happens at the moment a person on your side says it does.
Access scoped to the job
Your AI staff get the narrowest access the plan needs, connected to your accounts and revocable from your side at any time.
Sending stays gated
Going live turns the work on, not the sending. Every outbound action still waits for a person, from the first day onwards.
- Draft replies to new enquiriesWill go live
- Prepare bookings for approvalWill go live
- Send without a person approvingOff
- Access · inbox and calendarRevocable
- Reviewed by Sarah
- Switched on at that moment
- Paused again whenever you ask us to
It runs, with a human in the loop
Live does not mean unsupervised. Every email, booking, message or record change your AI staff want to make arrives in one approvals inbox as a draft, with the reasoning attached, and waits for a person to approve, edit or reject it. Approve it and it goes out; ignore it and nothing happens at all. Say the word and we pause the lot, and every decision, yours and theirs, stays on the record.
One inbox for everything
Calls, emails, bookings and updates queue in the same place, so approving your AI staff is one habit rather than five.
Approve, edit or reject
Editing a draft before you approve it is the normal case, and it is how your AI staff learn the wording you actually use.
Pause on request, audit at will
Ask us to stop and everything stops, and you can read back what was drafted, what was approved and by whom, whenever you want to.
Day Eight AI staffDrafted a reply to Tom’s enquiry and held a Thursday inspection slot. Approve, edit or reject?Day Eight AI staffEdited and sent. The held slot is still waiting on you.- Approved by a person, then sent1
- Still waiting on you2
- Sent without approval0
What holds at every stage
Six stages, one set of rules. These do not change between the first conversation and the thousandth approved draft.
A human approves before anything leaves
From the plan to the thousandth draft, a person on your side says yes before anything reaches a customer. It is the product, not a setting.
Your tools stay yours
Your accounts stay in your name. Access is connected per client, scoped to the job written in the plan, and you can revoke it yourself at any time.
Evidence before go-live
Nothing is switched on because it sounded good in a meeting. You read the sandbox log (drafts, bookings and failures) before a real customer sees anything.
Questions people ask before stage one
Nothing is ever sent, booked or changed without human sign-off.
How long does the whole thing take?
It depends on how much you want built and how fast the plan gets approved: a single AI receptionist is a much shorter job than a full agent team. You get a timeline in writing inside the plan, before you commit to anything, and we would rather quote it after the intake than guess at it on a website.
What do you need from me?
An honest intake conversation, access to the tools you want automated (scoped to the job and revocable from your side), and one person who can approve drafts once it is live. That last one matters most: approvals are quick, but they are not optional.
What happens if it gets something wrong?
A wrong draft is still only a draft. You reject or edit it, nothing goes out, and your customer never sees it. That is exactly what the approval step is for. We then fix the behaviour behind it and put the change back through sandbox testing before it returns to your inbox.
Can I turn autonomy on?
Not for anything that leaves your business. The preparing (drafting, researching, sorting, holding a slot) already runs unattended; the sending, booking and changing waits for a person, and that gate is not a switch we offer to flip. If you want an AI that emails your customers unsupervised, we are the wrong agency.
What does it cost?
A fixed price, quoted in the plan after your intake, so the number is based on your business rather than a package. There is no price list here because we do not know what you need until we have asked. The discovery call is free, and you read the price before any work starts.
How do I stop it?
You tell us, and we pause your AI staff: the queue stops filling and anything already drafted simply sits there unsent. Because the tool access is connected to your own accounts, you can also revoke it from your side without going through us, and the record of what happened stays readable either way.
Ready for your eighth day?
A free 30-minute discovery call. We’ll find the automation with the biggest payoff for your business. No obligation, no jargon.