The AI proposes, the server decides.
The model never touches the database or the money directly. Every turn is either a read or a proposal; the server executes, and checks the request against its own rules again.
Case · Front office · Bookings
Bookaflow
A front office that handles routine questions and bookings itself, and stops the moment something becomes binding. Every proposal waits for approval before it counts.
Four fields, one standard.
Education and learning materialsConsulting and strategyPublishers and editorial teamsResearch and think tanksTeachFoundryBookaflowValoraEducation and learning materialsConsulting and strategyPublishers and editorial teamsResearch and think tanksTeachFoundryBookaflowValora
Who it's forSmall booking businesses
WhatAI front office over WhatsApp
CoreThe AI proposes, the server decides
The challenge
A bot that makes promises gets held to them.
Recurring questions and bookings cost a small business owner hours every week. Putting a standard chatbot on the job is tempting, and risky. A bot that speaks on behalf of your business and makes a wrong promise (a discount, a free spot, a double booking) gets held to that promise. There's case law on this by now.
So the question isn't whether you deploy AI for the front office, but how: without a guest or the model being able to force what something costs, or having something booked that isn't right.
What we built
A front office that answers on its own, but never decides alone.
Bookaflow is a booking platform for small businesses with recurring activities. Yoga studios, workshops, courses. On top of the booking engine sits an AI layer: an assistant that answers customer questions over WhatsApp, states the real price, checks what a punch card or subscription already covers, and sets up a booking.
The starting point is one rule: the AI proposes, the server decides. The model never touches the database or the money directly. If a guest says they get a discount, the system ignores that and calculates the price itself from its own records. Neither model nor guest decides what something costs.
AI front office over WhatsAppproposal, approval, only then booked
“free?”
The AI proposes
The AI prepares it, it doesn't count yet.
The server decides
€ 14.00 · The rate is fixed · a human approves
Your approval
A human approves, only then booked
How it works
From message to booked, with a human in between.
+ 01 · Step
The guest messages.
A question or a booking, over WhatsApp, in plain language. The way a customer always used to ask.
+ 02 · Step
The AI prepares it.
Answers the question or sets up a booking, with the real price and what the punch card or subscription already covers.
+ 03 · Step
The server does its own math.
Price, coverage and availability come from the business's own records. Not from whatever the model or the guest claims.
+ 04 · Step
You give approval.
Anything that's binding or touches money waits in your approval list. One tap: approve or reject.
+ 05 · Step
Only then does it count.
After your approval, the booking is made. Never twice, and every step logged.
The double-checks
Where the safety comes from.
Not one lock, but a series of double-checks. Each one in the code, not in a promise.
Money never moves on its own.
Refunding, paying out and requesting a payment are hard-limited. They can never run on autopilot. A human always signs off before money moves, even if you open everything up yourself.
The server calculates the price.
Does a guest say “I get it for free” or “50 percent off”? The system ignores that and calculates it itself, from its own records. Neither the model nor the guest decides what something costs.
Never double-booked.
Every confirmation has a unique key. A repeated click doesn't book twice. It collapses into the booking that already exists.
Everything logged, nothing to erase.
Every turn (executed, proposed or failed) goes into a log that only ever adds. Afterwards you can trace exactly what happened and why.
A test as the emergency brake.
An automated test scans the AI code for dangerous actions: paying directly, booking directly. If one ever shows up, the build fails. The boundary is guarded by the machine, not by good intentions.
The AI proposes. The server decides. And you sign before it counts.
What it means for the business owner
Time back, without the reputational risk.
The routine disappears.
The questions that come back every week and the simple bookings are handled by the front office itself, day and night.
You keep control.
Everything that matters passes by you, in one list, with one tap of approval. No surprises afterwards.
No wrong promises.
The system can't promise anything that isn't true: no wrong price, no wrong discount, no double spot. That removes exactly the reputational risk an ordinary chatbot runs aground on.
In numbers
How the control is built in.
9 × 4
configurable
Tasks, each set independently: off, propose, assist or autopilot
0
exceptions
Money actions that can ever run on their own, hard-limited
33
tests
Automated tests, including the safety emergency brake
There's no published measurement yet on usage or time saved. What's certain is how the boundaries are built in: in code, and guarded by tests.
Start small, go far.
Want to look for yourself first? The supplier scan and the Chain check are free and need no conversation.
The supplier scanWant something like this for your front office?
In half an hour we'll see whether your routine work lends itself to this, and where the line should be.
Book an intro callHere's what that looks like in practice.
Three different worlds, one principle: the AI does the work, the human checks and approves.
Read the case