The quote drafter
Reads the enquiry and builds the first version from your rate card. You check it and send it. Trades, travel, print, events.
Put AI to work
Small agents that each do one job, inside the tools you already use
Nothing changes for the person doing the job.
The agent works inside the tools they already have open, and it drafts. You press send. An agent is a small worker with one task: it reads what came in and writes the first version of what goes out, and it logs every action it takes. The quote that used to leave your desk two days later leaves it the same evening.
01
What it does to your money
Speed of reply decides a lot of enquiries, and the one that arrives at 19:40 no longer waits until Monday.
Quotes leave the same day, so the follow-up happens while the customer still remembers asking.
The evening work stops being evening work. Your people take on more without staying late, and you postpone the next admin hire rather than the next truck.
Invoicing moves closer to the day the job was done, which moves your cash forward by the same distance.
02
One job each. Here are real ones.
The follow-up agent
Notices a quote has gone quiet for five days and drafts the nudge in your tone, with the reference and the number already in it.
The inbox sorter
Reads what arrives overnight and separates a real enquiry from a supplier invoice from a nonsense mail, putting each one where it belongs before you are awake.
The slip reader
Turns photographed receipts and fuel slips into coded rows in your reconciliation sheet, and flags the ones it cannot read instead of guessing at them.
The document filler
Takes the details off a passport or an ID and populates the forms that need them, so nobody types a number twice.
The morning brief
At 06:30 it tells you who to call today and why. Four names and a reason, not a dashboard.
03
How an agent gets built
We pick one job
The smallest job that eats real time every week. One job, not a platform.
We watch a person do it ten times
That is where the rules actually live. Nobody can write them down from memory, including the person doing it.
We build it and run it beside you
For the first stretch it drafts and a human sends. You see everything it produces before a customer does.
You decide what it may do alone
Some clients keep the human check forever on anything customer facing and let the agent act alone on internal work. That is a fine answer and we do not argue with it.
We add the second agent only when the first one is boring
Boring means it works and nobody talks about it any more.
04
What keeps it under control
It drafts, you send
By default nothing reaches a customer without a person seeing it.
Every action is logged
What it read, what it wrote, at what time, and what it skipped. You can read any run afterwards without asking us for it.
It has limits
How many items it may process, and what it may not touch.
It says when it is unsure
A flagged item waiting for a human costs less than a confident mistake, and it never invents what it cannot see on the page.
There is an off switch
One click and it stops. The old way of working still exists, because we never removed it.
What the agent can reach, and what it cannot
The agent runs inside your accounts, not ours. Your inbox, your drive, your accounts package and your sheets, under access you grant and can take back.
You grant it from your own admin screens. We never ask for your password and the agent never uses an admin account.
It gets the narrowest access that does its job: one inbox label and one folder, and nothing wider. A quote drafter has no reason to see payroll and does not get it.
You revoke it in one click from the same screen, without going through us, and the rest of your business carries on.
We do not ship your files to a platform of ours to be processed and stored. Text the agent sends to an AI model goes to that provider's business API, encrypted in transit, on a plan whose terms state the content is not used to train the model. The provider for your build is named in your proposal.
Every run is logged with a timestamp, in your own environment, where you can read it.
06
When an agent is the wrong answer
Count how often the job happens. Twice a month and the build will never pay for itself, so do it by hand and put the money somewhere else.
If being wrong is expensive and nobody would notice for a week, keep a person on it. Pricing and credit decisions belong to a human, and so does anything a regulator reads.
If the information the agent needs only exists in phone calls nobody records, there is nothing for it to read. That step stays manual until something writes it down.
And if nobody is putting the reference number in the WhatsApp caption, agreeing how files get named and where they go solves most of the pain for nothing. An agent built to work around a habit is the more expensive way to fix it.
07
See one working before you buy one
The demos on this site run in your browser on sample data we made up. Watch a slip become a reconciliation row, or a passport become a filled form, without uploading anything or giving us your name.
When you want to see it on your own material, WhatsApp us one real example. One enquiry and the quote you sent back is enough to show you what a quote drafter would produce for you.
The numbers, in one place
- Putting a standard agent in costs nothing.
- A complex build, one with many decisions to make, many exceptions, or a lot of your own material to learn from, is quoted separately, before we start, so you see the number first.
- All prices exclude VAT.
- What moves whether a build counts as complex: how many decisions the agent has to make, whether your tools have a proper way in, how many exceptions it must handle rather than hand to a person, and how much of your own material it has to learn from. Then a monthly subscription, priced by how much work the agent does: | What the agent does | A month | |---|---| | A simple agent: one job, one format, one destination | From R1,000 | | A standard agent: several formats, your own rules, writing into two systems | From R2,500 | | A heavy agent: high volume, many exceptions, several people relying on it | From R5,000 | The subscription covers the agent running, the model costs, monitoring, and the changes when a supplier alters a format or when your rules change. Nothing extra to buy when something shifts. You can stop. It is month to month, you own your data, and the process map stays yours. One agent doing one job is the cheapest way to find out whether this works for you.
Straight answers
Questions owners ask us
Not in the work we do. The jobs we automate are the ones that pile up after hours and on the weekend. The intent is that the same people handle more work without staying late, and the person who was retyping does the part that needs a human.
It will, sometimes. That is why it drafts and a person sends, why every action is logged, and why it flags what it cannot read instead of guessing. Design for the mistake and the mistake stays cheap.
No. The agent works inside what you already use. It needs a way to read and write, which most current tools have. If yours does not, we say so early rather than building something fragile.
It runs on your accounts and your subscriptions, so it keeps running. You own the code, the written build document, every login and the account it lives in, from the day it goes live. Another developer can read that document and take it over, and we will walk them through it on a call. There is no licence of ours in the middle and nothing to switch off.
A first single job agent is usually running beside your team within three to four weeks, including the watching period. You see draft output well before that.
Putting a standard agent in costs nothing to build. Then a monthly subscription, priced by how much work the agent does: from R1,000 a month for a simple agent doing one job in one format, from R2,500 a month for a standard agent handling several formats and writing into two systems, and from R5,000 a month for a heavy agent with high volume and several people relying on it. We show you which tier fits, and the number, before you sign.
The next step
Send us one enquiry and the quote you sent back, by WhatsApp to +27 73 932 6852 or to [email protected].
We will build the draft an agent would have produced and show it to you within two working days. Or call that number and book twenty minutes to talk it through first.
Or email [email protected]. No slides, and no invoice for the conversation.