Every enquiry you don’t answer goes to somebody else.
They arrive at nine at night and on Saturdays, while your team is retyping invoices and rebuilding Monday’s report. Six questions, your own numbers:
- What we ask
- How many hours a week the task takes, how many people do it, what an hour of their time costs, how many enquiries you get, how many are answered late, and what a customer is worth.
- What you get back
- What the task costs you a year, what automating most of it would save, and which day of each month the monthly plan has covered its own fee.
- What we assume
- That automation removes about 60% of those hours, not all of them. You can change that, and every other assumption, on screen.
Work it out yourself in one line: hours a week × people × 52 × what an hour costs you. Automation takes roughly 60% of that. Compare it to the monthly price above and you have your answer without talking to anyone.
We build the AI that does it instead. Anything that touches money or a customer still waits for your yes.
No email required. Nothing is stored.
Don't take our word for it. Give it something to read.
Paste an invoice. Watch what it reads — and what it refuses to pass on.
Read cleanly
- Supplier
- Example Office Supply Co.
- Invoice
- INV-2026-1001
- Amount
- USD 34,624.95
- PO
- PO-5001 — exact match
Every line adds up and the supplier is on file. In a real build this goes on to the next step — and still stops before anything that moves money.
Held back — a person needs to look
- Supplier
- Example Security Systems
- Invoice
- INV-2026-1007
- Amount
- USD 17,885.33
- Reason
- total does not match the line items
Nothing was passed on. Nobody outside your office ever saw it.
That ran in your browser. Nothing was sent anywhere and nothing was stored. It is a parser, not a chatbot — whether the line items add up is arithmetic, and a language model can be confidently wrong about a sum. In a real build the model reads the layouts nobody has seen before, and this same hard check still guards the money.
That is one check out of fifty — open the whole system and inspect every step
Three jobs we are asked for most.
Work that stops arriving on your desk.
Bell
Receptionist
Answers whoever messages you — chat, email, WhatsApp. Books what it is allowed to book, and holds anything that would commit you.
Available now
Reed
Enquiry Desk
Answers a new enquiry in under a minute, qualifies it, routes it to the right person.
Available now
Tally
Accounts Clerk
Reads every supplier invoice, matches it to the PO, drafts the bill — and stops dead when the totals disagree.
Available now
Also built on the same skeleton: Otto onboarding coordinator, Wren reporting analyst. Ask for either during Discovery.
Four messages arrive one evening — see what Bell does with each
What it costs, and what happens next.
The build is priced after Discovery, confirmed in writing, and paid once. These plans keep it running afterwards. We maintain what we built, because we know how it behaves when it breaks.
AI Care
One workflow, business hours
AI Plus
Up to three workflows, answered first
AI Care covers one workflow. Once you run a second or third, AI Plus covers all of them and adds tuning and first-in-line answers. Connected systems of several workflows go on an Operations plan. We do not publish prices: each is set after Discovery and confirmed in writing, and Discovery is credited against the build if you go ahead within 14 days.
The AI itself bills separately: we pass through what the models actually cost, with the markup stated on the invoice. We will not bury it in the monthly fee.
How long before it is actually running?
Discovery comes first and is paid and fixed. It produces a written scope, a fixed price and a delivery date, and you see all three before you commit to the build.
We quote the timeline in that same document. Nobody sensible quotes you a date before they have seen the process, and anyone who does is quoting a guess.
What do you need from us?
Less than you are expecting. An hour or two with whoever actually does the job today, and read access to the systems it touches.
You do not have to write a specification or know what is possible. We watch the process, write down what we saw, and you correct it — that document becomes the scope.
Am I stuck with you afterwards?
No, and the build is designed so you are not. You get the code, the documentation and the accounts it runs on. It is yours.
The monthly plan exists because automation drifts when nobody watches it — models get retired, a supplier changes their invoice layout, costs creep. Watching for it, and fixing it before you notice, is what the monthly plan buys.
What does it actually cost to run, on top of the fee?
The AI itself. We pass through what the models cost with the markup stated on the invoice, so you can see it. We will not bury it in the monthly fee.
It is usually small next to the hours it replaces, and the calculator above does not count on it either way. If a quote from anyone does not tell you what the running cost will be, ask before you sign.
What happens when it gets something wrong?
It stops. Anything that moves money, reaches a customer or commits your business waits for a person to say yes.
When the software is unsure it does not guess — it hands the job back with a note saying why, and every run is written down so you can see what happened and when.
How will we know whether it worked?
We agree the measure before we build, during Discovery, and write it down: usually hours off a specific task, or how fast enquiries get answered.
Anyone quoting you a percentage before they have seen your process is guessing. We agree the measure first, in writing, then hold ourselves to it.
Bring us the task that eats your week.
We will write back with how we would start and what it costs, usually within one business day.