Read orders out of WhatsApp
A message, a photo of a handwritten note or a PDF becomes a priced order, checked against stock, with the odd ones flagged for a human.
Pays back in ~6 weeks
AI AUTOMATIONS
The work your team should have stopped doing years ago
Twelve automations we build over and over, with the payback we typically see on each. Pick the one that sounds like your week — or tell us the one that isn't here.
Twelve we have built more than once
Every one of these started as somebody's Tuesday. The payback figures are what we have measured on comparable builds — your numbers land in the written scope, not here.
Answer every enquiry in minutes
New leads get a real first reply, an owner and a follow-up schedule — without waiting for someone to notice the inbox.
Pays back in ~4 weeksBuild the quote from the conversation
The agent drafts a quote from what was actually discussed, prices it off your catalogue, and waits for a human to send it.
Pays back in ~10 weeksPost invoices without typing them
Supplier invoices are read, matched to purchase orders and posted, with the mismatches queued instead of guessed.
Pays back in ~8 weeksChase what you are owed
Overdue accounts get a polite, escalating sequence on the channel each client actually answers, and stop the moment they pay.
Pays back in ~5 weeksClose the month without the marathon
Sales, stock movements and bank lines are reconciled continuously, so closing is a review instead of a reconstruction.
Pays back in ~12 weeksKeep stock honest across channels
One sale updates the store, the warehouse and the books at once. No more selling what shipped yesterday.
Pays back in ~6 weeksTurn documents into data
Delivery notes, contracts, ID documents and forms are read, validated and filed where the process expects them.
Pays back in ~7 weeksRoute work to the right person
Incoming jobs are classified, prioritised and assigned by the rules you actually use — including the exceptions.
Pays back in ~9 weeksOnboard the same way every time
A new hire or a new client triggers one checklist across every system, with nothing skipped and nothing done twice.
Pays back in ~10 weeksAnswer the questions your team repeats
An internal assistant that knows your procedures, price lists and policies — and says "ask a person" when it does not.
Pays back in ~8 weeksWrite the report nobody wants to write
Monday's report builds itself from live data and lands in the channel the team already reads, with the anomalies called out.
Pays back in ~4 weeksNothing in that area yet — try another, or tell us what you had in mind.
Automation that admits what it doesn't know
Most disappointing AI projects fail the same way: the demo handles the happy path and nobody planned for the other 20%. We plan for the 20% first.
-
01
We measure the manual version first
How many times a day, how long it takes, who does it and what it costs when it goes wrong. If those numbers do not justify the build, we say so.
-
02
We automate the boring 80%
The clear-cut cases run on their own. The edge cases are routed to a person with everything they need already assembled — not sent back to square one.
-
03
A human stays on the hook
Anything that spends money, signs something or talks to a client in your name waits for approval. Autonomy is earned case by case, never assumed.
-
04
We watch it in production
Every run is logged. When the model is unsure it says so, and we see the drift before your team does.
WHEN NOT TO
Three times we will talk you out of it
We charge by the hour, so recommending an automation that will not pay for itself would be good for this quarter and bad for the relationship. These are the three cases where the answer is usually no.
See how we price the ones worth doingA process nobody agrees on
If three people do it three ways, automation just picks a fight faster. Settle the process first — we will help, and it is cheaper than the build.
Something you do twice a month
Five minutes, twice a month, is an hour a year. It will never pay for the build, and we will tell you that on the first call.
A decision with real consequences
Pricing, credit, hiring, medical calls. AI can prepare the decision and show its working. It should not be the one making it.
WHERE IT PLUGS IN
It has to live inside the tools you already pay for
An automation the team has to remember to open is an automation the team stops using. Ours run where the work already happens: the WhatsApp thread, the ERP screen, the spreadsheet, the channel where your team already argues about Monday.
When a system has no API we work with exports, e-mail or the database directly — and we say up front which route we would take, so nobody discovers the hard part halfway through.
The questions that decide whether this is for you
-
No, and almost nobody does. Part of the work is reading what you actually have — spreadsheets with merged cells, PDFs, WhatsApp threads — and turning it into something a process can rely on. If a clean-up is genuinely needed before anything else can work, it goes in the quote as its own stage so you can see what it costs.
-
Whichever fits the job and your constraints: a hosted model when quality matters most, a local one when the data cannot leave your infrastructure, and plain deterministic code whenever a model would be an expensive way to write an if-statement. We tell you which is which and why.
-
Every automation is built with a confidence threshold and a human queue. Below the threshold the case goes to a person with the context already assembled; it is never silently guessed. Every run is logged, so when something is wrong you can see exactly what it saw and what it decided.
-
No. We use API tiers with training disabled, or self-hosted models when the data is sensitive enough to warrant it. The choice is written into the proposal, not buried in a vendor's terms.
-
A contained automation is usually live in three to six weeks, including the two weeks where it runs alongside the manual process so you can compare. Bigger ones are staged, and each stage ends with something already doing work.
-
That is the normal case. Most of these connect to an ERP, a CRM, a store, an accounting package or a stack of spreadsheets that already exist. When a system has no API we work with exports, e-mail or the database directly — and we say up front which route we would take.
START WITH ONE
Tell us the task and we'll tell you if it's worth it
Describe the thing somebody on your team does every day and hates. On the first call you get a straight answer on whether automating it pays for itself — and roughly what it would cost.