Rule 01
Deterministic by default
If a rule can express it, a rule expresses it. Language models are used where the input is genuinely unstructured, and their output is validated by code before anything acts on it.
ServicesBiggest time saved
Hours nobody billed, handed back to people who can.
The cost of a manual process is not the annoyance. It is the invoice nobody sent because a senior person spent Thursday copying fields between two systems. Automation projects fail when they start with a platform instead of that number. This one starts by costing your process in hours per month, ends about three weeks later with it running unattended, and is priced so you can buy one and judge us on what it gave back.
What this is worth, and who measured it
21×
A lead contacted within five minutes was 21 times more likely to qualify than one contacted after thirty. Most small firms reply the next working day.
4.9 hrs
Knowledge workers put 58% of the working day into coordinating work rather than doing it, and estimate better processes would give back 4.9 hours a week each. Multiply that by a charge-out rate.
These are findings from published research on the wider market, measured by the people named beside each one. They are not results we are promising you, and we do not have client results of our own to show yet, which we would rather say plainly than dress up.
Three sizes, and what moves the number
Small
$2,400 to $5,500
usually two to three weeks
One clean process with clear rules. Invoice chasing, an onboarding sequence, a report that builds and sends itself. The kind of job everyone agrees should not still be manual.
Medium
most firms$5,500 to $13,000
usually four to six weeks
A process that spans several systems, reads real documents, and needs a person to approve the part where being wrong is expensive. Exceptions handled properly rather than ignored.
Large
from $13,000
quoted after a paid scoping week
Several linked workflows, a trail that has to stand up to an audit, or a legacy system with nothing resembling an API. Scoped properly, because guessing at these is how they go wrong.
What moves the price
- How many systems the process touches, which is almost always the biggest single factor
- Whether those systems have decent APIs. A modern one is hours, a legacy one with no API can be weeks
- How much of the input is unstructured, because reading a PDF reliably is harder than reading a form
- How many exceptions are real. Every "except when…" your team mentions is a branch to build and test
- Whether a person has to approve something, and what has to be shown to them to make that quick
- Whether the trail has to stand up to an audit, which is a different standard of work
If you are not sure which size you are, the free twenty-minute call will place you within about a thousand dollars, and it costs nothing to find out.
The sprint, week by week
Week 1
Map it, cost it, fix the scope
We sit with the people who actually run the workflow and document it as it really happens, including the exceptions and the workaround everyone forgot was a workaround. Then we agree in writing what the automated version does, what it explicitly does not do, and what it costs to run per month.
- If the map shows the process should be changed rather than automated, we say so. Automating a bad process just produces bad outcomes faster.
Week 2
Build it, with AI only where it earns its place
The automation gets built and connected to the systems you already pay for. AI does the parts it is genuinely better at: reading an unstructured document, drafting a first response, classifying an enquiry, summarising a thread. Everything else is ordinary deterministic code.
Week 3
Run it beside the humans, then in front of them
The automation runs in parallel with your current process on real work and we compare the two outputs. Anything it gets wrong is fixed while the stakes are zero. Only once it agrees with your team does it move in front of them.
Handover
Yours, documented, and watched
A recorded walkthrough, a written runbook and the credentials. It runs on infrastructure we host and monitor, so a failure pages us rather than surfacing as a client complaint three days later. Thirty days of adjustment included.
How we decide what to automate
The honest constraint is not what is technically possible, it is what stays correct when nobody is watching.
One workflow at a time, fixed scope, fixed price. No platform migration, no year-long program, no seat licenses.
Rule 02
A human where it is expensive to be wrong
Money leaving the business, a client-facing document, anything with a filing deadline: a person approves it. The automation does the work and shows its reasoning, so approval takes seconds.
Rule 03
Fails loudly, never silently
Every automation states in advance what it does when the other system is down or the input is malformed, and who gets told. Silent failure is the most expensive thing an automation can do.
Rule 04
Your tools, not ours
We connect to what you already pay for. If a workflow can be done inside your existing practice system, that is the answer, and it is cheaper for you than anything we would sell instead.
What you get at handover
№ 01
The process map
Your workflow as it actually ran before we touched it, and as it runs now. Useful long after this project.
№ 02
The running automation
Hosted and monitored by us, connected to your systems, with the failure paths defined and the alerts pointed at a human.
№ 03
The runbook
What it does, what it costs to run, how to pause it, how to change the parts you will want to change.
№ 04
A recorded walkthrough
Ten minutes, so the person who joins next year can understand it without booking a call.
What this is not
What the sprint is not:
- Not a platform migration. We automate around the systems you have.
- Not unlimited scope. One workflow, agreed in writing. A second is a second sprint and it is usually cheaper.
- Not a license to a product of ours. You own the automation and the account it runs in.
- Not an excuse to make people redundant. Every one of these so far has been about removing work nobody wanted to be doing.
Fair questions
Where do the hours actually go, and what is getting them back worth?
Pick the job your team complains about most and we cost it in hours a month at your real rates, before anything is quoted. The arithmetic is deliberately boring: hours a week, times a fully loaded hourly cost, times fifty-two. Ten hours a week across a team at $30 an hour is about $15,600 a year sitting in one task, and firms are routinely surprised by that number because nobody has ever added it up. There is a calculator on our pricing page that does it with your figures and sends nothing anywhere. That number, the one you confirm out loud, is what the decision turns on.
Which workflow should we start with?
The one that is high volume, rules-driven and universally disliked. Client intake, quote generation, invoice chasing, customer onboarding and recurring reporting are the usual five. If you are not sure, the free opportunity call will get you to an answer in twenty minutes.
Why is this so much cheaper than the quotes we have had?
Partly scope, because we deliberately sell one workflow rather than a program, and partly overheads: we are a small studio in Chiang Mai, not a city agency with a sales floor. The market range for a single scoped workflow is $3,000 to $15,000, and for a whole AI automation project $5,000 to $50,000. Ours is $2,400 to $13,000 across three sizes, under both. We would rather undercut and say why than charge agency rates without an agency behind us.
What does it cost to run each month?
Whatever the third parties charge, estimated to the dollar before you commit. Typical workflows land between a few dollars and a few tens of dollars a month. If a design would be expensive to run, we redesign it rather than passing you a surprise.
Can you automate something that needs judgement?
Partly, and it matters to be precise about which part. A model can read the file, draft the response and explain its reasoning. A person still approves it. That removes most of the minutes while keeping all of the judgement, which is a much better trade than trying to remove the person.
Start here
One workflow. Three weeks. One price.
Tell us which job your team would most like to stop doing and we will tell you honestly whether it is a good candidate.
John Weeks
The person who replies is the person who does the work.
Next on the path: AI Visibility (Once the work runs itself, the question is whether anyone can find you.)