We set it up. Your team uses it. That is the whole service.
No platform to buy, no dashboard to learn, no software from us at all. We work out which parts of your week are worth handing to AI, set Claude or ChatGPT up on your own machines around your own documents, and train the people who will actually use it.
What we actually do
Three steps, and then the parts of the job people do not expect to be included.
Work out what is worth doing
Half a day watching how work really moves through your business — not how the process document says it does. You leave with the three tasks worth handing over, ranked by hours returned, and the ones you should leave alone.
Pick the right tool for each job
Claude and ChatGPT are genuinely better at different things, and almost nobody switches because nobody told them there was a choice. Getting this right is usually the biggest single jump in output quality, and it costs nothing.
Set it up on their machines
Correct account tier, configured around your templates, your clients and your wording, with the three tasks working end to end. On the computers your team already uses — not a portal they have to remember to open.
Sit with the people who will use it
Training means sitting next to someone while they do the task themselves, once, on real work. Not a recorded course and not a manual nobody reads. This is the part that decides whether any of it survives to month three.
Answer the data question in writing
Which account tier, whether your work is used to train a model, where it is processed, and a one-page note you can show your own clients or your professional body. For accounting and legal firms this is the first question, so it is part of the job.
Keep it from rotting
Optional and monthly. New staff need setting up, the tools change every few months, and new tasks surface once people get the idea. Without this, most setups quietly decay within a year.
Why this works when the last attempt did not
Most people have already tried. The tool was fine — the setup was missing.
It has seen your work
Out of the box it has never seen your templates, your clients, your pricing, or one example of what a good version looks like in your business. Half the job is putting that material in front of it properly.
Fixed price, in writing
After the call you get one number for the whole job, in writing, before you commit — and it does not move afterwards. The session comes off the setup if you go ahead. If we have misjudged the scope, that is our problem — not a change order.
You own all of it
The accounts are in your name, on your machines. There is no platform of ours in the middle, nothing to log into, and nothing that switches off if you stop working with us.
Your team is already using AI
The only question is whose account it is running on.
On the right tier, your work is not used to train a model
You can see who has access and remove it when someone leaves
It stops living on personal logins you do not control
You get the answer in writing, with where processing happens
A one-page note for your clients and your professional body
And a short list of what should never go in, whatever the tier
What is different about this
Mostly unglamorous things that decide whether it is still in use next year.
Built around your documents
Your templates, your house style, your past work as the reference for what good looks like. That is the difference between a generic paragraph and something you would actually send.
On the machines they already use
The desktop apps on their own computers. Nothing new to log into, nothing new to remember, no portal of ours sitting in the middle of your business.
The right tool per task
Claude for some jobs, ChatGPT for others. Most teams are using one for everything because nobody told them there was a choice.
One page per task, not a manual
Each task gets a single page anyone can follow. Long documentation goes unread, which is how setups quietly die.
A list of what not to automate
Plenty of what people want to hand over is not worth the cost of setting up or babysitting. Saying so is part of the job, and it is usually the most useful page in the report.
Adoption guaranteed for 60 days
If your team is not still using it 60 days after handover, we come back and fix it free until they are. Installing something is easy; being used in month three is the actual job.
Questions about the work
The ones that come up on almost every first call.
Are you selling us software?
No. We do not sell a product, a platform or a dashboard. You end up on Claude or ChatGPT accounts in your own name, set up properly around your own work. If you stop working with us, everything keeps running exactly as it is.
We already pay for ChatGPT or Copilot. Why pay you as well?
Because that is the problem rather than the reason not to. You are already paying for the capability and getting a fraction of it — so the cheapest improvement available is not another subscription, it is making the one you have actually work. Most engagements add no new tools at all.
Will our team need to learn new tools?
They will need to learn three tasks, on the machine they already use, and we will sit with them while they each do it once. There is no new system to adopt, which is deliberate — adoption is where this kind of work normally dies.
What happens to our data?
It depends entirely on the account tier, which is why it is part of the paid session rather than a footnote. Consumer tiers can use what you type to improve models unless that is turned off; business and enterprise tiers do not train on your data by default. You get the specifics in writing, including where processing happens, plus a one-page note for your own clients or professional body.
How long before we see anything working?
On the first call, if you bring a document — we will do a real task with it on the call. After the session, setup is usually days rather than weeks, because we are configuring tools that already exist rather than building software.
What if you think we should not do this?
Then we say so. A useful part of the session is the list of things not worth handing over. We would rather set up one task that is still in use next year than five that get abandoned by March.
Bring one real document
Twenty minutes on a call. We will do an actual task with it there and then, and tell you whether the rest is worth doing.