In ten working days you get one task working on your real data, with before and after numbers measured the same way. You keep the test set, a model scorecard, a runbook and a fixed price to go live. Or a plain recommendation to stop.
Last updated
Day by day
Ten days. One decision.
Every task is different and the days shift a little. The shape stays the same.
Before day one
Agree the task
One task, not a department. You name one owner. We sign an NDA and a data agreement before any data moves, and settle where the data may go.
Day 1
Measure today
We watch the work and write down the starting numbers: time per item, weekly volume, error rate and cost. Then the money number and the success line. No number worth chasing? We stop on day one.
Days 2 and 3
Build the test set
100 to 200 of your real cases with the right answers attached. Messy ones included. No data clean-up first.
Days 4 and 5
Test the models
Claude, OpenAI and open models on the same cases. Accuracy, cost per task and where the data goes, on one scorecard. Poor results? This is the point to stop.
Days 6 to 8
Build the first piece
Inside the tools your team already uses. A person approves anything important. Anything it was not tested on goes to a person, as it does today.
Day 9
Run it on real work
This week's work, next to your team. Measured again, the same way as day one.
Day 10
The readout
Before and after numbers, everything below, and a fixed price to go live. Or a plain recommendation to stop.
What you get
Yours to keep.
Before and after numbers
Time, cost and accuracy, measured the same way on day one and day nine.
A test set from your real work
100 to 200 of your cases. It stays with you, so you can check any AI tool you try later.
A model scorecard
Accuracy, cost per task and where the data goes, for every model tested.
A working first piece
Running on your data, inside your tools. It carries on to production. Not a throwaway demo.
A runbook and handover doc
How it works, how to check it and what to do when something looks wrong.
A fixed price to go live
A 4 to 6 week plan, with the result we must hit agreed in writing.
Who it is for
Built for one busy task.
One task eats your team's week.
High volume, heavy on text or documents. Invoices, purchase orders, supplier emails, contracts, claims, onboarding files.
A pilot looked great, then stalled.
We test what you have on real work and tell you why it stopped. Or start with the 5-day pilot check-up.
You want numbers before a budget.
You leave with your own numbers and a fixed price. Not a deck.
You can name an owner.
One person who knows the task, sample data, system access and about two hours a week.
Who it is not for
Sometimes the answer is no.
We would rather lose a project than build the wrong thing.
A rule will do.
If the right answer can be written down, write the rule. Cheaper, faster, easier to audit.
The volume is small.
A task that happens a few times a month rarely pays for an AI system.
The process is the problem.
AI will not fix an approval stuck with someone who is travelling.
A mistake costs too much.
Some decisions need a person to own them. AI can gather facts and draft.
You want a multi-year programme.
We start with one task. Large programmes suit a different kind of firm.
Questions
Straight answers. Anything else, just ask.
Skcript Applied Labs
Give us one workflow.
We will show it working on your data in 10 days.
Book a 20 minute call. We will tell you honestly whether it is worth doing.