AI, explained
Where AI is worth paying for, and where it isn’t.
Most of what you hear about AI is aimed at companies far larger than yours. For a small company the useful part is narrower: a few repetitive jobs it does well, at a cost you can measure.
Common questions
What people ask us first.
- Is this just hype?
- A good deal of it, yes. The parts that work are narrow and unglamorous: reading documents, sorting messages, drafting text a person then checks. We’ll tell you which category your idea falls into before you spend anything on it.
- Will it replace my staff?
- Not in the work we do. It takes away the parts of a job nobody wanted — the retyping, the sorting, the first draft — so the same people get further with the work only they can do.
- Is my company information safe?
- That depends on choices made before anything is built: what gets sent where, what’s kept, and what never leaves your systems at all. We make those choices with you and put them in writing.
- What does it cost to run?
- Each task has a price every time it runs. It’s usually small, but never nothing. We set a ceiling on it and report what was actually spent, so it doesn’t turn up as a surprise on a bill.
- What happens when it gets something wrong?
- It will get things wrong. Anything it’s unsure about goes to a person before it counts for anything, so mistakes are caught while they’re still cheap.
- How would I know whether it worked?
- We measure the job before we start. If the new way doesn’t beat the old one, we say so and you stop. That’s why the first version is small.
Where it pays
Four jobs it does well.
All four are repetitive, high-volume, and easy for a person to check.
Reading what arrives
Invoices, orders, delivery notes, forms. Instead of someone typing the details into your system, they arrive already entered, with anything unclear set aside for a person.
Sorting the inbox
Hundreds of messages a week, most of them routine. They get grouped, sent to the right person, and given a draft reply, so your team starts with something.
The writing you repeat
Quotes, standard replies, summaries of long email threads. A first draft appears in seconds; a person edits it and sends it. The judgment stays with you.
Finding what’s buried
Contracts, past jobs, years of email. Ask a question in ordinary words and get an answer, with a link to the document it came from so you can check it.
Where the time goes
Most of a week goes on work nobody chose.
Typing things in. Sorting what arrived. Writing the same reply again. None of it is why anyone started the business, and all of it has to get done.
- We find out where the time actually goes before suggesting anything
- Jobs that wouldn’t improve are listed too, with the reason
- You see what each one costs to run, with a ceiling on it
Where the hours go
Checking the work
A person still decides.
Anything it’s confident about goes through. Anything it isn’t goes to somebody on your team, with the reason attached. You decide where that line sits.
- Uncertain items go to a person, with the reason attached
- Every decision is logged and can be reviewed later
- You choose how cautious it is, and can change your mind
Read, entered, and sorted
Limits
Where we’ll tell you not to.
Most of what we’re asked about falls into one of these. Saying so early is cheaper for everybody.
Where being wrong is expensive and hard to spot
If a mistake could cost you a customer and nobody would notice for a month, this is the wrong tool.
Where an ordinary rule already works
If a simple condition covers it, we write the condition. It costs nothing to run and never surprises you.
The judgment people pay you for
If customers buy your expertise, don’t automate the expertise. Automate the paperwork around it.
Anything you’d have to justify later
If an insurer, a regulator or a customer might ask why a decision was made, you want an answer you can point to.
Starting
How we’d begin.
One job, a small version of it, and an honest comparison at the end.
Pick one job
The one that takes the most hours and that nobody in the building enjoys.
Measure what it costs now
Hours, people, mistakes. Without that number there’s nothing honest to compare against later.
Build a small version
On your real work, with a person checking what comes out of it.
Compare, then decide
If it beats the way you do it now, we grow it. If it doesn’t, we stop before it has cost you much.
Something in your week is eating hours.
Describe it in ordinary words. We’ll tell you whether it’s a job for AI, a job for a simple rule, or a job to leave as it is — before you spend anything.