How we choose

Most of the work is knowing where it pays.

Most of the work isn't building. It's knowing where AI pays.

Every owner has heard the pitch. Few have been shown the two or three places in their week where it would earn its keep. We start there.

We count the week before we name a tool.

We start with how work moves through your business. Who asks for what. Where it waits. Who chases it.

Then we follow three or four chains from start to finish. A quote to an order. A bank line to a closed month. A customer question to an answer.

Every chain stalls somewhere. Usually it's a person reading, matching or chasing by hand. We count those items, and we look there first.

We look at the owner's week too. Promises and decisions live in one head, and some slip.

Six departments, the same three patterns.

• Sales: enquiries answered a day late, and promises lost in the sent folder.
• Finance: bank lines matched in one rush, and sign-offs waiting on one desk.
• Operations: supplier dates nobody checks until one is missed.
• HR: applications read in order of arrival, and first-day steps that slip.
• Customer service: the same status question, forty times a day.
• Marketing: channel dashboards stitched together by hand.

Underneath, it's reading, matching and chasing, in every sector we work across.

Five kinds of work it suits.

It does well where the work is heavy on reading and shaped by rules:

1. reading a pile of email and picking out the few that need a person;
2. noticing a promise in a sent message and remembering its date;
3. comparing what arrived with what was expected;
4. drafting a first chase note for a person to approve;
5. spotting when a pattern breaks, and saying “worth a look.”

A good assistant could do each of these with enough hours. The hours are the problem. That's where it pays.

Five kinds it doesn't.

1. Decisions that need judgment you would never hand over.
2. Processes that already run fine.
3. Problems that are really a missing policy or a missing hire.
4. Volumes too small to be worth building for.
5. Data too messy to read reliably yet.

We'll tell you when AI is the wrong answer. Sometimes the fix is a checklist, a clearer owner or a tool you already pay for. We'd rather say so than build something you don't need.

What we build reads, drafts and flags.

It reads and warns. A person approves before any message leaves.

When something might be wrong, it says “worth a look” and stops there. It never acts in your name on its own.

The first conversation: one hour, no access.

It's with one of our co-founders, and it's free. Here is what to expect:

• No access. We don't ask to see your systems or your mail.
• No preparation. You don't need to send documents first.
• Your whole list. You may arrive with one job in mind. Bring the other five.
• Questions about your week. Where work waits, who chases, what lands on you at night.
• Names to leave with. The jobs worth solving, in plain words. The ones that aren't, too.

There's no pitch at the end. If nothing is worth building, we'll say so. If something is, the next step is yours.

What you end up with

You use AI where it counts and ignore the rest. Your people spend their week on the work that needs them.