AI assistant

Principles⁠

The assistant drafts, you decide⁠

Nothing reaches a client without a person approving it. An assistant that proposes is useful. An assistant that sends is a liability.

Draft against real data⁠

A useful first version knows your packages, your prices and this client's history. Generic text is faster to reject than to edit.

Everything lands in one queue⁠

Drafted replies, proposed proposals and flagged anomalies belong in a single inbox with approve and reject. Work you have to go looking for is work that waits.

Keep the reasoning attached⁠

A draft you can see the source of is a draft you can trust. Keep the conversation that produced it next to the result.

Automate the follow-up, not the relationship⁠

Chasing an unanswered enquiry is a scheduling problem. Deciding what to say to a client who is hesitating is not.

Learn from the outcome⁠

Whether a draft was approved, edited or thrown away is the only real signal about whether it was any good. Record it.

Practices⁠

Ask about any event⁠

Ask in plain language what is booked, what is outstanding, who is working and what changed. The answer comes from your own data, not from a general model.

Draft the proposal from the enquiry⁠

The assistant reads the enquiry and builds a first proposal from your block library, with quantities and prices already filled in. You edit and send.

Draft the run sheet from the event⁠

A first briefing written from the timings, numbers and packages on the event, ready to correct rather than to compose.

Work the inbox⁠

Proposed actions arrive in one queue. Approve the good ones, reject the rest, and discard leads that are not real. Two keys, no context switching.

Let it read the replies⁠

Incoming email is matched to the right event and the relevant thread, so a reply about Thursday lands on Thursday's event instead of in a shared mailbox.

Let it follow up on its own⁠

Enquiries and proposals with no answer are chased on schedule, drafted for you, so the pipeline keeps moving while you are running events.

Let it flag the unusual⁠

A scan looks for margins, costs and payment patterns that do not fit and raises them, so a problem surfaces in days rather than at year end.

Keep the draft history⁠

Each proposal and briefing keeps the conversation that produced it, so a colleague picking it up can see what was asked and why.

Check the outcomes⁠

Approved, edited and rejected drafts are tracked, which tells you where the assistant is earning its place and where it is not.