The voice assistant

Somebody with wet hands, a phone in a truck mount and four minutes between jobs is not going to fill in a form. They will say what happened.

Press the assistant and speak. “Record 22 miles for the Henderson job.” “Start a proposal for Jim Bob for a sink repair.” “What’s John Smith’s number?”

How it works

The product writes down what you said, works out which action you meant, finds the records your words named, and fills in the form.

For anything that writes, it shows you the filled-in form and waits. Nothing is saved because you said it. You look at what it understood, correct it if it got something wrong, and press save yourself.

Nothing you say is wasted

If it cannot work out an action from what you said, the words still go somewhere: onto a job as a note, or copied out for you to paste.

That is a deliberate rule. An assistant that answers sorry, I didn’t understand and throws away thirty seconds of a person describing a leak is an assistant nobody uses twice.

What it will and will not do

Four rules shape the whole thing, and they are worth knowing because they explain its limits:

• It never sees an identifier. It hears “Charlie” and the server does the searching. It cannot name a record by its internal id, so it cannot act on one by accident.
• It never writes. It proposes an action; a separate step, triggered by you, carries it out.
• Nothing is saved without you seeing it.
• The catalog is small and curated. It can do a specific list of things, and it will tell you when what you asked for is not on that list rather than doing something approximate.

Nothing that sends, charges or deletes is available to it. Those stay actions a person takes deliberately, on a screen.

What it costs

The assistant needs Professional, because transcribing speech and interpreting it spends against your monthly AI allowance.

The remaining allowance is shown before you use it, and asking spends it whether or not you keep what comes back.