The Meeting Preparation Agent: How One Sales Team Got Its Fridays Back

Most writing about AI agents stays at the level of the category. This is one agent, in detail — what it does, how it was built, and what actually changed.

It runs daily in the advertising sales team of one of Australia's largest media organisations.

The job it was given

The problem was described by a salesperson, in one sentence, and it is worth quoting close to how it was put:

"I walk into client meetings without knowing what we last promised them, what has changed since, or what they announced this week."

The honest preparation for a client meeting is around half an hour: the CRM for the account history, email for what was last agreed, internal chat for anything the team has flagged, and the trade press for whatever the client has announced. Multiply that by five meetings a day and preparation becomes a part-time job.

So mostly it did not happen. People went in cold, relied on memory, and occasionally got caught out on something the client had announced publicly a week earlier.

What it produces

Each morning, a brief lands covering every meeting in the next 48 hours. It draws from the calendar, Salesforce, email, internal chat and the industry press.

The structure matters more than the sources. It covers what was agreed last time, what has changed in the account since, what the client has said publicly, who is attending and what they care about, and what is currently open.

It is a briefing document, not an answer. Nobody is being told what to say.

What it got wrong at first

The first version was too comprehensive. It surfaced everything it could find about an account, which meant a brief nobody read on the way to a meeting.

The fix was not technical. It was deciding what a person can absorb in the four minutes before walking in, and cutting hard to that.

The second problem was recency. Early versions treated an email from two years ago as equal to one from Tuesday. In an account relationship, the old thing is occasionally the important thing — but usually it is noise, and a brief that cannot tell the difference gets ignored.

Both problems came from the same source: the team using it every day, saying what was useless. That feedback loop is the reason it works now, and it is the part that cannot be skipped.

Why it was possible

Nothing here requires an exotic model. What it requires is reach.

The brief is only useful because the agent can see the CRM, email, chat and the press at once. Any one of those alone produces something a person could have got themselves. Together they produce something nobody had time to assemble.

This is the reason the job survived every previous attempt at automation. No workflow tool spanned all of those systems. The agent is not cleverer than what came before — it can just reach further.

And because the underlying index respects permissions, a salesperson's brief contains only what that salesperson could already have found. That property is what made it deployable across a team rather than a pilot for one person.

What actually changed

The obvious framing is time saved: half an hour per meeting, several meetings a day.

That is real, but it is not what the team talks about. What they mention is that conversations now start from what was actually said last time, rather than from whoever's memory is better. The client does not have to re-explain their own situation. Nobody gets caught not knowing about an announcement.

The measure of it is simple: it would be noticed immediately if it stopped. Which is the only durable test of whether an agent has made it into production.

Whether it would work for you

The conditions this needed were unremarkable. A team with many external meetings. A CRM in reasonable order. Systems that could be indexed. And one person willing to say what was wrong with it every day for the first fortnight.

The last condition is the one most often missing, and it is free.