Claude Community Sydney: three lessons on AI for sales

On Tuesday night we spoke at the Claude Community Sydney meetup run by the Claude Community Australia. The room was full of developers and founders building on Anthropic's Claude. Our talk was about the one thing most of that room had never done: sales.

We covered how AI reached sales. In 2023 it was a chat window. In 2025 Claude Code made it real for developers. In 2026 Claude Cowork is the first time people in sales were properly included, with Claude working across the files, inbox, calendar and CRM they already use.

We took three lessons from the evening. All of them matter if you run a sales team or if you are thinking of getting a Claude workshop for yours.

Everyone had used Claude. Nobody had run a forecast call.

Early on we asked for a show of hands. Who has used Claude Cowork? Every hand went up. Keep your hand up if you have run a sales forecast call? They all went down.

That is the gap in a clear picture. The people building AI tools are amazing at building them and most have never sat in the meeting those tools are meant to fix. They have not spent a Monday morning chasing reps for deal updates or watched a forecast fall short because the notes from the last call were never written down. So a lot of AI for sales gets built for the problem that is easiest to see from the outside: write the email faster. The real problem, we believe, is usually somewhere else.

We showed the room a Monday from my time as a sales director. Nearly six hours of it went on the forecast. Reading through call notes, asking each rep where their deals were at, and putting it all together. None of it was moving the needle. We fixed it in four steps. A note-taker on every call. The transcript going straight into the CRM. Questions for each rep drafted by Wednesday. A forecast that builds itself on Friday. With those four in place, the forecast takes about thirty minutes of a Monday, and the rest of the morning goes back to a customer call and a prospecting block that did not even exist before.

It is not perfect. The daily digest we run every morning kept nudging us about a deal that had gone quiet because the conversation had moved to WhatsApp and never made it into the CRM. You build it, you use it, you fix what breaks.

The best demo of the night had no buyer

One founder showed us a voice agent speaking in the voice of a famous actor. It was cool and we loved it. But then the founder asked us a simple question: how do I sell this?

Building before validating is a founder's mistake and sales teams make their own version of it with AI all the time. They buy the tool before they have mapped the process. Someone sees a fancy demo, a licence gets signed and a month later nobody can say which part of the sales process it was meant to improve.

The order we keep coming back to is process first, people second, technology third. Write down how your team actually sells. Make sure the people running that process understand what changes for them. Only then pick the tool. Claude is powerful. But what it does not have is your context: how your team sells, where your deals stand, what a good next step looks like for you. Mapping the process is how you give it that. A year-old model wired into your CRM, your call notes and your definition of a good next step will beat the newest model with no context.

The data question beat every demo

The demos got the applause. The real conversations happened afterwards and one question kept coming up: where does my customer data go? Not what can it do, but where does it go, who can see it and what happens when a rep connects the pipeline to it.

That is the right question and sales leaders should ask it too. The way we set it up at Onwise, every system stops at a draft inside the client's own accounts, and a person decides what gets sent. The full answer, from which systems connect to what stays inside your accounts, needs a more in depth look. A full article written by Thibault is coming soon

Why the room matters to sales leaders

Adoption is no longer the problem. Salesforce's State of Sales research finds 87% of sales organisations already use some form of AI. Yet in Bain's executive survey, only 23% of executives can tie their AI initiatives to new revenue or lower costs. The tools are in. The results have not followed.

The room in Sydney explains part of why. Many AI tools are built by people who have never sold, then bought by teams that have never mapped how they sell. The fix is not a better model. It is putting the process first and letting AI take on the work around the deal, so the rep can spend more time selling.

What to look for in a Claude workshop for sales

If you are looking at Claude workshops or AI sales training for your team, ask these questions: does it start from how your team sells, or from the tool? Does it use your own deals, CRM and call notes, or made-up examples? And does every seller leave with something running on Monday, rather than a list of prompts?

That is how we built our one-day AI sales training: hands-on, on your live deals, with the process mapped before any tool is switched on. If you would rather try Claude on your own first, our free Claude Cowork course for sales takes about an hour and needs no technical background.

Thank you to Claude Community Australia for having us and to everyone who stayed for the questions afterwards.