Claude meets Salesforce: Why Claudeforce matters for your data

At the end of August, Anthropic and Salesforce deepened their partnership under the name “Claudeforce.” From now on, Claude works directly on your Salesforce data—with pre-built sales skills, in Agentforce, and even in Slack. Specifically, Claudeforce is launching with “Salesforce in Claude,” a plugin with 37 pre-built sales skills. This lets you evaluate revenue context, automate pipeline updates, and trigger actions directly from Claude—meeting preparation, deal health checks, or pipeline reviews, all without opening the Salesforce app. It’s currently running with pilot customers, and the open beta is announced for September.

Marc Benioff is bringing two market leaders together, the “#1 AI” and the “#1 CRM.” And honestly: it’s a clever, logical step. An agent working on your data makes a lot of sense. They say data is a company’s gold—but without the right treasure map, you won’t find the treasure. Huge, unused treasures lie dormant in the data lakes of most companies. This is exactly where an AI is strong: it can process this data and present it in the right format.

But: as soon as an AI tool accesses your CRM data that deeply, a series of questions arise that you shouldn’t skip. What exactly goes out? What is used for training—and what isn’t?

What Claudeforce actually is

Following the announcement on August 27, Claudeforce is launching with “Salesforce in Claude,” a plugin with 37 pre-built sales skills. This allows you to evaluate revenue context, automate pipeline updates, and trigger actions directly from Claude. This includes things like meeting preparation, deal health checks, and pipeline reviews, all without opening the Salesforce app. In addition, there is Claude in Agentforce and Claude in Slack. It is currently running with pilot customers, and the open beta is announced for September. This is the direction in which marketing and sales are moving, and it is fundamentally good.

The question that remains: What happens to your data?

As soon as an AI tool accesses your CRM data so deeply, however, a series of questions arise that shouldn’t be skipped. What exactly goes out? What is used for training and what isn’t? Is the data only used to complete the current task, or is something else happening? Do I have to be careful about what I write in the CRM in the future? And how do we control that customers aren’t suddenly contacted automatically or that sensitive deal information doesn’t pop up elsewhere?

To be fair: Anthropic has invested a lot here, with its own protection mechanisms and a permission concept to ensure customer data remains private. That is important. But the fundamental question remains with you: With a setup like this, Claude gets even more access to your data, and you should know exactly what that means.

A point from my own experience that is often underestimated: AI tools are smart and often find ways to get to data that you haven’t given them directly. A simple example: from a domain and a first and last name, the model independently constructs a possible email address and checks if it exists. This doesn’t happen out of malice, but it shows why control over access is not a detail, but the core.

The better way: the same idea, but your data stays with you

In the end, the choice is yours. You don’t have to choose between ‘agentic AI on my data’ and ‘data control.’ It depends on how the setup is built. At Aivie, your data is stored encrypted on Swiss servers. Encryption on Swiss servers is basically possible elsewhere too, because the difference lies deeper: we host the data on your behalf, with an independent open-source system instead of a shared cloud product. This is a path that more and more Swiss organizations are taking. Every customer has their own database that we operate for them. There is no exchange between customers, a private key is possible, and we can give you complete access and full transparency.

And when data goes to an AI, you can set very granularly what it is allowed to see and what it isn’t. For example, only statistics, but no personal data. For personal data, there is an additional layer that censors sensitive information like email addresses before anything even goes to the model. This way, you get the benefit of agentic automation—from evaluation to personalization at individual touchpoints—without sensitive data leaving the house in an uncontrolled manner.

The next level: your own marketing AI

For many, this setup is enough. But there are SMEs with increased data protection needs who want to go one step further. For them, we are currently working on our own marketing AI based on the same principle: a dedicated, fine-tuned model, hosted in Switzerland for the customer and operated by Aivie. Normally, these are extreme costs because such a model requires a lot of computing power. That’s exactly why we’re not building it in isolation, but together with SMEs to share the basis and make it affordable. Today, this is a pilot and not a finished product, but it is the direction we are going.

Conclusion: Data question first, then automation

Claudeforce shows: Agentic AI directly on company data is here, and it makes sense. The treasure in your data can be used. The only question is who keeps the treasure map and who ultimately has control over the data. Contact me and tell me about your challenges and wishes regarding AI and your data. My goal is to make the Swiss marketing AI better together with you.