Five things you will soon simply ask your CRM

Five tasks from an ordinary working week, and how Claudeforce changes them.

Between two conversations with donors sits a lot of lookup work. Putting a list together, going through a giving history, explaining a campaign figure to a colleague who never opens the CRM. That is exactly the work Claudeforce touches.

Salesforce and Anthropic announced Claudeforce on 26 August 2026. According to that press release, Salesforce in Claude, the part that lets you query your CRM inside a conversation, is running with selected pilot customers, with an open beta planned for September 2026. It ships with thirty-seven prebuilt skills, all written for sales: meeting prep, deal health review, pipeline review. Skills for other domains will only start rolling out in late 2026.

The movement in the other direction has been under way longer. Claude serves as the reasoning model for the Atlas Reasoning Engine, runs by default behind Agentforce Vibes and Agentforce Coworker, and is available in Agent Builder. In Slack, Claude has been the default model since the announcement.

One technical detail matters more than the rest for our sector. Processing runs through Amazon Bedrock inside the Salesforce Trust Boundary, so your data stays within that perimeter. For an organisation working with donor data, that changes the conversation with your DPO.



The five examples below are about the work this brings within reach. Part of it can be tested in Agentforce today, the rest depends on when the beta reaches your organisation.

1. Pulling a list without building a report

You want to know who gave at least a hundred euros last year and nothing this year, in the province of Antwerp, leaving out everyone who already received a tax certificate for another project.

Today that means picking a report type, setting filters, adding fields, and starting over twice because something is missing. Or asking the colleague who knows how, and she has other plans this week.

Soon you will ask that in one sentence and get the list.

Do one thing alongside it: ask for the count as well, and hold it against a number you already know. If you had roughly four hundred donors above a hundred euros last year and the answer says two thousand, your question was understood differently than you meant it.

2. A first draft of a thank-you letter or a call note

A gift of five thousand euros comes in from someone who has been giving for nine years. You want to thank them properly, and you do not want it to read like a form letter.

What you do now: open the record, go through the giving history, check when you last spoke to this person, and only then start writing.

The lookup work can disappear. You ask for a first draft based on what sits in the record, with the correct facts already in it.

The tone stays your own work. And check the amounts and the years before you send, because a wrong figure in a thank-you letter does more damage than a late one.

3. Having a number explained to someone who never opens the CRM

Your director asks why the May campaign brought in twelve percent less than last year. You export some data, build a table in Excel and write a mail around it. An hour and a half.

The gain here sits mostly with him. He will ask the question himself, in plain language, and get an explanation with the underlying figures attached. Within his own access rights, so he sees nothing beyond what he was already allowed to see.

That saves you the hour and a half and saves him the wait.

4. Preparing a conversation with a donor

At eleven you have a long-standing donor on the phone. You want to know what she gave, when, on which campaign, whether she attended the May visit, whether there was ever a complaint, and who had the last contact.

Today that is five tabs and ten minutes. Per call.

This is also the part that works best right now, because the prebuilt skills shipping with Claudeforce are built around exactly this kind of preparation. They were designed for sales conversations, but the lookup work is identical.

5. An inventory of the mess in your data

You know there are duplicate records in your system. You also know that sorting them out has been on the list for three years without ever reaching the top.

Ask which records closely resemble each other, which fields stay systematically empty, which campaigns have no source recorded. That inventory is ready in minutes.

The cleanup itself stays manual work. But the difference between knowing something is off and knowing exactly what, is the difference between an intention and a task.

Where it still falls short

Two things worth knowing up front.

The bundled skills reason in sales logic. A bequest with a horizon measured in years, or a monthly donor you should be careful not to contact too often, does not fit a model built on win probability and close dates. Skills for other domains have been announced for later.

And the answer always sounds equally confident, including when it is wrong. A misunderstood question does not produce an error message, it produces a tidy list. So anything you use externally, in a report to a funder or in a board summary, you verify.

What you need to get started

A Salesforce environment, the connection your administrator sets up once, and data that is reasonably in order. That last point decides more than the rest: on a system full of duplicates and empty fields you get fluent answers on a shaky basis.

At Fundflow, a product of Impactella, that structure is part of the CRM itself. Gifts, tax certificates, mandates and campaigns each exist as their own component, with the Belgian rules built in. Whatever AI layer arrives on top will be talking to that model.

Want to try the five examples above on your own data? Let us know through fundflow.be and we will set it up with you.

Source for the announcement, the skills and the availability dates: the joint Salesforce and Anthropic press release of 26 August 2026 ( salesforce.com, also on investor.salesforce.com).