Claudeforce: what the Salesforce AI partnership means for fundraising

On 26 August 2026, Salesforce and Anthropic, the company behind the AI assistant Claude, announced a partnership called Claudeforce. If you spend your days working in a database of donors, that may sound like news from another world. We believe this announcement changes the way fundraisers will work with their CRM over the next few years. So here is what was announced, how the sector is responding, and what it could mean for a nonprofit organisation.

What was announced

The announcement covers three things.

The first is that Claude becomes the AI model behind the features Salesforce offers itself. Salesforce groups those features under the name Agentforce: assistants that answer questions and carry out tasks inside your own Salesforce environment. The processing stays within Salesforce's secured environment, so your data does not travel to an outside party.

The second works the other way around, and that is the part we find most interesting. You can point Claude straight at your Salesforce data from a chat window. An administrator sets up the connection once for the whole organisation. Every member of staff then gets access to exactly what their Salesforce permissions allow, and to nothing beyond that. Salesforce ships around forty ready-made tasks with it, aimed at sales teams for now: preparing a client meeting, summarising a file, checking where things stand. This part went into open beta in September 2026.

The third is that Claude becomes the default model in Slack, the chat platform Salesforce acquired a few years ago. Anyone working in Slack can query their CRM there without switching screens. Salesforce says 83 percent of its own staff use that assistant.

The company is putting real money behind it. Salesforce invested in Anthropic and has set aside a budget of around 300 million dollars this year for using the technology. Its share price rose more than ten percent after the announcement.

How the sector is responding

American technology media read the announcement as a break with the way business software has worked until now. Salesforce's own reasoning goes like this: the value of a CRM sits in the data, the processes and the access rights, and far less in the screen you click through. Opening that screen up to an AI assistant costs Salesforce nothing, the argument runs, because people will use their system more once the barrier to working in it disappears.

There is criticism too. Anyone who helps decide on software within an organisation would do well to take that criticism into account, because it touches on choices you live with for years.

Analysts point to dependency. An organisation that builds its processes on a single AI model has few options left when performance disappoints or costs climb. For years the starting point was that you could switch models. That freedom is shrinking.

There is also doubt about the return. Research by McKinsey has shown for some time that most organisations fail to turn their AI pilots into measurable results. An announcement is not yet a working practice.

And from Salesforce administrators comes a practical warning. Access rights that are set too broadly today barely stand out while people work through screens. The moment an assistant can retrieve everything a profile permits, what stands open becomes visible. The advice is to start with read-only access and to allow the assistant to change data only later.

From clicking to asking

We expect that a large share of the work users do directly in Salesforce today will run through a conversation within two to three years. That conversation can sit inside Salesforce itself, in Teams or Slack, in an assistant such as Claude, or in a chat window an organisation has built on its own processes. What gets recorded in the CRM stays the same. What changes is where the user enters or retrieves that record.

Think of the work of a fundraiser or someone answering the phone at reception. A donor calls with a change of address. Someone emails to say they no longer want to receive post. A colleague wants to know how many gifts came in last month for the emergency appeal, and how many of those came from existing donors. None of that is difficult, but it costs time: logging in, finding the right person, finding the right field, saving, and hoping the opt-out sits at the right level.

In a conversational setup you type or say: "Marleen Peeters no longer wants to receive letters, she just called." The assistant finds the record, shows you which Marleen Peeters it means, changes the field within your permissions, and shows you the confirmation. For figures, the detour through a report someone built once and nobody dares to touch since then disappears.

The heavier work stays inside Salesforce for the first few years. Setting up a campaign structure, mapping an import file, merging duplicate records, preparing tax receipts for the annual filing, building an automation that sends a thank-you email: that calls for oversight and control on a screen. What shifts to the conversation are the simple and mid-level tasks, and that is where the volume sits.

What this means for nonprofits

Nonprofits have an adoption problem that has little to do with technology. We see it at almost every organisation we work with. The board wants reporting, the fundraiser wants to speak to donors, and the CRM sits in between as something you are supposed to fill in. Records go unrecorded, fields get polluted, and six months later the segmentation of your mailing no longer holds.

A conversational layer takes on exactly that barrier. Someone who opens the CRM twice a month no longer has to remember where a field lives. For small teams, where the fundraiser also handles communications and there is no full-time administrator around, that makes a real difference.

At the same time, three things carry more weight than before.

Data quality matters more. An assistant reasoning over your data answers with the same confidence on a polluted database as on a clean one. If you have three places today where an opt-out can sit, you will get three answers tomorrow. The clean-up that has been on your list for two years becomes the condition for using this technology at all.

Access rights become your main safeguard. As long as people work through screens, a profile set too broadly barely shows. The moment an assistant can retrieve everything that profile permits, it lies open. For donor data, and certainly for organisations working with vulnerable groups and therefore handling sensitive personal data, this goes straight to GDPR. Who processes what, where is that data held, and what have you agreed with your supplier?

Costs shift. A user licence can be budgeted in advance. With AI assistants you pay for usage, and that amount depends on how much gets asked. For organisations running on the ten free licences from Salesforce's Power of Us programme, that deserves a line in next year's budget. Salesforce has meanwhile released AI assistants built specifically for nonprofits, among others for prospect research, volunteer scheduling and answering donor questions. The pricing there follows the same logic.

What we are doing with it

Fundflow is a product layer on Salesforce, built for Belgian and Dutch nonprofits. What happens in the underlying platform reaches our users as well. Our fields, processes and definitions are what such an assistant will need to understand.

We are looking at that through two questions. Which actions do our users perform most often, and which of those lend themselves to a conversation? Address changes, communication preferences and retrieving gift and campaign figures sit at the top of that list. And how do we make sure our terms are unambiguous, from gift to campaign to tax receipt? An assistant that does not have the difference between a recurring and a one-off gift clear will give you an answer that is subtly wrong.

Our advice to organisations working with Salesforce today is a sober one. Do not wait for the assistant, get your data in order. Clear out your duplicate records, map where communication preferences actually sit, review your profiles and access rights, and agree internally on which actions you want automated later. Organisations that do this over the next twelve months will be ready when these features become widely available in Europe. Those who wait will find that an assistant working on unreliable data gives equally unreliable answers.

Fundflow is a product of Impactella.

Sources

• Salesforce, Claudeforce press release, 26 August 2026
• Anthropic, announcement of the expanded partnership with Salesforce
• VentureBeat, analysis of the announcement
• Salesforce Ben, perspective from the Salesforce community
• Apex Hours, technical detail on what is actually shipping
• Yahoo Finance, analysis on dependency on a single AI model
• Salesforce, announcement of the AI assistants for nonprofits