If you were at the Salesforce Agentforce World Tour in London, or following the conversation that came out of it, you probably noticed that for most organisations the conversation had moved past the roadmap. The focus now is on what’s being deployed, what’s already in place, and what comes next. In other words, the roadmap conversation is mostly over and the implementation conversation has now begun.
As a result, the question most firms are sitting with right now — and what Agentforce made harder to ignore — is whether the data fuelling agentic CRM is actually ready to support autonomous action.
For most organisations, if you’re being honest about it, the answer is probably not yet.
Table of contents
From “Copilot” (advice) to “agent” (action)
It’s worth being precise about what agentic CRM means, because the evolution from copilot to agent is more significant than it might first appear.
Copilot-style AI, which has been the dominant paradigm for the last few years, is fundamentally assistive. It drafts the email, surfaces the summary, suggests the next step but you still review the output and decide what to do with it. The AI is doing the work, but a human is still in the loop on every decision.
Agentic AI removes that checkpoint. An agent doesn’t wait for you to review the draft and click send. Instead, it takes the action, updates the record, routes the follow-up, and moves to the next task without anyone stepping in to approve it. For most professionals, the CRM has always been a system you go to when you need something from it. Agentic AI changes that relationship in that the system acts before you ask.
That’s a fundamentally different relationship with enterprise software. And it’s part of a broader evolution of enterprise technology and user experiences that’s been building for a while. The move toward what’s called ‘headless CRM,’ where the CRM functions less as a central destination and more as an intelligence layer feeding into wherever your teams work, sets up exactly this kind of agentic model. The agent isn’t pulling up a CRM screen. It’s drawing on a connected data layer that sits beneath your tech stack, and acting from it.
Some organisations are already there. For the rest, readiness for agentic CRM depends on what’s underneath. Those with a clean, connected data layer underneath their Salesforce instance can deploy agents now.
The data layer is the new bottleneck
Salesforce has built a capable framework for autonomous execution. But a framework is only as effective as what it’s working from, and for most firms, what it’s working from is incomplete in ways that have been tolerable up to now.
CRM data in professional services has always been an imperfect record. In fact, 76% of CRM users and administrators said less than half of their organisation’s CRM data is accurate and complete. Contacts go stale while at the same time meeting notes are missing because the partner who ran the call never logged it. The account looks active in the system, but the actual engagement history lives in three different people’s inboxes and hasn’t been formally recorded in months.
This isn’t really a failure of discipline, but a reflection of how organisations operate. The individuals generating the most valuable relationship activity, partners, senior practitioners, fee earners, need to dedicate their time to client delivery and business development, not data entry. That gap between the firm’s interactions with its clients and what exists in its CRM is something professionals have learned to work around it.
And for a long time, that workaround was manageable with BD teams triangulating from other sources or partners carrying the relationship context that never made it into the system. Institutional knowledge covered what the CRM didn’t.
Agentic CRM removes that workaround. When an AI agent is taking autonomous action based on what it reads in your CRM, incomplete data doesn’t slow the agent down. Instead they shape the decision, invisibly, before anyone notices.
Agents need context, not just records
Consider what agents can and can’t read from a typical CRM record.
An agent can read that a contact is listed as Chief Executive Officer at a key account. What it can’t read from a standard CRM record is who at your firm has a meaningful relationship with that person or how recently they were in contact.
Without that context, an agent reading only the CRM record might identify a contact as lapsed — no recent activity logged, no outreach recorded — and automatically send a re-engagement email on behalf of your managing partner, regardless of whether that contact is in fact the partner’s closest client, and the relationship is active but happening in email and in person.
Acquiring a new client costs significantly more than retaining an existing one. When an agent sends an automated re-engagement touch to a senior partner’s closest client contact because the CRM shows no recent logged activity, the cost isn’t just an awkward conversation. It’s the slow erosion of trust in an organisation that the client had spent years deciding to rely on.
Firms consistently overestimate how ready their data actually is. Research shows that 43% of data leaders cite data readiness as the top barrier to AI, while simultaneously claiming their data is ready. When autonomous agents are making decisions before anyone reviews them, that gap between perceived and actual readiness is where the damage happens.
That’s why data foundation work is a prerequisite to agent deployment. Agents act on what they find. If what they find is incomplete, the actions they take will reflect that. Passively captured, automatically enriched relationship data is what makes autonomous action reliable. An agent with access to it knows who at the organisation last spoke to a contact, when, and in what context.
Conclusion: don’t build agents on bad data
Organisations who are best positioned to get the most out of agentic CRM are the ones whose agents are working from data that accurately reflects the firm’s real relationship activity.
That’s a different kind of readiness than most firms are currently focused on. The conversation around AI adoption in professional services has been dominated by questions about tools, platforms, and use cases. The harder question, and the more consequential one at this stage, is whether your underlying data is ready for autonomous action.
Automated data capture needs to be in place before agents go live, not something you plan to sort out afterwards. If relationship activity is still being logged manually, or not logged at all, your agents will inherit every gap in your data foundation. And the gap doesn’t disappear when you add an agentic layer on top of it. It compounds, because now it’s influencing autonomous decisions instead of producing an incomplete report for someone to review.
The Agentforce World Tour showed how quickly the conversation around agentic AI has moved from roadmap to deployment. It also underscored the importance of getting the data right before agents run; a decision that will determine whether your investment pays off.
Your agents will be exactly as capable as the information you give them. That’s a straightforward principle, but getting it right has significant implications for where your attention should be focused right now.
The gap between your firm’s real relationship activity and what lives in your CRM doesn’t disappear when you introduce agentic CRM. It compounds. Book a demo with our team to see what ready looks like.
BOOK A DEMO
