Only 34% of Salesforce customers have adopted Agentforce and the company has shed more than $200 billion in market value. The barrier is not belief in AI agents. It is the messy, disconnected data most businesses feed them.
The agent is not the hard part. The hard part is the data underneath it, and that is the part everyone skipped.
Salesforce went all in on Agentforce, its agentic AI product, and the market has not rewarded the bet. Only 34% of customers have adopted it, and the company has lost more than $200 billion in market value, with analysts saying the product is not ready for prime time. The interesting part is why.
It is not that businesses doubt AI agents can work. The barrier is underneath. Agents need clean, structured, connected data to make decisions, and most enterprises are sitting on fragmented CRM records, disconnected systems and inconsistent customer information. Point an agent at that and it makes confident decisions on bad inputs. Most deployments have stalled at proof-of-concept because the foundation was never there.
Salesforce itself has read the room, buying data integration firm Informatica and adding tools that pull customer data from external sources. In other words, the company selling the agents is now scrambling to fix the plumbing that makes them work.
Why it matters
Every Australian business being pitched an AI agent right now should hear this clearly. The tool is not the constraint. Your data is. If you cannot answer basic questions about your own customers, your own channels and your own numbers, an agent will not save you. It will automate the guessing. The businesses that get value from agentic AI are the ones who did the unglamorous work first, connecting their systems and cleaning their records, so the agent has something true to act on.
Share of Salesforce customers who have adopted Agentforce, despite the all-in push. Source: MarTech.
What to do about it
Fix the foundation before the agent. Connected, accurate customer data is the prerequisite, not a nice-to-have.
Audit your data quality honestly. Duplicate records, missing fields and disconnected tools are where agents fail quietly.
Start narrow. One clean use case with good data beats an enterprise rollout on a messy foundation.
Do not buy the pitch over the plumbing. If a vendor cannot tell you what data the agent needs to work, they are selling you the guessing.
The biggest name in the category just proved the point. The agent is easy to buy and hard to feed. Sort the data first.