Salesforce Bets $5.66B on AI Agents. Which Customer Record Will They Trust?
the Braze deal shows why duplicate customer data now costs more
Most organizations believe customer AI succeeds or fails on the engagement tool: the right platform, the right channels, the right model deciding what to send and when.
The larger problem is the customer record underneath. This week, Salesforce agreed to acquire Braze for roughly $5.66 billion in enterprise value, tying Braze's real-time engagement and AI decisioning to Data 360, Customer 360 and Agentforce, according to the announcement. Salesforce framed the deal around agentic systems that can understand customers, decide and act in real time, Agile Brand Guide reported.
That puts a recurring question in front of every data team. When an agent acts on "this customer," which record is it reading: the CRM contact, the billing account, the support profile or the engagement platform's own copy?
And it gets worse as stacks grow. Every acquisition, every new engagement tool and every new agent adds another system that keeps its own version of the same person.
Why Does Every Customer Exist Three Times?
Each copy was created for a good reason. Sales needed a contact. Finance needed a billing account. Marketing needed a profile that updates every time someone opens the app.
Nobody designed the duplicates. They accumulate, one integration at a time.
Picture a retention agent asked to send a win-back offer to customers who cancelled last month. It finds the cancellation in billing, an active profile in the engagement platform and a recent support ticket in the CRM under a slightly different email address.
It sends the offer. To a customer who renewed yesterday on a different account.
This is not a marketing problem. It is a master data problem.
The Cost Isn't One Bad Message. It's Agents Repeating It at Scale
A person who spots a duplicate fixes it and moves on. An agent acting in real time repeats the same mistake across every channel, every hour, before anyone reviews it.
Most organizations already know this is hard. In Teradata's survey of 1,000 senior technology and data leaders, 78% said they find it challenging to unify data and knowledge across business functions.
The stakes are rising with it. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value or inadequate risk controls.
Before an agent can safely act on a customer in real time, someone has to:
Find every system that holds a version of each customer
Match records that belong to the same person, even when names, emails or IDs differ
Decide which source wins for each field when two of them disagree
Assign an owner to the customer entity, along with its consent and contact rules
Track which agents, campaigns and reports depend on each record
None of these tasks are particularly difficult. The problem is that they are usually done once, for one migration or one campaign, while new systems keep creating new copies every week.
How to Give Your AI Agents One Trusted Customer in 30 Days?
You do not need to replace your CRM or pause your engagement roadmap to do this.
Within the first month, focus on the customers your agents will act on first: active accounts, recent cancellations and anyone with a consent or contact preference on file.
Start by discovering every system that holds customer records and resolving duplicate and conflicting records into one trusted version of each person. Then attach an owner, and the contact and consent policies, to that record.
Next, expose this trusted record through the same connectors your agents and engagement tools already use, so an agent checks it before it acts.
DataManagement.AI accelerates this process by unifying master data, governance and operational trust into one continuously maintained system. It resolves duplicate and conflicting records, automatically discovers metadata and maps lineage across the modern data stack, and ownership and policy travel with every entity, so every agent and every channel reads the same customer.
Instead of cleaning up after a campaign misfires, your teams can let agents act in real time, knowing every system is reading the same person.
Customer AI Is Actually a Master Data Problem
Deals like this will keep coming. Engagement platforms, CRMs and agents are converging, and each one promises to know your customer better.
But every one of them can only act on the customer record your organization has already agreed is the right one. If that agreement does not exist, the agent simply picks.
Organizations that scale customer AI successfully do not just buy faster engagement. They make sure every system, and every agent, knows exactly who the customer is.
Warm regards,
Shen and Team