CONTEXTWhat it actually solves
Strip the marketing and Data Cloud is two things bolted together. It is a customer data platform that ingests structured and semi-structured data from many sources, resolves identities, and produces unified profiles. And it is an activation layer that lets those profiles be queried in real time from Sales Cloud, Service Cloud, Marketing Cloud, Einstein, and Agentforce. The first half has existed in various forms for a decade. The second half, activation inside the tools people already work in, is what makes Data Cloud genuinely different. A representative sees the unified profile without leaving the case, and that removes a friction a data lake never could.
01The unified service profile
When a customer calls, the agent should already know who is calling, what they bought, how they have been using it, what their open tickets are, and whether they are at risk of leaving. Most service organisations hold all of this data. It lives in five systems and surfaces across three browser tabs. Data Cloud's job is to assemble the picture and place it inside the Service Cloud case record. Ingest accounts, invoices, usage telemetry, and case history; resolve identity by email and account; surface the profile as a Lightning component on the case layout. At one client this single build cut average handle time by twenty-eight percent, because agents stopped saying they would check and started saying they could see it.
Real-time segmentation
Traditional segmentation runs overnight, by which time the behaviour has moved on. Streaming ingestion lets you define a segment such as "viewed pricing in the last hour and is on a paid tier above Pro" and have it update as events arrive. Outreach fires within minutes of the qualifying event. Conversion on triggered campaigns typically runs two to four times their batch equivalents.
02 / 03Scores written back, and agent context
Einstein Studio trains models on Data Cloud objects without exporting the data, and the output, a churn score, a propensity, a next-best-action, is written back to the CRM record where sales and service already work. The reason it succeeds where older predictive efforts failed is that the model sees the full unified profile, not just the handful of CRM fields. The same unified profile becomes the first thing an Agentforce agent retrieves: with it, a service agent in the console can reason over billing history and recent behaviour rather than the narrow slice a single cloud holds.
The predictive score is not the win. The win is the score being visible at the moment of decision, inside the tool the person is already using.
04 / 05Data products, both directions
The teams that get the most from Data Cloud treat it as a two-way street. Pull data in, resolve it, and publish clean, deduplicated, identity-resolved datasets back out to the warehouse, the data science team, and the BI tools. The data engineering team stops fighting the Salesforce admin team because they finally share a substrate. This is the least discussed pattern and one of the most durable.
| Workload | Data Cloud | Warehouse |
|---|---|---|
| Identity resolution across sources | Strong | Weak |
| Real-time activation in Salesforce | Strong | Not native |
| Streaming segmentation | Strong | Batch only |
| Long analytical joins, many tables | Costly | Built for it |
| Storing every event forever | Expensive | Cheap |
ANTI-PATTERNReplacing the warehouse
Once a quarter someone asks whether Data Cloud can replace their Snowflake. The honest answer is no, not yet, and probably not ever. Data Cloud is optimised for customer-centric data, people, accounts, and their interactions. It is not a general analytical warehouse, long multi-table joins are slower and dearer than they should be, and storing every event from every system because you might query it later runs up a six-figure bill quickly. Keep the warehouse, use Data Cloud for the slice that needs activation inside Salesforce, and let the two talk through zero-copy or scheduled sync.
The two biggest cost surprises we see are high-cardinality streaming events such as every keystroke or scroll, and full-table calculated insights running every fifteen minutes. Audit both quarterly, and turn off anything no activation actually reads.
STARTWhere to begin
Pick one of the five. The unified service profile is usually the safest first build, because the value is visible to a single team and the plumbing teaches you the platform. Once it has run reliably for a quarter you have earned the right to attempt the harder patterns. What you must not do is licence Data Cloud, try to ingest everything, and then hunt for a use case. We have watched six-month projects produce nothing that way. Name the use case first, and let it pull only the data it actually needs.