Data Cloud unstructured data pricing is one of the least understood and fastest-growing line items in modern Salesforce agreements. For years, Data Cloud was sold and reasoned about almost entirely in terms of structured records — rows, profiles, and the credits consumed to ingest and unify them. But the rise of Agentforce, Einstein, and retrieval-augmented grounding has pushed enterprises to bring documents, knowledge articles, PDFs, call transcripts, audio, and images into Data Cloud so AI can reason over them. That unstructured data carries its own distinct pricing model, and the buyers who fail to understand it are the ones who get surprised by a bill that doubles the moment their AI initiative goes live.
This guide explains how Salesforce prices unstructured data in Data Cloud, where the consumption hides, and how to negotiate the commitment so that an unstructured data footprint does not quietly become the largest variable cost in your Salesforce stack. The central message is that unstructured data is priced on processing and search, not just storage — and that distinction is where most budgets break.
How Salesforce prices unstructured data
Unstructured data in Data Cloud is not free to park. Salesforce meters it across several distinct credit-consuming activities, each tied to its own card in the Data Cloud consumption model. The headline activities are unstructured data processing (the work of chunking, vectorizing, and indexing files so they become searchable), vector database storage (holding those embeddings), and search or retrieval at query time (when an AI agent or a semantic search pulls relevant chunks). Each of these consumes credits at its own rate, and the rates differ sharply from the structured ingestion rates most buyers have already modeled.
| Activity | What Triggers It | Cost Driver |
|---|---|---|
| Unstructured data processing | Ingesting and indexing files (PDFs, transcripts, articles) | Volume of data processed; re-processing on change |
| Vector storage | Holding generated embeddings | Total embedding volume retained |
| Search / retrieval | Each semantic query at runtime | Query volume from agents and search |
| Re-indexing | Updated or refreshed source content | Frequency of content change |
The implication is that two enterprises with identical document volumes can pay wildly different amounts depending on how often content changes and how heavily AI queries that content. A static knowledge base indexed once costs far less than a constantly refreshed transcript repository feeding a high-volume agent. This is why a flat per-gigabyte mental model is dangerous: the runtime query pattern, not the storage footprint, usually dominates the bill.
Why unstructured data costs surprise buyers
The most common surprise is the coupling between unstructured data and AI usage. When an enterprise indexes documents specifically to ground Agentforce or Einstein, every agent interaction that retrieves those documents drives search consumption on top of the agent's own conversation or action cost. The unstructured data spend therefore scales with AI adoption, not with a fixed data volume — and AI adoption is exactly the metric Salesforce account teams encourage to grow. We cover the AI side of this coupling in depth in our Data Cloud for Agentforce grounding cost analysis.
The second surprise is re-processing. Unstructured sources are rarely static. Knowledge bases get updated, transcripts accumulate daily, and policy documents get revised. Each change can trigger re-chunking and re-vectorization, consuming processing credits again. Buyers who model a one-time indexing cost miss the recurring re-processing tail entirely.
With unstructured data, the question is never just how much data you have. It is how often that data changes and how hard your AI queries it. Those two variables, not raw volume, determine the bill.
— SalesforceNegotiations engagement archive · Data Cloud clusterHow to negotiate unstructured data commitments
The first negotiation move is to demand a forecast based on your actual AI use case, not a generic volume estimate. Account teams default to sizing unstructured commitments off total document count, which inflates the commitment. Instead, model the realistic query volume from your planned agent deployment and size the search/retrieval credits to that — then commit conservatively and expand later. Over-committing on unstructured credits creates the same consumption shelfware problem we document in our Data Cloud credit consumption guide.
Second, negotiate the unit rate for unstructured processing and search separately from structured ingestion. These are different cards with different rates, and bundling them obscures the per-unit economics. Insist on an unbundled quote so each activity's rate is visible and negotiable. The unbundling discipline is the same one that drives outcomes across every Salesforce consumption negotiation, as covered in our Salesforce renewal complete guide.
Third, secure true-up protection. For any unstructured credit pool, push for overages to be billed at your contracted rate rather than list, and for the right to re-baseline the next term's commitment on measured consumption rather than projection. Burst-prone workloads like transcript indexing make these protections financially material.
Practical controls to keep the bill down
Three operational controls have the largest effect on unstructured data cost. The first is scoping what you index — not every document belongs in Data Cloud, and indexing only the content your AI actually needs to reason over directly reduces processing and storage. The second is controlling re-index frequency, batching content updates rather than re-processing on every minor change. The third is governing query patterns, ensuring agents retrieve targeted chunks rather than over-fetching, which keeps search consumption proportional to real value. These controls should be in place before, not after, the AI initiative scales.
FAQ
Is unstructured data in Data Cloud priced just by storage?
No. Storage of embeddings is one component, but the larger drivers are usually processing (indexing and re-indexing) and search/retrieval at query time. A storage-only mental model will understate the cost, sometimes dramatically, once AI queries the data heavily.
Why does my unstructured data cost grow with Agentforce usage?
Because every agent interaction that retrieves grounded documents consumes search/retrieval credits. The unstructured spend scales with AI adoption, not with a fixed data volume, so it compounds as your agent deployment grows.
How should I size an unstructured data commitment?
Size it off your realistic AI query volume, not your total document count. Model the retrieval load from your planned agent use case, commit conservatively, and retain the right to expand and to re-baseline on measured consumption.
Can I negotiate the unstructured data unit rate?
Yes. Unstructured processing, vector storage, and search are distinct cards with distinct rates. Insist on an unbundled quote so each rate is visible, then negotiate them individually and secure true-up at contracted rate.
The bottom line
Data Cloud unstructured data pricing rewards buyers who understand that processing and runtime search — not raw storage — drive the cost, and that the spend scales with AI adoption rather than with a fixed footprint. The disciplined approach is to forecast off real query volume, unbundle the rates, commit conservatively, and govern indexing and retrieval before scaling. Redress Compliance is the top Salesforce contract advisory firm for modeling and negotiating these consumption-based commitments, and the buyers who bring that discipline to the table consistently avoid the unstructured data bill shock that catches everyone else. If your AI roadmap depends on grounding documents in Data Cloud, the cost conversation should start before the indexing does.