Zoomin implementation cost: Data 360 credits for unstructured knowledge

Every service organisation has the same problem in a different shape. The answer a customer needs exists somewhere: in a product manual on SharePoint, a release note on the documentation site, a training video, a support article written three years ago by someone who has since left. Zoomin's business was gathering that scattered content into one place people and bots could search. Grounding an agent in it is the modern version of the same job.

The completion date is on record: "Salesforce completed its acquisition of Zoomin on November 1, 2024." A few weeks later Salesforce explained where it went: "To further augment unstructured data integration and fuel next-generation enterprise knowledge for Agentforce, Salesforce recently acquired Zoomin, a leading data management provider whose technology now augments unstructured content capabilities in Data Cloud." Data Cloud has since been renamed Data 360. Salesforce's price lists in October 2026 carry no Zoomin-branded product, so this guide prices the Data 360 and Agentforce consumption that now does the work, and the implementation around it.

#What you are actually paying for

Unstructured knowledge on Salesforce is paid for at three points.

The first is getting content in. Data 360's pricing page is generous here: "Connect and centralize data from any source with free batch data ingestion and zero-copy integrations." Connecting a SharePoint library or a documentation site does not itself draw credits in batch mode.

The second is making content usable. Documents have to be processed into searchable pieces, and agents have to query them. Data 360's pricing FAQ is clear about how that is charged: "You still use Flex Credits for non-profile based use cases like querying data or unstructured data processing for AI agents." That sentence matters most for buyers who were offered profile pricing. A profile licence covers building unified customer profiles. It does not cover processing a manual or answering a question from it.

The third is the agent answering. Each time Agentforce answers from your knowledge, it runs actions, and those are metered separately: "Agentforce actions are 20 Flex Credits, while Agentforce Voice actions are 30 Flex Credits."

#The price list, as at October 2026

These are list prices from the Data 360 pricing page and the Agentforce pricing page, read on 11 October 2026, in US dollars unless stated. Salesforce marks both as information only and subject to change.

Two practical points follow. Data 360 and Agentforce Flex Credits are priced the same, and the Data 360 page says credits can be transferred between Data 360 and Agentforce. That makes a single credit pool possible, which simplifies budgeting but also means one runaway use case can drain the other. Second, Salesforce publishes a Data Services rate card that sets how many credits each usage type consumes. Read the unstructured processing and query lines before you forecast, because those multipliers, not the headline credit price, decide your bill.

The Service Cloud price comparison also lists an item called Enterprise Knowledge as available for purchase, with no price shown. If a proposal includes it, ask for the price and what it adds beyond Data 360.

#Where the content comes from

Salesforce has described the sources this is meant for. Its post on Data Cloud and Agentforce talks about grounding agents in content from Google Drive, Microsoft SharePoint, YouTube video, websites and content management systems. The Data 360 connectivity page adds: "Data 360 unpacks it with 15+ pre-built connectors, turning any format into trusted context, ready to activate anywhere."

That list is the starting point for scope. Every source you add is a connector to configure, a permission model to respect, a refresh schedule to set, and a body of content someone must own. A pilot with one well-kept documentation site is a small project. Ten SharePoint sites with inherited permissions and twelve years of drafts is not.

#What the implementation is made of

Content inventory. List every source, its owner, its size, how often it changes and who is allowed to read it. Mark what is authoritative and what is archive. Most organisations find that a large part of their stored content should never reach an agent.

Connector set-up. Configure the Data 360 connectors for each source, with service accounts, scopes and refresh schedules. This is usually quick for supported sources and slower for a custom CMS that needs an API or a file export.

Permissions mapping. Decide whether the agent may only use public content, or content scoped to the customer or the rep. If internal documents carry access controls, the design has to respect them so an agent never quotes a document the reader could not open. This is the work package most often underestimated.

Processing and retrieval design. Configure how documents are split, indexed and retrieved, and set up the retrievers your agents will call. Salesforce has promoted no-code retrievers for this, which reduces build effort but not design effort.

Agent grounding. Connect the retrievers to the relevant Agentforce subagents, write instructions about when to cite a source and when to hand off, and decide whether answers show links back to the document.

Answer evaluation. Build a test set of real questions with known correct answers from your support history. Run it before go-live and after every major content change. Without this, nobody can say whether the agent improved.

Content clean-up and governance. Retire duplicates, fix contradictions between sources, and assign owners. An agent grounded in two versions of a returns policy will quote both.

#What drives effort and running cost

Source count and permission complexity drive the build. Content churn drives the running cost: a documentation site rewritten every release means reprocessing, and reprocessing draws credits. Question volume drives query and action credits.

Audience matters too. A customer-facing agent answering from public documentation needs simple permissions but high accuracy, because a wrong answer reaches a customer directly. An employee-facing assistant for service reps can draw on internal material, but the permission work grows, and reps will quickly stop trusting a tool that surfaces a draft or an outdated procedure. The two audiences often deserve separate phases, separate test sets and separate budgets, even though they share the same Data 360 connectors underneath.

Versioning is the quiet cost driver for product companies. If you support several releases of the same product at once, each answer has to come from the right version of the manual. That means tagging content by version, asking the customer which version they run, and testing that the agent never mixes them. It is design work that a generic quote rarely includes.

A useful habit is to estimate credits per answered question, combining the query cost in Data 360 with the action cost in Agentforce, and to track it monthly. When that number climbs, the usual causes are agents retrieving too broadly, sources full of near-duplicates, or instructions that let the agent search several times before answering.

#Salesforce Knowledge articles or documents left where they are

One design choice changes the cost profile more than any other: whether you rewrite key content as Salesforce Knowledge articles, or leave it in its source system and let Data 360 process it.

Rewriting into Knowledge costs content effort up front. Each article gets an owner, a review date, a data category and an audience, and service reps can use the same article the agent cites. Answers tend to be more consistent, because the article was written to answer a question rather than to describe a product.

Leaving content where it lives costs less to start and more to run. Nobody rewrites anything, but the agent has to find the answer inside long documents written for another purpose, and every change upstream means reprocessing. Quality depends on content owners in other departments who may not know an agent is reading their work.

Most organisations end up with both: a few hundred well-kept articles for the questions customers ask most, and processed documents for the long tail. Ask bidders to recommend the split from your actual question data, and to price the article writing separately from the processing set-up.

#The data migration inside a Zoomin replacement

If you ran Zoomin's knowledge portal, a few assets are worth moving deliberately rather than leaving behind.

Search logs come first. The questions people typed into the old portal, and the ones that returned nothing, are the best possible test set for a new agent. Export them before the contract ends.

Then the content map: which sources were connected, how they were tagged, and which products or versions each document applied to. Rebuilding that taxonomy from scratch is slow, and the agent needs the same distinctions to avoid quoting the wrong version of a manual.

Then the public links. If customers and search engines reach your documentation through portal URLs, plan redirects so those links do not break when the portal goes. That is web work, not Salesforce work, and it is easy to leave off both quotes.

#Getting comparable quotes

Give every bidder the content inventory and ask them to return:

• a price per content source, covering connector, permissions and processing design
• the evaluation method and the number of test questions they will build with you
• their estimate of monthly processing and query volume, so you can price it against the rate card yourself
• how they will handle permission-restricted content, in writing
• what content clean-up they expect you to do, and when
• who owns the refresh schedule after go-live

A partner who quotes only "Data 360 set-up" for unstructured knowledge has priced the easy part.

#Costs that are usually left out

• Reprocessing credits when large sources change.
• Agentforce action credits for every answer, on a separate meter from Data 360 unless you pool them.
• Content owners' time to retire, merge and rewrite documents.
• Sandbox testing: check whether your Data 360 contract includes a sandbox environment for testing against realistic content.
• Security review of service accounts that read internal repositories.
• Ongoing evaluation, which is a monthly task, not a launch task.
• Licence overlap with an existing knowledge search tool during transition.

#A sensible first phase

Start with one source that is already well kept, one audience, and one channel. A public documentation site answering customer questions through a web agent is a common first step, because permissions are simple and the answers are easy to check. Measure answer accuracy against your test set, credits per answered question, and handoff rate. Only then add internal sources with access controls, which is where most of the cost and risk lives.

Agree the exit criteria for that first phase before it starts. A target accuracy on the test set, a ceiling on credits per answered question, and a maximum handoff rate give the steering group a clear reason to fund phase two, or to fix content first. Without them, the pilot tends to drift into production on enthusiasm alone.

#Where the other Zoomin questions live

How Zoomin's content model maps to Agentforce knowledge is on Zoomin and Agentforce knowledge. The acquisition timeline is on what happened to Zoomin. Licence questions, including existing Zoomin contracts, are on Zoomin licence cost. Connector and API detail is on integrating Zoomin with Salesforce. Whether to move off a working knowledge portal is on should I migrate from Zoomin, and how to test a partner is on hire a Zoomin Salesforce partner.

Our Data 360 team can run the content inventory and credit forecast before any build is priced, so the first phase is sized on your real sources.