A practical guide to Agentforce deep research

Disclosure: This article is published by eesel AI, a competitor of Salesforce Agentforce. We encourage you to read Salesforce's own materials for their perspective.

The idea of an AI agent that handles research on demand, prepping for meetings, digging into market trends, or pulling up customer histories, is genuinely appealing. Salesforce addresses this with its Agentforce platform, which includes a deep research capability built for enterprise users already invested in the Salesforce stack.

But what does "Agentforce deep research" actually mean in practice?

This guide breaks down what the feature is, what it takes to get it working, where it runs into limits, and what the full pricing picture looks like. We'll look at how its data model shapes what it can and can't research, and where a broader tool integration approach gives you a more complete picture across your entire knowledge base.

What is Agentforce deep research?

The deep research agent is a behind-the-scenes research tool designed to process data and handle longer analytical tasks across sales, support, and other teams. Salesforce describes it as capable of market analysis, gathering data for strategy sessions, and delivering productivity insights.

It runs on the Atlas Reasoning Engine, which Salesforce describes as how "Agentforce understands, decides, and acts autonomously to provide trusted, accurate answers for every request." The engine takes a question, determines what data it needs, runs agentic loops to refine the answer, and returns a response. For deep research tasks, it works primarily with data already inside the Salesforce ecosystem, including CRM data, customer support history in Service Cloud, and anything pulled into Data 360 (formerly Data Cloud).

A concrete example: a sales manager asking it to "summarize all high-value opportunities in the pipeline and flag key risks based on recent support tickets" would trigger the agent to pull from Sales Cloud and Service Cloud. That works well when all of your relevant context already lives in Salesforce.

Setting up Agentforce deep research: requirements and reality

Getting Agentforce deep research running is not a quick configuration. Salesforce's own help docs make clear that most Agentforce capabilities require Enterprise Edition+, with additional add-on licenses varying by agent type.

The central role of Data 360

For Agentforce to do meaningful research, it needs good data in the right format. Salesforce's data layer for this is Data 360 (formerly Data Cloud), which creates the search index and retriever that agents draw from.

To use the Agentforce Data Library, Salesforce's help docs require you to first "turn on Data 360" and assign a user with a Data Cloud Architect permission set plus System Administrator profile. If your most useful company information lives in documents, wikis, or communication tools outside Salesforce, it won't reach the agent until this setup is complete.

Connecting external knowledge

Your company's information typically lives across many platforms. Product specs may be in Confluence, decisions get made in Slack, and strategy documents live in Google Docs. Agentforce can reach those sources, but not directly.

Salesforce's connector documentation lists a Confluence connector (currently in Beta, requires enabling through the feature manager and an Atlassian API token), a Google Drive connector, and a MuleSoft Direct connector for broader integrations. All of these flow data into Data 360 first, then make it available to agents through the Answer with Knowledge action.

Each connector requires dedicated setup, adds to the Data 360 cost, and introduces maintenance work. A research agent working without full access to these sources returns answers shaped by whatever data reached Salesforce.

eesel AI is built for this multi-platform reality. One-click integrations for Confluence, Google Docs, Slack, and 100+ other tools connect your company's knowledge in minutes, without a data migration project.

Agentforce deep research in action: use cases and key limits

Once configured, Agentforce deep research handles specific tasks well. Its limits appear most clearly when questions require information from outside the Salesforce data model.

Use cases within Salesforce

Within Salesforce's native data, the agent handles requests like:

• Analyzing sales performance: "Show me win rates for deals that included a product demo last quarter."

• Summarizing customer history: "Give me a summary of Acme Corp's last five support cases and their outcomes before my call."

• Identifying at-risk customers: "List all enterprise customers with dropping product usage and no support tickets in the last two months."

Salesforce's Agentforce 360 announcement cites several customer results: Reddit deflected 46% of support cases and cut resolution times by 84%, OpenTable resolved 70% of diner and restaurant inquiries autonomously, and 1-800Accountant achieved a 90% case deflection rate during tax week. These are Salesforce-native deployments.

The knowledge boundary

The most consistent pattern in user feedback is that research quality depends entirely on what's inside Salesforce. G2 reviews (1,095 reviews, 4.3/5 overall) show this clearly. One reviewer wrote:

Although it is very useful but Agentforce depends on the data in Salesforce so if data is messy and contains duplicate records then Agentforce's response will also not very accurate sometimes it gets hallucinated so admin knowledge is required.

(Aman R., Software Developer, Mid-Market, G2)

Real research tasks often require a broader view: technical documentation in internal wikis, ongoing product discussions in team chat, marketing plans and competitive analysis, and customer feedback from non-Salesforce platforms. An agent that can't access those sources returns answers from a narrow slice of your organization's knowledge.

Salesforce Ben, a Salesforce trade publication, tested Agentforce across six use cases in a six-month review and found it "only delivered well on two." Setup improvement was noted between the initial review and follow-up, but the pricing question remained: "Agentforce pricing is still clear as mud to me."

eesel AI chat is built for the multi-platform reality most teams operate in, connecting across your full knowledge stack without requiring a migration.

The full cost of Agentforce deep research

The Agentforce pricing page lists two consumption models and several add-on tiers.

Direct pricing options

Salesforce currently offers two usage-based models. One org cannot use both:

• Flex Credits: $500/100K credits. Launched May 2025. Covers customer-facing agents, employee agents, and voice.

• Conversations: $2 USD per conversation. The original model, still live alongside Flex Credits as of May 2026.

Both models scale with usage volume. The pricing page's own example estimates a service-case workflow at $0.30 per case (3 actions x $0.10), adding up to $1,800/month at 360,000 credits, with the explicit disclaimer that this "does not include other costs like Data 360 credits or other consumption services."

Additional costs not on the pricing page

Salesforce notes in its pricing page footnotes that Flex Credit figures do not include Data 360 credits or other consumption services. Several additional cost areas are not publicly disclosed on the Agentforce pricing page:

• Data 360 licensing: Required to use the Data Library and connect external knowledge sources. Not on the Agentforce pricing page.

• MuleSoft licensing and development: Needed to connect non-Salesforce systems beyond the documented connectors. Not on the Agentforce pricing page.

• Professional services: Salesforce links to an "expert guidance" datasheet from the pricing page, but no implementation fee is published publicly. Contact Salesforce for a quote.

• Internal admin time: Setup, testing, and ongoing configuration by a Salesforce-certified admin.

A capable tool with a narrow aperture

Agentforce deep research is a genuinely powerful capability within its scope. For organizations already running on Salesforce, with clean CRM data and the resources to set up Data 360, it can automate meaningful research tasks and deliver real productivity gains.

The constraints are real and worth understanding before committing. Most Agentforce features require Enterprise Edition or higher. External knowledge sources require Data 360, and the Confluence connector is currently in Beta. The costs go beyond the per-action or per-conversation rates. The research it delivers is only as complete as the data Salesforce can reach.

Before committing to this path, ask whether your team needs an AI assistant that works where they already are, pulling answers from across your entire knowledge base. For that use case, a platform-agnostic tool may fit better.

Unlock your team's knowledge with eesel AI

If you want an AI assistant that can answer questions from Confluence, Google Docs, Slack, Zendesk, and 100+ other tools without a Salesforce license or a data migration, eesel AI is worth a look. At $0.40 per task, you can have a working AI agent answering questions from your full knowledge base in minutes, not months.

Start your free trial today