AI and Agentforce

Salesforce Agentforce Use Cases: How Businesses Can Put AI Agents to Work

By Ranjeet Srivastava16 min read
1 view
An illustration of AI agents at work: a central agent hub linked to four panels for sales, customer service, marketing and operations, with a person reviewing the work at the bottom.

By now most teams have seen a Salesforce Agentforce demo. The harder question comes after it: where would an AI agent actually earn its place in your business, what would it need to see to do the job well, and who still has a hand on the wheel when it does? This guide walks through eight practical Salesforce Agentforce use cases across sales, customer service, marketing and operations. For each one we cover the business problem, how the agent workflow runs, the data and integrations behind it, where people stay in control, what to expect and what to watch for.

After the use cases we step back to the shared foundations, the oversight model and the implementation considerations, including the limits that vendor pages tend to skip. Product names and capabilities here were checked against Salesforce’s own documentation in November 2026. Salesforce renames things often (agent topics became subagents in April 2026, for example), so confirm current names and licensing with Salesforce before you plan a budget.

Key takeaways

  1. Start narrow: The best first use cases have a clear start and end, data that already lives in Salesforce, and an obvious point where a person takes over.
  2. Data decides quality: Clean records, current knowledge and well-built integrations matter more to an agent’s answers than any setting in the agent itself.
  3. Oversight is designed, not assumed: Guardrails, fixed rules for decisions that cannot vary, approval modes, handoff flows and testing each keep people in control.
  4. Each team gets a different shape of agent: Sales agents qualify and update, service agents resolve and escalate, marketing agents build and re-engage, and operations agents schedule and support.
  5. Plan the cost and the change: Salesforce offers consumption-based and per-user options, and the work to prepare, test and maintain agents is part of the real cost.
  6. Treat vendor results as vendor-reported: Measure your own baseline before launch so you can see what the agent changes in your business.

What Salesforce Agentforce does, in plain terms

Salesforce Agentforce is Salesforce’s AI agent platform. An agent is given a job, instructions and a set of actions. When a request arrives, the Atlas Reasoning Engine breaks it into smaller tasks, evaluates each step and proposes a plan until the answer or action is complete. The actions are the same tools your Salesforce team already builds with: Flows, prompt templates, Apex code and MuleSoft API connectors.

That is the practical difference from a traditional chatbot, which follows a predefined conversation tree. An agent understands the request, builds a plan and chooses among the actions it has been given. It is also why Flows do not go away: they stay the reliable hands that do the steps which must run the same way every time, while the agent decides which step comes next.

You will meet these terms throughout the use cases:

  • Agentforce Builder and Agent Script: the workspace and language for building agents. Agent Script combines natural-language instructions with programmatic rules, so an agent’s behavior does not rely solely on interpretation by a language model.
  • Subagents and actions: a subagent is a focused area of an agent’s job, such as order status or meeting booking (called a topic before April 2026). Actions are what it can do.
  • Data 360: the data layer, formerly Data Cloud, that brings together CRM data and external data so an agent can ground its answers.
  • Einstein Trust Layer: the set of data protections that includes dynamic grounding, zero data retention and toxicity detection, with an audit trail of agent actions.
  • Omni-Channel: the routing layer that can hand a conversation from an agent to a person.

Eight Agentforce use cases at a glance

Each use case below follows the same pattern, so you can compare them side by side and see which fits the team that feels the most pressure today.

Eight Salesforce Agentforce use cases by area, what each relies on and where a person stays involved
AreaUse caseWhat it relies onWhere a person stays involved
SalesInbound lead response and qualificationLead and account records, approved sales content, rep calendarsReps take over interested leads; sensitive topics go straight to a person
SalesPipeline updates and deal hygieneOpportunity fields, calls, emails and meeting notesReps approve updates in suggestive mode
ServiceCommon support requestsKnowledge, cases, orders, messaging channelsComplex or sensitive cases escalate to a rep with the history attached
ServicePhone calls with a clean handoffVoice channel, identity checks, cases, Omni-ChannelLive agents take escalations with the case already open
MarketingCampaign creation and optimizationData 360 profiles, consent data, brand contentMarketers approve audiences, content and sends
MarketingDormant lead re-engagementLead history, scoring, suppression listsSales takes sales-ready leads; marketing sets the limits
OperationsField service scheduling and technician supportWork orders, technician skills, parts, manualsDispatchers choose among options; technicians review summaries
OperationsEmployee HR and IT helpPolicies, tickets, identity and HR systemsSpecialists handle sensitive and high-privilege requests

Salesforce Agentforce use cases for sales teams

1. Inbound lead response and qualification

A prospect fills in a form at nine in the evening. The first reply arrives the next afternoon, after the interest has cooled, and your reps are spending their days researching and logging activity instead of talking to buyers. The cost shows up as leads that were never worked properly, not as a line item anyone can point to.

Agent workflow
A new lead arrives. The agent replies on your website or by email, answers product questions from approved content, handles common objections, checks the lead against the qualification criteria you define and offers real calendar slots. When the lead is ready, it books the meeting, records the conversation on the lead and hands over to a rep with context. Salesforce describes its sales development agent as one that answers product questions, handles objections and books meetings for reps.
Data and integrations
Lead, contact and account records, the qualification rules, product FAQs, case studies and sales plays the agent may quote, rep calendars, and the email and web chat channels it speaks on.
Human oversight
You set guardrails for when and how the agent reaches out, as Salesforce’s sales agent page puts it. Decide which topics go straight to a person (pricing exceptions, legal questions, complaints) and review a sample of conversations every week.
What to expect
Faster first responses, more consistent messaging and fewer leads that slip through. Reps spend their time on conversations with people who have shown interest.
Watch for
Lead ownership and routing rules that were never written down, and consent and opt-out rules for email. Start with inbound. Salesforce’s own AI SDR guide, published in April 2025, says fully autonomous outbound is more complex and that the most sophisticated AI SDRs were then better suited to inbound leads.

2. Pipeline updates and deal hygiene

Forecast calls go badly when opportunity records are stale. Reps update Stage and Next Steps late, if at all, because the real information sits in call recordings, emails and their own notes. Managers end up asking for updates instead of coaching.

Agent workflow
The agent reads calls, meetings, emails and seller notes, turns what it finds into structured updates to fields such as Stage and Next Steps, and either applies them or proposes them. Salesforce describes a suggestive mode, where reps review, edit and approve updates, and an autonomous mode, where updates apply automatically, with source context shown for each.
Data and integrations
Opportunity fields and their definitions, email and calendar capture, call recordings or transcripts, and the rep workspace where proposed updates are reviewed.
Human oversight
Begin in suggestive mode. Move a field to autonomous mode only when the updates have been right for long enough and a wrong value is cheap to fix. Keep the source context visible so a rep can see where a value came from.
What to expect
Cleaner pipeline data with less typing, and forecast conversations that start from better information.
Watch for
Fuzzy field definitions. If your team disagrees about what Stage 3 means, the agent will apply the confusion at scale. Also check the privacy and consent rules for recorded calls.

Other sales jobs follow the same pattern. Salesforce lists prospect research, account briefs, quoting in natural language and sales coaching role-plays among its sales agent capabilities. Build the first two use cases well and the rest become variations.

Agentforce use cases for customer service

3. A service agent for common support requests

Routine questions queue up behind complex ones. Customers wait for answers that already exist in your knowledge base, and the reps who finally pick up the case ask for details the customer gave an hour ago.

Agent workflow
Per Salesforce Help, an Agentforce Service agent processes incoming cases and autonomously resolves common inquiries on channels such as enhanced messaging. The customer asks, the agent finds the answer in knowledge and customer records, or takes an action such as checking an order or booking an appointment. If the request is complex or sensitive, an Omni-Channel Flow escalates it to a service rep or another destination.
Data and integrations
Knowledge articles, case, contact and order records, your messaging channels, and the order, billing or fulfillment systems the agent may need to query or update through Flows, Apex or MuleSoft APIs. Omni-Channel routing has to be configured for the handoff.
Human oversight
Write down what the agent must never handle, such as refunds above a threshold, legal complaints or vulnerable customers, and make those routes rules, not suggestions. Test with real past conversations before launch, then review transcripts and escalations every week.
What to expect
Answers at any hour, fewer repetitive cases for your reps and consistent responses grounded in your own knowledge, with reps who pick up escalated cases already informed.
Watch for
A thin or outdated knowledge base, which produces thin answers. Judge success by resolution quality and customer satisfaction, not only by how many conversations the agent closed.

4. Phone calls with a clean handoff

Phone menus frustrate callers, and a transfer to a person often means starting over. The handoff is where customers decide whether the experience was any good.

Agent workflow
A caller reaches an Agentforce Voice agent, which verifies the account, opens a case and tries to resolve the request. When a person is needed, the agent transfers the call with the work already done. Salesforce’s own guide to voice escalation describes moving from an emergency bailout to a deliberate handoff, using rules such as collecting an email address before escalating or creating the case first.
Data and integrations
The voice channel and telephony setup, customer and case records, knowledge, an identity-verification step and Omni-Channel routing to live agents.
Human oversight
Write the escalation rules down and test that they hold. Salesforce notes that the default behavior hands off as soon as a caller asks for a person, so dynamic escalation needs deliberate configuration, and where a rule must never be skipped, Agent Script lets you enforce it as logic instead of relying on model instructions. Add supervisor review of recordings and transcripts.
What to expect
Salesforce expects better-prepared handoffs and fewer unnecessary transfers. Confirm that against your own call data after a pilot.
Watch for
Voice is less forgiving than chat: interruptions, accents and delays all show. Start with a few narrow call types, and check recording and consent rules for the regions you serve.

Agentforce use cases for marketing

5. Campaign creation and optimization

Marketers are asked for more personalized content, more segments and more channels than a manual process can deliver. Building audiences, writing variants and tuning delivery by hand takes weeks, and by the time the learnings arrive the moment has passed.

Agent workflow
The marketer states a goal. The agent proposes an audience from Data 360 using a plain-language prompt, drafts the brief, messages and journey, and after approval watches performance and adjusts delivery and creative within the limits you set. Salesforce describes Marketing Cloud Next as agentic execution across the campaign lifecycle, with two-way conversations across channels such as email, SMS, web and WhatsApp.
Data and integrations
Unified customer profiles in Data 360, consent and preference data, your brand guidelines and approved content, product information and the channel connections you send through.
Human oversight
A person approves audiences, content and sends, especially anything with regulated or contractual claims. Keep consent checks as hard rules, and limit how much an agent may change on its own.
What to expect
Faster campaign builds, more tailored content and a quicker path from results to adjustments.
Watch for
Product and licensing confusion. Marketing Cloud Next, Engagement+ and Account Engagement+ serve different starting points, and existing Marketing Cloud Engagement and Account Engagement customers keep what they have. Confirm which fits before you plan.

If marketing is your starting point, our guide to Salesforce Marketing Cloud implementation covers the platform side in more depth.

6. Re-engaging dormant leads for B2B demand generation

Every B2B CRM holds thousands of leads that went quiet. Nobody has the hours to follow up on cold lists, so the pipeline value in data you already own goes unclaimed.

Agent workflow
The agent selects dormant records using rules you set, sends personalized follow-ups, keeps the conversation going when a lead replies and passes sales-ready leads to sales with the context attached. Salesforce describes its SDR agent for B2B demand generation as one that re-engages dormant CRM records, follows up on cold lists and nurtures leads toward sales-ready conversations.
Data and integrations
Lead and contact history, campaign history, lead scoring, consent and suppression lists, and the product content the agent may use.
Human oversight
Marketing and sales agree on what sales-ready means and how fast sales must respond. Suppression rules and daily contact limits are hard rules, and someone samples the conversations.
What to expect
Pipeline from records you already hold, and marketing and sales working from the same customer record.
Watch for
Email reputation and consent. Writing to an old list too fast can hurt deliverability, and regional rules differ. Clean duplicates and bad addresses first, and ramp volume slowly.

Agentforce use cases for operations

7. Field service scheduling and technician support

Scheduling eats dispatcher hours, cancellations leave gaps in the day, and technicians spend time on paperwork that should go to repairs. Customers wait on the phone just to change an appointment.

Agent workflow
Following Salesforce’s Agentforce for Field Service announcement: a customer asks to book, reschedule or cancel in natural language, and the agent infers the job type, creates a work order and service appointment, finds technicians by location and skill and offers time slots. When cancellations or early finishes open gaps, it suggests replacements using the Field Service optimization engine. Technicians can get pre-work briefs, troubleshooting guidance and drafted post-work summaries.
Data and integrations
Work orders, service appointments, technician skills and territories, parts and inventory, product manuals and service history, plus Salesforce Field Service itself and any parts or ERP system that holds availability.
Human oversight
Dispatchers choose among the suggested options for schedule gaps, technicians review and refine summaries, and you decide which job types the agent may book without a person.
What to expect
Less phone tag, fewer idle gaps and technicians who arrive better informed.
Watch for
Skills, territories and SLAs that live in people’s heads rather than in the system. Some of these capabilities launched in stages in 2025, and Salesforce’s own announcement tells buyers to decide on fully released features, so confirm current availability.

8. Employee HR and IT help

HR and IT teams answer the same questions all week: how much leave is left, how to reset access, who approves a purchase. Specialists spend their day on routine requests while the complicated ones wait.

Agent workflow
An employee asks in Slack, Microsoft Teams or a portal. The agent answers from policy and knowledge, creates or updates the request and runs permitted actions, such as submitting a leave request or resetting access, then hands exceptions to a specialist. Salesforce lists leave requests, HR policy questions, onboarding, help desk tickets and access resets among its employee and IT service examples.
Data and integrations
HR policies and knowledge, ticket records in Salesforce or your service desk tool, and connections to your HR system and identity provider for the actions that change something.
Human oversight
Route sensitive topics such as grievances and medical matters to people by rule. Require approval for high-privilege actions such as access changes. Salesforce says its Agentforce Coworker works within employee permissions, sharing rules and configured access controls, with approval and audit capabilities for supported operations.
What to expect
Quicker answers to routine questions and specialists who spend their time on exceptions.
Watch for
The sensitivity of HR data and the privileges behind IT actions. Decide who owns the policy content, because an agent that quotes an outdated policy confidently is worse than no agent.

The data and integrations every use case depends on

Across all eight use cases, the same few foundations decide whether an agent is useful or merely impressive in a demo.

  • Clean CRM data. Duplicate contacts, inconsistent picklists and empty required fields become wrong answers. A Salesforce health check is a sensible first step.
  • Current, owned knowledge. Service, HR and sales agents are only as accurate as the articles and content they are allowed to quote, and someone has to own keeping them current.
  • Unified and external data. Data 360 extends what an agent can see beyond CRM records, which matters for marketing, account research and anything that crosses systems.
  • Reliable actions. Flows, Apex and MuleSoft APIs do the real work in other systems. Salesforce also lists MCP support for connecting agents to tools through verified partner servers. Well-built Salesforce integrations decide how much an agent can safely do.
  • Right-sized permissions. An agent should reach only the records and actions its job requires, in the same way you would scope a new hire.

Keeping people in control: human oversight that works

Oversight is not one switch. The stronger deployments layer several controls and match them to how costly a mistake would be in that use case.

  1. Define the scope. Give the agent a narrow job, clear instructions and guardrails. Salesforce describes agent guardrails as a mix of user-defined safeguards and Salesforce-managed protections that block off-topic conversations and reduce the risk of hallucination and bias.
  2. Put fixed rules in code. Salesforce’s architect guidance acknowledges that language models can produce different outcomes from identical inputs. Its answer is a hybrid model in which anything you can express as a rule, such as a refund limit or a required check, runs as deterministic logic, and the model handles judgment and language.
  3. Choose the approval mode. Use suggestive modes where a person approves changes, and allow autonomy only for low-risk, well-tested actions.
  4. Design the handoff. Decide when the agent must stop, who takes over, how fast and with what context. A handoff that loses the conversation is a defect.
  5. Protect the data. The Einstein Trust Layer adds dynamic grounding, zero data retention and toxicity detection, and an audit trail records what agents did.
  6. Test and monitor. Generate test cases from real examples, inspect the responses and watch the agent after launch. Agentforce provides testing tools and observability for this.

An agent that cannot hand off gracefully is not finished. Test the escalation path as hard as you test the answers.

Implementation considerations: from first use case to rollout

Choosing the first use case

Four questions narrow the list quickly:

  • Is the job narrow, repeatable and easy to describe in one paragraph?
  • Does the data it needs already live in Salesforce, and is it reasonably clean?
  • Is there a named person or queue ready to take over every exception?
  • Can we measure how the work performs today, so we can tell what changed?

A rollout sequence that holds up

  1. Pick one job. Write it in a paragraph, with its start, its end and the point where a person takes over.
  2. Audit the foundations. Check the data, knowledge, automations, permissions and integrations the job depends on.
  3. Design boundaries and handoffs. Decide what the agent may do, what it may never do and who receives each exception.
  4. Build and test. Build in Agentforce Builder, reuse existing Flows and Apex, and test against real past cases.
  5. Pilot with review. Launch to a small group with a person reviewing outputs, and widen only as the results hold.
  6. Measure and expand. Compare against your baseline, tune the agent and then move to the next use case.

Licensing, cost and ownership

Salesforce’s Agentforce pricing page lists consumption-based options, including Flex Credits and conversations, alongside per-user options such as add-ons and an Agentforce user license, and it notes that the page is subject to change. Volume, agent type and your existing editions all shape the bill, so model your expected conversations and actions before you commit. Also budget for the work around the license: preparing data, building, testing and maintaining the agent. Each agent needs a business owner who answers for its behavior and an admin who maintains it.

Measuring what changed

Pick the measures before launch. Useful ones include first response time, resolution rate, escalation rate, quality of handoffs, time reps spend on admin and the completeness of CRM data. Review the same numbers at 30, 60 and 90 days so you can separate what the agent changed from ordinary variation.

Risks and limitations to plan for

  • Wrong or invented answers. Guardrails and grounding reduce this risk. They do not remove it, so keep human review where an error would be costly.
  • Poor data in, poor answers out. An agent exposes every gap in your records and knowledge.
  • Autonomy creep. It is easy to let an agent do more than it has earned. Expand permissions only after the evidence supports it.
  • Failed escalation. Without a reachable human or a clear next step, customers get stuck. Test out-of-hours routing as well.
  • Unpredictable usage costs. Consumption pricing rewards forecasting. Watch volumes from the first week.
  • Moving names and staged features. Salesforce renames products and releases capabilities in stages, and some pages describe the same feature differently. Check current documentation and your edition before you commit.
  • Consent and compliance. Outreach, recorded calls and HR data each carry rules that differ by region. Involve legal and security early.
  • Vendor-reported results. Customer figures on vendor pages describe someone else’s business. Treat them as prompts for your own baseline.

Where Brazen Group fits

Brazen Group is a Salesforce implementation and consulting partner. For Agentforce, the work usually starts before any agent is built: assessing the data, automations and integrations a use case depends on, then narrowing the list to the few worth doing first. From there we build agents with guardrails, escalation paths and tests, and support them after launch. Our Salesforce Agentforce AI service covers that path, and our RevOps automation work helps when the real problem sits across sales, marketing and service handoffs. If you are comparing AI approaches first, see Claudeforce vs Agentforce, and for the wider picture, what Agentforce means for sales and service operations.

Putting Agentforce to work

The businesses that get value from Salesforce Agentforce tend to treat an agent like a new team member: a clear job, the right information, a manager who reviews the work and a way to ask for help. Start with one narrow use case, prepare the data behind it, design the handoff before the answers and measure against where you began. Then widen the circle.

Sources

Product names and capabilities were checked on November 19, 2026 against these Salesforce pages:

Frequently asked questions

What is Salesforce Agentforce?

Salesforce Agentforce is Salesforce’s platform for building, deploying and managing AI agents that work with your Salesforce data and processes. An agent is given a job, instructions and a set of actions, such as running a Flow or calling an API, and uses a reasoning engine to plan the steps. Salesforce also ships prebuilt agents for service, sales, marketing and commerce.

How is an Agentforce agent different from a chatbot or a Flow?

A traditional chatbot follows a predefined conversation tree, and a Flow follows a fixed set of steps. An agent reads the request, builds a plan and chooses among the actions it has been given. In practice they work together: agents call your existing Flows, Apex and APIs for steps that must run the same way every time, and Agent Script lets you combine fixed logic with model reasoning in one agent.

Which Agentforce use case should we start with?

Start with a narrow, repeatable job where the data already lives in Salesforce, a person is ready to take over exceptions and you can measure the result. Inbound lead response, common service requests, pipeline updates and employee help desk questions often fit. Avoid starting with a job that needs judgment on every case or with data nobody trusts yet.

What data and integrations does Agentforce need?

It needs the CRM records, knowledge and content relevant to the job, with consistent fields and few duplicates. Beyond that it depends on the use case: Data 360 for unified or external data, Flows, Apex or MuleSoft APIs for actions in other systems, and channels such as messaging, voice, email or Slack. Permissions matter too, because an agent should only see and do what its job requires.

How do people stay in control of Agentforce agents?

Through several layers: the agent’s scope and guardrails, deterministic rules written in Agent Script for decisions that cannot vary, suggestive modes where a person approves changes, Omni-Channel flows that hand conversations to a rep, the Einstein Trust Layer, and testing and monitoring before and after launch. The right mix depends on how costly a mistake would be in that use case.

What drives the cost of Agentforce?

Salesforce lists consumption-based options, including Flex Credits and conversations, and per-user options such as add-ons and an Agentforce user license. Your cost drivers are usually how many conversations or actions the agents handle, which type of agent you deploy and which Salesforce editions you hold, plus the work to prepare data, build, test and maintain the agents. Ask Salesforce or your partner for current pricing for your edition and use case.

Want to Find the First Agent Worth Building?

Tell us which team feels the pressure and where the work piles up. We’ll help you choose a first use case, check the data and integrations behind it and design the guardrails, through our Salesforce Agentforce AI consulting.

Talk to Our Agentforce Team
Salesforce AgentforceAgentforce Use CasesAI AgentsSales CloudService CloudMarketing Cloud