Most lists of Agentforce use cases are long and undifferentiated. "Automate customer service" sits next to "personalise every interaction" with no indication of which one a team could ship next quarter and which one needs two years of data work first.
This is a shorter list. Every use case below is something we would put in front of a client today, with a note on how it is built and what tends to go wrong. At the end are the ones we would steer you away from for now.
How every use case here is built
An Agentforce agent is a set of subagents (called topics until April 2026). Each subagent has a classification description the reasoning engine uses to route a message to it, a scope, instructions, and a set of actions.
Actions are where the work happens. The main types:
- Flow for record lookups and rule-based updates. This is the default for most actions.
- Apex (invocable methods) when the logic is complex, needs callouts, or has to enforce sharing and validation precisely.
- Prompt templates from Prompt Builder when the action itself needs an LLM, such as summarising a case or drafting an email grounded in record data.
- MuleSoft APIs and External Services for systems outside Salesforce, such as an ERP or an order management platform.
Since the new Agentforce Builder and Agent Script arrived, you can also make parts of an agent deterministic: run an identity check before anything else, force a transition to an escalation subagent when a condition is met, or hide actions until a variable is set. A good rule is that anything you would put in a policy document goes in script or code, and anything about tone goes in instructions.
Grounding is the context the agent answers from: CRM records returned by actions, Knowledge articles and files indexed through an Agentforce Data Library, or Data 360 (formerly Data Cloud) data such as unified profiles and calculated insights.
Service use cases
Order and case status for verified customers
The agent verifies the customer (an emailed or texted code, or an authenticated portal session), then answers "where is my order" or "what is happening with my case".
- Build: a verification subagent with deterministic logic, then two or three Flow actions that query Orders or Cases filtered on the verified Contact or Account ID. Return a small, structured set of fields, not the whole record.
- Watch for: actions that query by anything other than the verified ID. Verification on its own does not restrict what an action can return.
Knowledge answers with clean escalation
The agent answers product and policy questions from Knowledge articles and hands off to a person when it cannot.
- Build: an Agentforce Data Library over published external articles, the Answer Questions with Knowledge action, and an escalation subagent that routes through Omni-Channel to a queue with the transcript attached.
- Watch for: stale or overlapping articles. The agent will quote whatever is retrieved. Fix content before tuning instructions.
Case triage and summarisation for agents
This one is aimed at service staff rather than customers. On case creation, a prompt template classifies the case, suggests a priority and drafts a summary, and the human agent sees it in the console.
- Build: a record-triggered Flow calls a Field Generation or Flex prompt template grounded on the case, its contact and recent related cases. Write the output to fields the human can accept or overwrite.
- Watch for: writing AI output straight into fields that drive routing or entitlements without a human check in the first weeks.
Sales use cases
Inbound lead follow-up with the SDR agent
Salesforce's Sales Development (SDR) agent engages inbound leads by email, answers questions from approved content, handles common objections and books meetings for a rep.
- Build: configure the standard agent, restrict the content it can ground on, define the lead criteria it works, and connect calendar booking. Most of the effort is in deciding which leads it may contact and when it must stop.
- Watch for: unclear hand-off rules. Reps need to know exactly when a lead has moved from the agent to them.
Pre-meeting account briefs
An employee agent in Sales Cloud produces a one-page brief before a call: open opportunities, recent cases, last activities, contract renewal dates.
- Build: a Flex prompt template grounded on the Account and related records through a Flow or Apex data provider, exposed as an action. Because an employee agent runs as the rep, it only sees what the rep can see.
- Watch for: pulling 200 activities into the prompt. Pre-filter and summarise in the data provider, not in the prompt.
Opportunity updates from call notes
The rep pastes or dictates notes, and the agent proposes updates to next step, close date and stage for the rep to confirm.
- Build: a prompt template extracts the changes, an Apex action validates them against stage rules, and the action requires user confirmation before it writes.
Sales Coach role play
Agentforce Sales Coach lets a rep rehearse a call against a simulated buyer using deal context, then gives feedback. It is low risk, uses data you already have, and is a reasonable way to get reps comfortable with agents.
Commerce use cases
Salesforce's commerce agents (the Shopper, Buyer and Merchant agents) are available for Commerce Cloud storefronts.
- Shopper or Buyer agent for product discovery and order questions. Grounded on the product catalogue and order history. It works well when product data is structured and attributes are consistent. It works badly when half the catalogue's specifications live in PDF datasheets.
- Merchant agent for catalogue and promotion setup. Useful for merchandisers who build many similar promotions or product descriptions. Keep publication behind human approval.
- B2B reorder. A buyer asks to repeat last month's order with two changes. The agent reads the order history, builds a cart and leaves checkout to the buyer. Pricing still comes from the commerce price books, not from the agent.
Internal operations use cases
- HR and IT policy answers. An employee agent over a Data Library of policy documents. Split documents by policy, keep effective dates in metadata, and retire old versions from the index.
- Deal desk summaries. A prompt template summarises a quote approval request (discount, non-standard terms, margin) for the approver. The approval decision stays in the approval process.
- Handover summaries. When a case or account changes owner, the agent writes a structured handover note. It saves real time and the risk is negligible.
Which to start with
| Use case | Data dependency | Risk of a wrong answer | Typical effort | Start here? |
|---|---|---|---|---|
| Order or case status | Orders and Cases linked to Contacts | Low | Low | Yes |
| Knowledge answers with escalation | Current external articles | Low to medium | Medium | Yes |
| Case triage for service staff | Case history | Low, human reviews | Low | Yes |
| Pre-meeting account briefs | Account, Opportunity, Activity data | Low | Low | Yes |
| SDR agent on inbound leads | Lead data, approved content | Medium | Medium | Second wave |
| Commerce shopper agent | Structured catalogue | Medium | Medium to high | When catalogue is clean |
| Opportunity updates from notes | Stage rules | Medium | Medium | Second wave |
The pattern is consistent: start with use cases where the agent reads more than it writes, and where a person sees the output before it matters. If you have not yet checked whether your org can support any of these, the Agentforce readiness guide covers that.
Use cases to avoid for now
- Refunds, credits and goodwill without hard caps. If you must, put the limit in Apex and require a human above it. An instruction saying "never refund more than X" is not a control.
- Configuring and pricing complex quotes. The rules live in Salesforce CPQ or Revenue Cloud for a reason. An agent can explain a quote; it should not assemble one with 40 lines and three bundles.
- Calculations that must be exact. Tax, pro-rated billing, commission. Do the maths in code, return the result, let the agent explain it.
- Writes across several systems in one turn. Without idempotent APIs and a compensation strategy, a retry can create two orders. Put that orchestration in MuleSoft or Apex, not in the reasoning loop.
- Anything regulated that needs a qualified human. Medical, legal or financial advice, even when the data is available.
- Unauthenticated access to account data. Anonymous visitors should only get public content.
Each of these can become viable later, once you have a tested action layer and a governance process. The controls involved are covered in Einstein Trust Layer and Agentforce security, and the choice between Flow and code for actions in Apex vs Flow.
Fixholics designs and builds Agentforce agents for service, sales and internal teams, starting with the use cases above and the action and test layer underneath them. If you want help choosing a first use case or building it properly, see our Agentforce services or our Service Cloud practice.
Frequently asked questions
How do you measure whether an Agentforce use case is working?
Track the share of conversations resolved without a hand-off, the reasons for each escalation, accuracy on a sample of sessions reviewed by a person, and consumption per conversation. Agree the targets with the business owner before launch, and review a sample of sessions every week rather than relying on dashboards alone.
Can Agentforce update records or only answer questions?
It can update records, create cases, book appointments and call external systems through actions built on Flow, Apex, prompt templates, MuleSoft APIs or External Services. Write actions should be narrow, validated in code, and set to require user confirmation where a mistake would matter.
Is Agentforce only for Service Cloud?
No. Service is where most early deployments sit because the volume and the Knowledge base already exist, but there are standard agents and patterns for sales development, sales coaching, commerce storefronts and internal employee support. The build approach is the same platform in each case.
PartnerWritten by the Fixholics architecture team: certified Salesforce consultants who deliver this work for clients worldwide.