- Agentforce is Salesforce's AI agent platform: agents that understand a natural-language request, draw on your Salesforce data and knowledge base, and take actions (flows, Apex, queries).
- The use cases that work best today: first-line customer support, an internal assistant for Sales / CSM / support teams, help preparing and following up on meetings, and pipeline analysis for leadership.
- An agent's quality depends mainly on three things: a clean knowledge base, well-scoped actions and tightly controlled access rights.
- Start small: one agent, a few measurable use cases, a pilot group. Meir IT's Agentforce quickstart (€4,000 excl. VAT) covers activating one agent, permissions, 5 use cases and training.
After the wave of demos, the question is no longer "What is Agentforce?" but "What actually works for companies like mine?". This article answers with real use cases we have deployed at Meir IT, named FY25 Agentforce Top Partner by Salesforce, and with the method we use to get started.
What is Salesforce Agentforce?
Agentforce is the AI agent layer of the Salesforce platform. An agent receives a request (from a customer in a chat, or from an employee in Salesforce or Slack), identifies the topic, then chooses among the actions you have authorised: searching the knowledge base, reading records, launching a flow, creating a case, updating an opportunity. Responses are generated by a large language model, governed by the Salesforce Einstein Trust Layer (masking of sensitive data, no data retention by the model provider, audit logging).
The difference from a traditional chatbot: you do not script every conversation. You describe topics, instructions and actions, and the agent reasons to chain the steps together. The trade-off is that you need to define carefully what it is allowed to do and say.
What are real-world Agentforce use cases?
1. Replacing a support chat with an agent connected to the knowledge base
Yomoni used Zendesk for its chat. With Meir IT, the knowledge base was moved into Salesforce and the chat is now served by Agentforce: the agent answers from the Salesforce knowledge articles and, when the answer does not satisfy the customer, it creates a case for the support team. Customers get an immediate answer to frequent questions, and advisers pick up the requests that genuinely need a human, with full context.
2. An internal assistant for Sales, CSM and support teams
At Natural Origins, Agentforce serves internal teams: sales reps, Customer Success Managers and support staff can ask Salesforce questions in natural language (for example "Which of this customer's orders are late?" or "Summarise this account's history") and take actions without navigating several screens. The gain is less spectacular than a public chatbot, but very tangible day to day.
3. Helping consultants before and after their meetings
Neovia wanted to help its consultants respond better to prospects and clients, and to improve after every call. The agent draws on an internal knowledge base made up of articles and PDFs, supplemented by content scraped from Neovia's websites. Consultants get sourced answer suggestions and structured feedback on their conversations.
4. Giving the board a fast read on the pipeline
For Technilum, the agent helps leadership analyse open and won opportunities: trends, at-risk deals, period-on-period comparisons, by asking questions rather than building reports. This requires reliable opportunity data: an Agentforce project is often a chance to tidy up the CRM.
Agentforce use cases compared
| Use case | Users | Data required | Typical actions |
|---|---|---|---|
| Support chat (Yomoni) | Customers | Public knowledge base | Answer, create a case |
| Internal assistant (Natural Origins) | Sales, CSM, support | Accounts, orders, cases | Query, summarise, update |
| Consultant coaching (Neovia) | Consultants | Articles, PDFs, web content | Suggest answers, analyse a conversation |
| Pipeline analysis (Technilum) | Leadership | Opportunities | Summarise, compare, flag |
How do you get started with Agentforce?
Our method, proven on our projects, comes down to five steps:
- Pick a single agent and a pilot group: for example the website chat, or the CSM assistant. Not both at once.
- List 5 precise use cases, written as real questions or requests, with the expected answer or action. They serve as both specification and test set.
- Prepare the sources: up-to-date knowledge articles, documents, reliable fields. An agent will never do better than the data it reads.
- Define permissions: which objects and fields the agent can read, which actions it can run, and when it must hand over to a human.
- Test, measure, expand: rate of useful answers, cases created, time saved. Only add use cases once the first ones are solid.
Beyond the first agent: specialised Agentforce offerings
Salesforce regularly expands the Agentforce family with business-specific products. We offer fixed-price quickstarts for several of them; their exact scope depends on your licences and Salesforce edition, and is confirmed during scoping.
| Offering | What it is for | Meir IT quickstart (excl. VAT) |
|---|---|---|
| Agentforce (custom agent) | A customer-facing or internal agent for your use cases | €4,000 |
| Agentforce IT Service | Internal IT service management (incidents, requests) with agents | €5,000 |
| Agentforce Operations | Automation of internal operational processes | €3,000 |
| Agentforce Coworker | AI assistant for employees in their day-to-day work | €3,000 |
| Agentforce Help Agent | Help and support agent for users | €3,000 |
| Slackbot | AI assistant in Slack, connected to your data | €2,000 |
On the front-office side, Agentforce also combines with Service Cloud (quickstart €4,000 excl. VAT) for support, with Sales Cloud for pre-sales and with Revenue Cloud for quotes: see our Revenue Cloud / Agentforce Revenue Management guide. For marketing, read our article on Marketing Cloud Next.
The mistakes that derail an Agentforce project
- An outdated knowledge base: the agent will confidently give wrong information. Cleaning up articles is a prerequisite, not an option.
- Vague use cases ("help the sales team"): impossible to test. Write real questions.
- Too many permissions: an agent should only see what the user concerned is allowed to see, and only act within a defined scope.
- No way out to a human: always plan for escalation (creating a case, transferring to an adviser), as at Yomoni.
- No measurement: without KPIs from the start, you cannot decide whether to expand or adjust.
See more projects in our case studies.
Frequently asked questions
What is Salesforce Agentforce?
Agentforce is Salesforce's AI agent platform. An agent understands a natural-language request, draws on your Salesforce data and knowledge base, and performs authorised actions (answering, creating a case, launching a flow, updating a record).
Can Agentforce replace a chat tool like Zendesk?
Yes, this is a use case Meir IT implemented for Yomoni: the knowledge base was moved into Salesforce, Agentforce answers from that knowledge base and creates a case when the answer does not satisfy the customer.
Do you need Data 360 to use Agentforce?
Not necessarily for a first agent: it can rely on Salesforce objects and the knowledge base. Data 360 (formerly Data Cloud) becomes useful for working with large documents or external data, or for unifying several sources.
How long does it take to launch a first agent?
With a limited scope (one agent, five use cases, a pilot group) and a usable knowledge base, a first agent can go live within a few weeks. The timeline depends mainly on the quality of the source data.
How much does an Agentforce quickstart with Meir IT cost?
The Agentforce quickstart is offered at a fixed price of €4,000 excl. VAT: activation of one agent, permissions, 5 use cases and training. Agentforce licences and consumption billed by Salesforce are extra.
Is the data sent to the AI protected?
Agentforce relies on the Salesforce Einstein Trust Layer, which governs exchanges with large language models (data masking, no retention by the model provider, audit logging). The user's Salesforce access rights also apply.