Short answer: Your Salesforce org is ready for Agentforce when four things are true: your data is clean and connected, the agent has reliable actions it can call (Flows, Apex or APIs), permissions limit what it can see and do, and your knowledge content is current. Most orgs have gaps in at least two of these, and those gaps decide whether a pilot succeeds.
Agentforce — now marketed by Salesforce as Agentforce 360 — is one of the biggest changes to the Salesforce platform in years. Your Salesforce account team is probably already talking to you about it. But an AI agent does not fix an org. It amplifies whatever is already there. Duplicate records become wrong answers. Undocumented automations become unpredictable actions. Loose sharing rules become data exposure.
This guide explains what Agentforce actually needs to work, walks through the six pillars of Agentforce readiness, gives you a checklist you can run today, and shows how to choose a first use case that will prove value instead of stalling.
What Agentforce actually needs to work
Agentforce is Salesforce's platform for building AI agents that can take action inside your business, not just answer questions. A service agent can look up an order, process a return and update the case. A sales agent can qualify an inbound lead and book a meeting. To do that reliably, every agent depends on four layers working together.
| Layer | What it does | What it depends on in your org |
|---|---|---|
| Reasoning | Interprets the request and plans the steps (the Atlas Reasoning Engine) | Clear topics, instructions and guardrails defined in Agentforce Builder |
| Grounding | Gives the agent facts to work from | Accurate CRM records, Data 360, knowledge articles and documents |
| Actions | Lets the agent change something in the real world | Well-built Flows, Apex classes, prompt templates and APIs |
| Guardrails | Controls what the agent can see and do | Profiles, permission sets, sharing rules and the Einstein Trust Layer |
The reasoning layer is Salesforce's job. The other three are yours. That is why two companies with the same licences can get completely different results: the agent is only as good as the data, automation and security model it is built on top of.
The 6 pillars of Agentforce readiness
We assess Agentforce readiness across six pillars. Each one can block a pilot on its own, so it helps to look at all six before you commit to a use case.
1. Data quality and Data 360
An agent answers from your records. If your data is wrong, the agent is confidently wrong — at scale, and in front of your customers.
Signs you are not ready:
- Duplicate accounts and contacts, with no matching or merge rules enforced
- Key fields (status, product, region, owner) empty or inconsistently filled
- Customer data split between Salesforce and other systems with no unified view
- No clear owner responsible for data quality
What fixing it involves: deduplication, field-level cleanup, validation rules, and deciding which customer data should flow into Data 360 (formerly Data Cloud) so the agent has a complete picture. Read more in our Salesforce data quality checklist for AI agents and our guide to whether you need Data 360 for Agentforce.
2. Actions and automation
Agents act by calling actions — mostly Flows, Apex and APIs. If your automation is tangled, undocumented or built for humans clicking through screens, the agent has nothing safe to call.
Signs you are not ready:
- Legacy Workflow Rules or Process Builder still running alongside Flows
- Critical processes that only work because someone knows which buttons to press
- Flows with no error handling, or Apex with low test coverage
- No inventory of what each automation does and who owns it
What fixing it involves: auditing existing automation, rebuilding the key processes as clean, single-purpose Flows or invocable Apex the agent can call, and adding proper error handling. See our guide to building the actions your Agentforce agents can call.
3. Security, permissions and sharing
An agent runs with a set of permissions. If your sharing model is loose, the agent can surface records a user should never see — and it will do so faster than any person could.
Signs you are not ready:
- Many users with “View All” or “Modify All Data”
- Org-wide defaults set to Public Read/Write for sensitive objects
- Permission sets granted ad hoc, with no recent access review
- No agreed rules on what customer-facing agents are allowed to disclose
What fixing it involves: a least-privilege review, a dedicated permission model for each agent, and testing that agents cannot reach data outside their scope. See our guide to permissions and sharing rules before you deploy an AI agent.
4. Knowledge and unstructured content
For service and internal help use cases, much of what the agent says comes from knowledge articles, PDFs and policy documents. Outdated content produces outdated answers.
Signs you are not ready:
- Knowledge articles with no review dates or owners
- Policies stored in shared drives instead of Salesforce Knowledge or the Agentforce Data Library
- Several conflicting versions of the same answer
- Articles written as internal notes rather than customer-ready answers
What fixing it involves: a content audit, retiring stale articles, assigning owners, and restructuring the most-used answers so an agent can retrieve them accurately. See our guide to knowledge articles: why your agent is only as good as your content.
5. Integrations with systems outside Salesforce
Many valuable agent tasks need data that lives elsewhere: order status in the ERP, billing in a finance system, tickets in another support tool. If that data is not reachable, the agent can only give partial answers.
Signs you are not ready:
- Agents would need data from systems with no API or integration
- Point-to-point integrations that break after releases
- Batch syncs that leave Salesforce hours or days out of date
What fixing it involves: identifying which external data each use case needs, then exposing it through APIs, middleware or Data 360 connectors so the agent can read it in real time.
6. Governance, testing and monitoring
An agent is not a one-time project. It needs an owner, a testing process before changes go live, and monitoring once it is in production.
Signs you are not ready:
- No named business owner for the agent's outcomes
- No test scenarios or acceptance criteria for agent responses
- No plan to review conversations and fix failure patterns
- No sandbox process for changing agents safely
What fixing it involves: a simple operating model — owner, test library, release process and a review cadence using Salesforce's testing and observability tools. See our guide to what a Salesforce partner does in an Agentforce rollout.
Agentforce readiness checklist
Use this checklist for a quick self-assessment. Answer yes or no honestly. Every “no” is a gap to close before or during your pilot.
Data
- Duplicate rules and matching rules are active on Accounts, Contacts and Leads
- The fields your use case depends on are more than 90% complete
- You know which customer data lives outside Salesforce and how to reach it
Actions and automation
- No processes still run on Workflow Rules or Process Builder
- You have an inventory of active Flows and Apex, with owners
- The processes your use case needs can be triggered without a human clicking through screens
Security
- Access to sensitive objects follows least privilege
- User access has been reviewed in the last 12 months
- You have defined what an agent may and may not disclose
Knowledge
- Knowledge articles have owners and review dates
- There is one current answer for each common question
Governance
- A named business owner will be accountable for the agent
- You have a sandbox and test process for changes
- You know which metric will prove the pilot worked
How to read your result: 12 or more “yes” answers means you are close to pilot-ready. Between 7 and 11 means a short readiness phase will save you time and money. Fewer than 7 means start with groundwork, not an agent.
Why Agentforce pilots stall
Most stalled Agentforce projects fail for the same handful of reasons, and almost none of them are about the AI itself.
- No clear use case. The team switched Agentforce on to “do something with AI” and never defined the job the agent should do or the metric it should move.
- The data is not trustworthy. The agent gives inconsistent answers, users lose confidence in the first week, and adoption never recovers.
- There are no safe actions to call. The agent can talk, but every real task depends on a process that only works manually.
- Security surprises late in the project. The security team sees the agent just before launch and blocks it over access concerns.
- No testing or ownership. Nobody owns the agent after go-live, so failure patterns are never fixed.
Every one of these is cheaper to fix before the pilot than during it. Read the full breakdown in our guide to why Agentforce pilots fail.
How to choose your first Agentforce use case
Your first agent should be the one most likely to succeed, not the most ambitious. Score each candidate from 1 to 5 on four factors and start with the highest total.
| Factor | Question to ask | Scores high when… |
|---|---|---|
| Business value | Will this save meaningful time or revenue? | High volume, repetitive work with a clear cost today |
| Data readiness | Is the data the agent needs clean and reachable? | Data lives in Salesforce and is well maintained |
| Action availability | Do the actions already exist as Flows or APIs? | Processes are automated and documented |
| Risk | What happens if the agent gets it wrong? | Errors are low-impact and easy to catch |
Three first pilots that often score well:
- Service case deflection — answering common “where is my order” or “how do I” questions from good knowledge content, with clean handoff to a human
- Inbound lead qualification — asking qualifying questions, enriching the record and routing or booking a meeting
- Internal IT or HR help — lower risk because users are employees, and a good way to build confidence before customer-facing agents
See more examples by department in our guide to the first five Agentforce use cases worth piloting.
Do you need Data 360 for Agentforce?
It depends on the use case. Agents can work with standard Salesforce records, but Data 360 (formerly Data Cloud) is the data layer Salesforce designed to ground agents — especially when they need a unified view of customers across several systems, or need to search large volumes of unstructured content. Salesforce's packaging of Data 360 with Agentforce has changed more than once, so check what your current licences include before you plan the architecture.
For a fuller answer, including when you can start without it, read our guide to whether you need Data 360 for Agentforce.
What does Agentforce cost to run?
The licence price is only one part of the total cost. When you budget for Agentforce, account for four things:
- Usage costs — Salesforce charges for agent activity, and the model has changed since launch, so estimate volume carefully
- Licences and add-ons — the editions, Data 360 capacity and any other products your use case needs
- Readiness work — data cleanup, automation rebuilds and security changes
- Build and run effort — configuring, testing and improving the agent after launch
Readiness work is usually where budgets are most underestimated. Our guide to how Agentforce pricing works breaks down the current model and how to estimate your costs.
What an Agentforce readiness assessment includes
An Agentforce readiness assessment is a short, fixed-scope review that answers one question: can Agentforce deliver value in your org, and what has to change first? At MoreYeahs it takes one to three weeks depending on org size.
| Phase | What happens | You receive |
|---|---|---|
| 1. Discovery | Workshops with business and IT owners to collect candidate use cases and goals | A longlist of use cases |
| 2. Org and data review | Review of data quality, automation, security model, knowledge content and integrations | Scores across the six pillars |
| 3. Use-case shortlist | Each candidate scored for value, readiness, action availability and risk | 2–3 prioritised pilot candidates |
| 4. Gap report | Specific fixes needed for the top use case, with effort estimates | A clear list of what to fix |
| 5. Pilot roadmap | Sequenced plan from readiness work to a live pilot | Timeline, team and cost estimate |
Book an Agentforce Readiness Assessment. Find out in one to three weeks whether Agentforce can deliver value in your org, which use case to start with, and exactly what to fix first. Talk to a Salesforce architect.
From readiness to pilot to scale
Successful Agentforce programmes move in three stages:
- Readiness (1–3 weeks): assess the org, pick the first use case and agree the success metric.
- Pilot (typically 6–10 weeks): fix the gaps for that use case, build and test the agent, and launch to a limited group of users or customers.
- Scale (ongoing): measure results, expand to more channels and use cases, and put governance in place to keep agents accurate as your business changes.
Read how we run each stage in our guide to what a Salesforce partner does in an Agentforce rollout.
