Skip to content

Founder-led. Limited engagements.Talk to Us →

[Rewrite] Agentforce Readiness Assessment: 12 Checks Before You Build

Twelve checks to run before you build an Agentforce agent, from edition and data quality to permissions, testing and cost, with a simple score to decide whether to pilot now or fix something first.

By 8 min readSalesforce · Agentforce · AI Readiness · Governance
[Rewrite] Agentforce Readiness Assessment: 12 Checks Before You Build
In this article
  1. Why assess before you build
  2. The 12-point Agentforce readiness checklist
  3. 1. Your edition and licences support Agentforce
  4. 2. You have one narrow, repetitive use case
  5. 3. Someone owns it and has a baseline
  6. 4. Your data can already answer the question
  7. 5. Each fact has one source of truth, and knowledge is current
  8. 6. The agent user has least-privilege access
  9. 7. Sensitive data is mapped, and you know what the Trust Layer does
  10. 8. Rules that must always hold live in actions, not instructions
  11. 9. The actions exist, and integrations fail cleanly
  12. 10. You have a test set and a regression routine
  13. 11. You have a usage model before the pilot
  14. 12. Handoffs and feedback are designed
  15. Score yourself
  16. India: edition and data residency checks
  17. What the assessment should leave you with
  18. FAQs
  19. How long does an Agentforce readiness assessment take?
  20. Is my Salesforce org ready for Agentforce if our data isn't clean?
  21. Do you need production access for the assessment?
  22. What is the most common problem a readiness check finds?
  23. Does a readiness assessment commit us to building an agent?

An Agentforce readiness assessment checks whether your Salesforce org can support an agent before you spend money building one. It covers twelve things: edition and licences, one narrow use case, an owner with a baseline metric, data quality, permissions, sensitive data, where controls live, actions, testing, cost, and handoffs. If most of these hold, run a pilot. If two or three fail, fix those first, because the agent will only make them visible faster.

Why assess before you build

Salesforce Agentforce makes it quick to build something that looks finished. That demo usually runs as an admin, on clean sandbox data, with polite prompts. Pilots that stall rarely fail because of the model. They fail because the data can't answer the question, nobody owns the result, or permissions are too loose for customers. This checklist finds those problems while they are cheap to fix.

The 12-point Agentforce readiness checklist

1. Your edition and licences support Agentforce

Start with the commercial basics. Salesforce Foundations, the free bundle that includes Agentforce Builder and Prompt Builder according to the Agentforce pricing page, is offered to Sales Cloud and Service Cloud customers on Enterprise Edition or higher. Professional Edition orgs need an upgrade conversation before any build conversation. Also confirm that Agentforce is turned on in Setup, which Data 360 (Data Cloud) features your use case depends on, and whether you have Flex Credits or a per-conversation agreement. [CHECK: current Agentforce edition availability on the Salesforce Help "Available in" line]

2. You have one narrow, repetitive use case

"Help customers with anything" is not a use case. "Answer where-is-my-order questions for recent orders, and hand off anything involving a refund" is. The test: can you write twenty real requests and the correct outcome for each? If the list keeps spreading, you have a theme, not a use case. Also split pairs hiding as one: "billing questions" is usually explaining an invoice (read-only) and changing a payment method (a transaction), with different risks.

3. Someone owns it and has a baseline

A pilot needs a named business owner who can say "ship it" or "stop", and one metric you can measure today, before the agent exists: handling time for one case type, or accuracy against a reviewed test set. Without a baseline, the pilot can't pass or fail.

4. Your data can already answer the question

Give ten typical requests to someone who has never seen your org and ask them to answer using only Salesforce. If they keep saying "there are two accounts with this name", the agent will hit the same walls in front of customers. You don't need a clean org, only a reliable slice: populated fields, one record per customer or order, and status values that mean what they say. Salesforce's own Agentforce troubleshooting module lists Data 360 sync delays, paused data streams, null or outdated fields and broken mappings as common causes of bad agent answers.

5. Each fact has one source of truth, and knowledge is current

If order status lives in both Salesforce and an ERP (SAP, NetSuite, Tally, Zoho Books) and they disagree, the agent can't know which is right. Pick one source per fact. An outdated knowledge article is worse than none, because the agent quotes it with confidence.

6. The agent user has least-privilege access

Built and tested as an admin, everything works. Under a properly restricted agent user, actions start returning empty results. Salesforce's guidance is to give the agent user the minimum object permissions for each object it touches and to always debug a flow by running it as the agent user. Don't copy a support rep's permission set; the agent then inherits access it has no use for.

7. Sensitive data is mapped, and you know what the Trust Layer does

List the fields holding personal or regulated data and decide whether the agent needs them at all. The Einstein Trust Layer adds data masking, zero data retention with the model provider, toxicity detection and grounding that respects permissions. It does not decide what the agent should be allowed to see in the first place. That is your permission design from check 6.

8. Rules that must always hold live in actions, not instructions

Salesforce's Agentforce guide is explicit: instructions go to the LLM, are non-deterministic, and do not replace coded business rules in the action. "Verify the customer before sharing order details" written as an instruction is a request. Written as an action with a pass or fail result, it is a control. The same applies to approvals. In our open-source case resolution POC, built on synthetic data, refunds wait for human approval rather than relying on the prompt.

9. The actions exist, and integrations fail cleanly

Mark every action the use case needs as exists, needs changes, or must be built. Agents mostly call Flows and Apex, so this is where the real effort usually hides.

  • Screen flows that assume a record page or a confirmation screen won't work as agent actions unchanged.
  • Silent failures leave the agent to improvise, and it may say something happened when it didn't.
  • Descriptions matter: the engine picks actions by name, description and inputs. Salesforce recommends 15 or fewer actions per subagent to keep latency down.
  • External calls need authentication, rate limits and response times checked. Multi-second calls are slow in a live chat.

If the actions don't exist yet, the pilot is an automation project with an agent attached. Scope and budget it that way.

10. You have a test set and a regression routine

Build a test set from real cases, chats or emails, each with the expected outcome. Agentforce's Testing Center takes test cases as a CSV with the utterance, expected subagent, expected actions and expected response, so tests can be re-run after every change. Include messy inputs: two questions in one message, missing details, attempts to pull the agent off-topic. Our AgentOps POC shows one way to turn those runs into a release gate.

11. You have a usage model before the pilot

Price usage before anyone falls in love with the demo. On Salesforce's pricing page (checked 17 November 2026), Flex Credits cost $500 per 100,000, a standard Agentforce action uses 20 credits (about $0.10), a voice action uses 30, and customer-facing agents can instead be priced at $2 per conversation. Estimate conversations per day and actions per conversation, compare that with the value of the work, and move routine steps to a plain Flow. Our Agentforce pricing breakdown goes deeper; confirm your contract rates with your account team.

12. Handoffs and feedback are designed

Decide what context travels with a handoff to a person; an escalation that makes the customer start again is worse than no agent. Name who reviews agent conversations each week, and give frontline staff a simple way to report bad answers.

Score yourself

Score each check 0 (not true), 1 (partly) or 2 (clearly true), for a maximum of 24.

ScoreWhat it usually meansNext step
20–24, no zerosReady for a scoped pilotRun an Agentforce proof of concept on the riskiest assumption
14–19CloseFix the low scores first; they are your plan
Any zero on checks 1, 3, 4 or 6Not yetFix that before you configure a single subagent

These thresholds are a practical reading, not a scientific model. A self-score also misses what you don't know to look for, such as a Flow that quietly depends on a user's profile.

India: edition and data residency checks

Two questions come up with Indian orgs.

Edition. Check your edition under Setup → Company Information before scoping anything. If the org is on Professional Edition, check 1 ends the conversation until you have priced an upgrade.

Data residency. Salesforce runs Hyperforce in India. Its Hyperforce FAQ says customer data is generally stored in the country where the org is located, provided the services you use are available in that Hyperforce country, and warns that some products have components running elsewhere. So the audit question is not "are we on Hyperforce India?" but "is every service this agent uses, including Data 360 and the model calls, available in the India region?" [CHECK: whether Agentforce, Data 360 and LLM inference run in the Hyperforce India region for India-hosted orgs] Put the answer next to your obligations under India's Digital Personal Data Protection (DPDP) Act before any personal data reaches the agent.

What the assessment should leave you with

A ranked use-case list with a first pilot, a data gap list (field, issue, fix, owner), a proposed agent permission set, an action inventory, a starter test set, a usage model to price with Salesforce, and a rollout plan. Then comes a scoped pilot, and Agentforce implementation for whatever the pilot proves.

FAQs

How long does an Agentforce readiness assessment take?

It depends on the org and how many use cases are on the table. A focused assessment of a few candidates usually takes days to a couple of weeks, not months. [MEASURE: actual duration of our completed readiness audits]

Is my Salesforce org ready for Agentforce if our data isn't clean?

Probably, if the slice your first use case touches is reliable. You need the specific objects and fields that use case reads to be populated and current, not a clean org. Targeted cleanup of that slice is often enough for a pilot.

Do you need production access for the assessment?

Read access to production metadata and representative data helps, because sandboxes are usually cleaner than production. Without it, we use a full or partial copy sandbox and note what that hides.

What is the most common problem a readiness check finds?

Actions. Existing Flows often weren't built to be called by an agent, so they need clear inputs and outputs and proper error handling first.

Does a readiness assessment commit us to building an agent?

No. A good assessment can conclude that a Flow, a data cleanup or waiting is the better move. That is cheaper to learn now than after a build.

If you want this checklist applied to your org, with a clear recommendation on what to pilot first and what to leave alone, see our Agentforce Readiness Audit.

Keep reading

Agentforce Readiness Audit: What We Check and What Usually Breaks

The Agentforce readiness audit checklist we use: use cases, data, permissions, actions, testing, cost, and change management, plus what usually breaks.

Salesforce Agentforce: What It Does, Costs, and When It Fits

A plain guide to Salesforce Agentforce in 2026: how an agent routes, plans and acts, what Flex Credits, conversations and per-user licences cost, and where a Flow or a custom agent is the better answer.

5 Signs Your Org Is Ready for an Agentforce Pilot

Five practical signs your Salesforce org is ready for an Agentforce pilot, the red flags that mean not yet, and a quick self-assessment to score yourself.

Subscribe to the edition

One useful systems note a week from real Salesforce and automation builds. Unsubscribe in one click.

Letters

Letters to the editor

No letters yet. Questions and counterpoints are welcome.