Service · Two to four weeks · Fixed scope

Get ready for AI agents before one meets your customers

The Agentforce Readiness Sprint is a short, fixed-scope assessment. We look at your data and permissions the way an agent would, pick the use cases worth doing first, and design the guardrails before anything goes live.

You know this one

The demo was impressive. The agent answered every question, updated the record and booked the follow-up, all in a clean demo org with 40 perfect accounts.

Then somebody asks what happens when it runs on yours. The org with three records for the same customer. The org where “Status” means four different things depending on who filled it in. The org where the integration user can, for historical reasons, delete anything.

A person works around all that without thinking. An agent takes it at face value, at speed, with your name on the reply.

An Agentforce readiness assessment that ends in a plan

Most readiness work stops at a slide that says “improve data quality.” Ours ends with a short list of things to do, in order, and a pilot you could start.

We cover four areas:

  • Data. Would an agent find one record per customer, fields that mean what they say, and enough written context to ground an answer?
  • Access. What can the agent’s user see and change? Usually it’s more than anyone intended.
  • Use cases. Of everything an agent could do, which two or three are worth doing first: frequent, low-risk, easy to check?
  • Guardrails. Which actions run alone, which need a dry run, which wait for a person, and how you’ll know afterwards what the agent did.

What you get

  • A readiness scorecard: data and access findings, each with the records or permissions behind it.
  • A use-case short list: two or three candidates, each with what it needs, what could go wrong, and how you’d measure it.
  • An action catalog: the specific actions each use case needs, and which already exist in your org.
  • A guardrail design: what runs unattended, what previews first, what needs approval, and what gets logged.
  • A pilot plan: scope, audience, success measure and a stop condition, so the pilot can end as well as start.
  • A read-out with the people who’ll own it.

How it runs

  1. Week 1: data and access. We run a data health scan and review the permissions of the users an agent would run as. If you’ve had a Data Health Check recently, we start from that and the Sprint is shorter.
  2. Week 2: use cases. A working session with the people who do the job today. We list candidates, score them, and pick.
  3. Week 3: actions and guardrails. For each chosen use case, the actions, the checks and the approvals.
  4. Week 4: pilot plan and read-out.

Smaller orgs with one clear use case finish in two weeks.

What we need from you

  • An admin for about four hours a week.
  • Two or three people who do the work the agent would help with, for one 90-minute session.
  • Someone who can say yes to a pilot.
  • Read access to the objects and permission setup in scope.

What it isn’t

  • It isn’t the build. The Sprint ends with a plan. Building the agent is separate work, which we can do with you or hand to your team or partner.
  • It isn’t a licensing review. We don’t advise on which Salesforce products to buy. Ask your account team, then ask them again in writing.
  • It won’t promise what a beta will do. A lot of what was shown at Dreamforce is beta or pilot. We plan around what’s generally available and say plainly where something isn’t.
  • It isn’t a verdict on AI. If the honest answer is “fix these two things first and come back in a quarter,” that’s what the report says.

The apps that help, if you want them

  • Sounding runs the data health scan and scores agent readiness.
  • Sluice gets you to one record per customer, with an undo.
  • Wake keeps a field history, so there’s a record of what changed and a way to restore it.
  • Watchbill shows what a user, including the agent’s user, can do and what grants it.

None is required. See the full picture on the agent readiness page.

Questions people ask

What does “Agentforce ready” mean in practice?

It means an agent can find the right record, trust what the fields say, act only where it should, and leave a trail. In practice that’s four checks: duplicates, field meaning and completeness, the permissions of the agent’s user, and a log of changes. Most orgs pass one or two on day one.

How clean does our data need to be before we start with AI agents?

Clean enough for the first use case, not perfect everywhere. An agent answering order-status questions needs reliable orders and contacts. It doesn’t need your lead data from 2019 to be tidy. The Sprint narrows the cleanup to what the pilot touches.

What should our first Agentforce use case be?

Something frequent, low-risk and easy to check. Answering a common question from data you trust is a good first use case. Changing records, sending things to customers or anything involving money makes a better second or third. The working session in week two is where you choose.

How do we know what the agent changed?

Decide that before the pilot. The platform has its own tracing for what an agent did and why. For what a record looked like before and after, you need field history on the fields that matter. The guardrail design in the Sprint covers both.

We haven’t bought Agentforce yet. Is this too early?

No, and it may be the cheapest time to do it. The data and access findings are useful whichever agent you end up with, and the use-case short list gives you something concrete to take into a vendor conversation.

Find out what an agent would trip over

Thirty minutes to talk through where you are. Or start with the checklist and see how far you get on your own.