AI agents in Salesforce

AI agents that work with your Salesforce processes.

Add AI agents to the sales and service work your team already manages in Salesforce. We configure the information they can use, the actions they can take and when they should ask a person to step in.

Discuss your project
Generated rear view of a woman working at a laptop with a softly blurred Claude interface.
Browse servicesCurrent page: AI agents in Salesforce

What we can deliver

AI agents in Salesforce

01

Sales assistance

Help sales teams research prospects, qualify leads and prepare follow-ups. You choose whether the agent suggests a next step or can update Salesforce itself.

02

Customer service agents

Use your knowledge articles and customer records to answer questions and classify requests. Pass the conversation to your team when the agent cannot resolve it.

03

Agentforce & approved actions

Define the agent’s scope, instructions and permitted actions through Flow, Apex, prompt templates or supported APIs. Validate inputs, identify the customer when needed and enforce record permissions before an action runs. Require confirmation or approval for changes with business consequences.

04

Evaluation and improvement

Test typical conversations, follow-up questions, missing records, misleading instructions and failed actions in a suitable test environment. Check that the agent selects the intended action and respects access rules, not just that its reply sounds plausible. Keep regression cases for later prompt, model or workflow changes.

05

Knowledge & CRM grounding

Choose the knowledge articles, documents and CRM records the agent may use. Review retrieval, freshness and access so answers are based on the right information.

06

Human handover & monitoring

Define when to pass a conversation to a person and what context goes with it. Review failed actions, answer quality and usage after launch to guide improvements.

07

Prompt Builder & staff assistance

Use reusable prompt templates for record summaries, email drafts and field generation where supported. Ground the output in authorised Salesforce data and let users review it in their normal workspace. Keep model-generated suggestions distinct from validated record values.

08

Channels, identity & handover context

Choose the supported employee, web, messaging or voice experience for the use case. Define session context, customer verification and the information passed to a human. Agree what happens outside service hours or when no representative is available, rather than promising an immediate handover.

How we approach it

How we work

We start with one useful workflow and define what success looks like. We review the available data and licensing, configure the agent and its actions, then evaluate the results before a wider rollout.

How we work

Questions worth asking

Can an agent take actions without approval?

Yes, for actions you have authorised it to take. We decide those permissions with you and add review steps for actions that need a person’s approval.

Can the agent connect to other systems?

Yes, where suitable APIs and permissions are available. We can connect Salesforce workflows to your website, mobile app and business tools, or design a broader custom AI solution.

Does every agent need the same Salesforce products?

No. Requirements depend on the agent, channels, data sources and actions. We check Agentforce entitlements, usage charges and whether features such as data libraries require Data 360. An external model connection is a separate architecture decision.

Introducing Agentforce

Watch on YouTube

Insights

Practical guides

Explore all insights

OOB Consulting

What could work better
in your business?

Tell us what you want to build, connect or improve.

Discuss your project