Work with us
AI tools / AI inside business systems (CRM, ERP, HR) / Salesforce Einstein / Agentforce

How to govern Salesforce Einstein / Agentforce

How to govern Salesforce Einstein and Agentforce: AI that drafts, scores and increasingly acts on customer records inside your CRM.

Salesforce · Reviewed when the vendor ships a capability change or a customer declares the tool

How Salesforce Einstein / Agentforce is used

We see six common ways businesses use Salesforce Einstein / Agentforce, and we govern each one on its own terms. If your team uses it differently, that is fine: you describe the use in your own words and Certrak works out the governance from what it actually does.

General productivity assistance

Consequence to people: Low

Drafting, summarising, brainstorming, internal Q&A - day-to-day help with internal work

Internal informationEvery output reviewedInternal use only
  • Review rhythm: Annual review
  • Governance is mostly about: Safe everyday use; Checking what it produces; What it may be given

Customer communications drafting

Consequence to people: Medium

Drafting customer emails, support replies, or other outbound communications - reviewed before sending

Personal informationEvery output reviewedCustomers, no decisions about them
  • Review rhythm: Quarterly review
  • Governance is mostly about: Safe everyday use; Checking what it produces; What it may be given

Sales lead scoring and prioritisation

Consequence to people: Medium

Ranking leads, suggesting next actions, prioritising sales effort

Personal informationEvery output reviewedCustomers, no decisions about them
  • Review rhythm: Quarterly review
  • Governance is mostly about: Safe everyday use; A person stays accountable; What it may be given

Customer segmentation and personalisation

Consequence to people: Medium

Grouping customers for targeted treatment, personalisation, or recommendations

Personal informationMost outputs reviewedCustomers, no decisions about them
  • Review rhythm: Quarterly review
  • Governance is mostly about: Safe everyday use; A person stays accountable; What it may be given

Customer-facing chatbot or automated support

Consequence to people: High

Interacting with customers directly, without a person reviewing each response

Personal informationNo human reviewCustomers, no decisions about themSends messages outside the businessActs on its own
  • Review rhythm: Monthly check-in
  • Governance is mostly about: A person stays accountable; Safe everyday use; What it may be given

Autonomous task agent

Consequence to people: High

General-purpose AI agent that takes actions across your tools and systems - email, calendar, messaging, files

Personal informationMost outputs reviewedInternal use onlyChanges records in other systemsActs on its own
  • Review rhythm: Monthly check-in
  • Governance is mostly about: What it may be given; Safe everyday use; A person stays accountable

Governing Salesforce Einstein / Agentforce in practice

  1. Name one person who owns Salesforce Einstein / Agentforce in your business. Governance starts with a name.
  2. Write down what Salesforce Einstein / Agentforce is used for. Certrak recognises 6 common uses: General productivity assistance, Customer communications drafting, Sales lead scoring and prioritisation, Customer segmentation and personalisation, Customer-facing chatbot or automated support and Autonomous task agent. Yours may differ; write down yours.
  3. Decide what Salesforce Einstein / Agentforce may and may not be given. Personal information is in scope for some uses; write the boundary down before it is needed.
  4. There is no human review of outputs. Decide whether that is acceptable for each use, and what would make you change it.
  5. Salesforce Einstein / Agentforce can send messages outside the business and create, change, or delete records in another system without a person approving. Record which of these you allow, and who can change that.
  6. Look at this again every month, and sooner on any of these: industry incidents, vendor model updates, autonomy level changes and data classification changes.