Seven things AI agents should never do with customer data, each with the failure mode, a real incident or study, and the control that enforces the rule.
Updated
Explore how AI can assist customer data operations while keeping access, decisions and accountability understandable. These articles cover agents, the Model Context Protocol, segmentation, private inference and governance. Start with the architecture and permissions an agent needs, then examine the data it may read and the actions it may take. The practical guides discuss review points, audit evidence and failure handling. Pricing discussions separate model inference from the wider platform and operating costs. Use the policy template to name owners and decision boundaries before connecting an agent to live customer data. Follow the source references for product-specific details. These resources support an evaluation of useful operational assistance, with human review where the consequences require it.
Explore this collection
ai
Agentforce pricing explained: $2 per conversation, Flex Credits per action and per-user add-ons, with the break-even maths against running your own LLM.
Updated
How to run AI agents on customer data without exposing it: private inference, least-privilege MCP tools, approval gates, audit logs and a 15-point checklist.
Updated
AI governance for marketing: a ten-section policy template covering approved uses, data rules, human review, EU AI Act Article 50 disclosure, logging and RACI.
Updated
AI marketing operations roadmap for regulated teams: four stages from data and consent to selective autonomy, with prerequisites, controls and honest metrics.
Updated
AI segmentation, step by step: a plain-language request becomes a reviewed, approved, audited segment, with sample audit fields and vendor comparison.
Updated
Explainable AI for marketing and customer decisions: what GDPR Art. 22, the EU AI Act and US credit rules require, which techniques work, and decision logging.
Updated
MCP for Salesforce and HubSpot: what each server exposes, how auth and scopes work, and what an agent can safely read, draft, write and activate.
Updated
On-premise LLM deployment guide: when private inference is justified, open-weight licences, serving stacks, GPU sizing, latency targets and a cost method.
Updated
What an MCP server is, how the Model Context Protocol works, which martech and CRM vendors ship one, the security risks, and a vendor-evaluation checklist.
Updated
Further reading
Published: · Updated:
Publication date recorded for the approved release. Editorial updates do not establish legal approval or a customer result.