
Zendesk CX Trends 2026: How to Make Customer Context Reviewable
Understand Zendesk's 11,000-plus respondent study and turn the contextual-intelligence themes into a reviewable context receipt.
View details
Salesforce's State of the AI Connected Customer examines how consumers and business buyers describe changing expectations, trust, and AI-enabled service. The study covers more than 16,000 respondents worldwide. Its answers represent perceptions at the time of the survey, not observed service quality or a product test.
The scope card below separates the report's combined population from the more fully disclosed consumer survey.
| Item | Detail |
|---|---|
| Publisher | Salesforce |
| Edition | State of the AI Connected Customer |
| Released | 31 October 2024 |
| Report population | 16,585 consumers and business buyers on the official report page |
| Consumer method | Double-blind YouGov online survey of 15,015 consumers, fielded 16 July–16 August 2024 and weighted to adults in 18 countries |
| Business-buyer boundary | Included in the report total; the accessible methodology note does not expose equivalent field and weighting detail for this subgroup |
| Access | Official landing page; form or regional routing may apply |
Trust matters more as AI becomes more capable.
Salesforce frames trust as a central condition for AI-mediated customer relationships. It does not show that disclosure alone creates trust.
Customers want transparency around agent interactions.
The publisher reports demand for clarity when people interact with AI.
Better context raises both value and risk.
Salesforce connects customer context with more relevant experiences under its report framing.
Stated openness is not adoption evidence.
Survey responses about willingness to use agents describe attitudes under the question wording. They do not prove successful deployment, resolution quality, retention or revenue.
This is Easy AI's interpretation, not a Salesforce finding: before an AI agent handles a customer step, write a small trust contract. Record what the agent identifies itself as, which data it may use, which action it may take, what it must explain, and when a person must accept the case. Replay the same cases after any data or instruction change.
Completed fictional trust contract:
| Field | Entry |
|---|---|
| Customer step | Answer delivery-status questions from the verified order record |
| Disclosure | First response states that an automated assistant is answering from the current order status |
| Data and action | May read order ID, status and last scan; may not change address, cancel or promise a delivery time |
| Explanation | Show the status timestamp and carrier event used for the answer |
| Human acceptance | A named service queue accepts delay disputes, identity mismatch and every customer request for a person |
| Failure and decision | On missing or conflicting status, state the limit and hand off; proceed with archived replay only, no live action |
The contract is fictional and claims no Salesforce or Easy AI outcome.
This is vendor-sponsored, self-reported research. Geography, respondent mix, wording, and field dates affect interpretation. The report does not independently compare vendors, audit deployments, prove causal ROI, establish legal compliance, or demonstrate Easy AI capabilities.
Use the official report page and Salesforce's official research summary and methodology note. Easy AI does not own the study or reproduce its report file.
Does a global sample describe Vietnam specifically? No. Apply a finding to Vietnam only after checking local customer language, channels, consent expectations, and operating constraints.
Is customer willingness enough to deploy an AI agent? No. A team still needs a bounded job, reliable source data, disclosure, testing, correction, and human handoff.
Once those boundaries are clear, define an AI agent, set human-handoff triggers, and document safe identity resolution.

Understand Zendesk's 11,000-plus respondent study and turn the contextual-intelligence themes into a reviewable context receipt.
View details
Use NIST AI 600-1 to map one generative-AI risk into a completed control register while preserving the profile's voluntary scope and limits.
View details
Understand Intercom's 2,470-person customer-service study and use a bounded intent record to test deployment maturity without assuming outcomes.
View details
Read HubSpot's 1,500-plus-marketer survey, then apply its themes with a completed content-journey audit and clear vendor-source limits.
View details
Read Salesforce's 4,050-person survey, then use a completed agent evidence card to scope one sales pilot within clear vendor-source limits.
View details
Use the OECD report's SME adoption findings to complete a one-page pilot brief while preserving G7 evidence limits and official access.
View details