# Agentic CRM

> VI: https://5ac.vn/wiki/agentic-crm/ · EN: https://5ac.vn/en/wiki/agentic-crm/ · lang: en · spec_as_of 2026-10-19 (spec no formal spec) · cluster C10

A CRM where AI agents read customer context and take controlled actions along the pipeline — propose next steps, trigger follow-ups, draft replies, execute behind human approval.

---
## Definition

Agentic CRM is a CRM system where AI agents do more than log customer notes: they read customer context, propose the next action, trigger follow-ups, draft replies and coordinate the sales step ahead under human approval. The boundary against generic "AI CRM" is the action loop. "AI CRM" is the umbrella for any AI feature bolted onto a CRM — email drafting, call summaries. "Agentic" claims more: observe → infer → propose → execute-with-approval → record, where the agent takes coordination work rather than only making the salesperson type faster.

## Mechanism

The data foundation is four layers preserved inside one system: **customer history** — channel, past requests, purchases, unresolved issues; **opportunity state** — pipeline stage, missing documents, the risk that could sink the deal; **action cadence** — the next task, its deadline, its named owner; **access and handoff rights** — who may view, edit, send and close. Miss one layer and AI just accelerates the fragmentation. The 90-day rollout described in the 5ac series: days 1–30, standardize lead sources, sales stages, required fields and viewing rights; days 31–60, agents run in proposal mode — summarizing leads, reminding owners, drafting messages — while sales still approves every send; days 61–90, controlled execution opens for low-risk work such as creating tasks, updating confirmed statuses and reporting the pipeline, while quotes, discounts and delivery commitments stay behind human approval.

## Application context

For Vietnamese SMBs the timing signal is national: the 2026–2030 SME digitalization program targets support for 300,000–500,000 businesses adopting AI, and market research cited across the 5ac series puts the annual saving from replacing the right repetitive work at roughly USD 84,000 — a salesperson whose agents summarize, remind and draft reclaims 5–7 hours a week. Priority use cases follow revenue: answering new leads before they cool, reviving overdue opportunities with a recap of the last exchange, and pipeline reports nobody has to key in by hand. Three minimum control points before any automation: every new lead enters one shared pipeline; every opportunity has an owner and a next-action date; every file sent attaches to the customer record. The standing risks are equally concrete — wrong data produces wrong suggestions, loose permissions produce wrong actions, and without logs nobody can say where the error started.

## Sources fetched

- [Agentic CRM for Vietnamese SMBs: Digital Transformation 2026](/en/blog/agentic-crm-smb-vietnam-digital-transformation-2026/) — action loop, 90-day rollout, 300k–500k program, USD 84k figure — accessed 2026-10-19
- [AI CRM Is a Customer Context Layer for Vietnamese SMBs](/en/blog/ai-crm-customer-context-layer-vietnam-smb/) — four data layers, task-layer vs context-layer boundary — accessed 2026-10-19
- [AI CRM: From One-Off Prompts to Stateful Sales Pipelines](/en/blog/ai-crm-sales-workflows-from-prompts-to-pipeline/) — stateful pipeline, ownership and next-action discipline — accessed 2026-10-19

## Related

- [AI Agent](/en/wiki/ai-agent/)
- [Agent governance](/en/wiki/agent-governance/)
- [Agent memory](/en/wiki/agent-memory/)