Product Vision

Problem

Small businesses often depend on repeat customers and personal relationships, but follow-up work is easy to postpone or forget. A business owner may remember who bought recently, but not consistently remember who has gone quiet, what they bought, whether a related offer is relevant, or when it is appropriate to contact them again.

This creates a gap between good customer relationships and day-to-day execution.

Dokene exists to make that follow-up process systematic without making it impersonal.

Primary user

The initial target user is a small-business owner or operator who:

• serves customers directly;
• has recurring or relationship-driven sales;
• has limited time for CRM administration;
• may work alone or with a very small team;
• uses WhatsApp heavily for customer communication;
• needs help remembering and prioritizing follow-ups more than a large enterprise CRM.

Core product question

Dokene should help the operator answer:

Who should I contact now, why, and what should I say or do next?

Initial product loop

Customer activity
      ↓
Stored customer + purchase history
      ↓
Follow-up eligibility evaluation
      ↓
Deterministic policy checks
      ↓
AI-assisted Next Best Action / draft
      ↓
Human review and approval
      ↓
Message delivery
      ↓
Delivery/audit history
      ↓
Future follow-up state

Initial capabilities

The MVP should support:

• customer records;
• purchase-history records;
• last-purchase and follow-up timing;
• configurable follow-up intervals or next-follow-up dates;
• consent and do-not-contact state;
• follow-up candidate generation;
• AI-assisted drafting using structured business context;
• message templates;
• manual approval before sending;
• WhatsApp delivery through a provider abstraction;
• message status tracking;
• audit history;
• tenant-aware access control.

Product principles

Reduce cognitive load

The product should remember routine follow-up state so the operator does not have to.

Preserve human judgment

The system may recommend and draft, but the operator remains in control by default.

Be useful before being autonomous

A reliable recommendation queue with manual approval is more valuable than premature full automation.

Respect customer intent

Consent, opt-out, contact policy, and reasonable contact frequency are hard constraints. They are never overridden by AI-generated recommendations.

Prefer explainable actions

A follow-up recommendation should be understandable: for example, because a customer bought a repeat-purchase product 60 days ago, because a configured follow-up date is due, or because a relevant seasonal event is approaching.

Keep provider choice replaceable

Dokene should not become synonymous with one LLM, one WhatsApp intermediary, or one hosting provider.

Differentiation

Dokene is not positioned primarily as an AI chatbot or bulk-marketing tool. Its primary value is relationship follow-up orchestration for small businesses.

Longer-term, this can evolve toward a broader Next Best Action engine that combines:

• customer history;
• timing;
• business rules;
• product relationships;
• special dates;
• previous communication outcomes;
• consent and risk constraints;
• AI-assisted recommendation quality.

Success criteria

Early product success should be measured by outcomes such as:

• fewer overdue follow-ups;
• less manual tracking work;
• useful drafts accepted with little editing;
• low accidental-contact and duplicate-send rates;
• high operator trust in recommendations;
• reliable tenant and security isolation;
• clear auditability of every externally visible action.

Raw message volume is not a primary success metric.

Explicit non-goals for early releases

• replacing mature enterprise CRM suites;
• building omnichannel support before WhatsApp is reliable;
• unrestricted autonomous outbound campaigns;
• building microservices for speculative scale;
• making the LLM the source of truth for customer state;
• storing provider secrets in prompts, logs, or ordinary application tables;
• optimizing for massive message throughput before product-market evidence exists.