How Daisy Works

A three-plane architecture designed for Shopify — with policy guardrails, explainability, and progressive trust at every layer.

Three-Plane Architecture

Control Plane

Merchant-facing admin app

  • Agent configuration & feature flags
  • Dashboards & analytics
  • Rule overrides & approvals
  • Human-in-the-loop actions
  • Billing management

Decision Plane

Daisy's autonomous engine

  • Price & promo recommendations
  • CX next-best-action
  • Explanation generation
  • Policy checks & confidence gating
  • Assortment & bundling suggestions

Execution Plane

Pushes decisions into Shopify

  • GraphQL Admin API mutations
  • Checkout UI extensions
  • Shopify Functions (discounts, validation)
  • Customer account extensions
  • App proxy storefront endpoints

Shopify-Native Integration

Daisy executes decisions through Shopify's own extension points — no third-party scripts or hacks.

Checkout Extensions

UI extensions for upsell, reassurance, promo banners, and dynamic messaging at checkout.

Customer Account

Reorder suggestions, order insights, support workflows, and retention nudges in the customer portal.

Shopify Functions

Discount logic, cart validation, and shipping customization running inside Shopify's runtime.

App Proxy

Low-latency storefront endpoints for recommendations, search re-ranking, and conversational bootstrap.

Admin App

Embedded admin app with agent config, dashboards, approval queues, and billing — via App Bridge.

Theme Extensions

App blocks and snippets for Liquid-compatible storefront injection and dynamic content.

Progressive Trust

Start by watching. Graduate to recommendations. Upgrade to full autonomy — on your timeline.

Mode 1

Observe

Agents watch and learn from your data. No actions taken. Perfect for building confidence.

Mode 2

Recommend

Agents surface recommendations in your admin dashboard for human review and approval.

Mode 3

Auto

Agents execute decisions autonomously within your policy guardrails. Full visibility retained.

Trust & Governance

Every agent decision passes through a policy engine with configurable guardrails.

Policy Engine

Configurable rules that block unsafe pricing, margin violations, and policy breaches.

Confidence Gating

Low-confidence decisions are held for review. High-confidence actions execute automatically.

Human-in-the-Loop

Approval workflows route high-impact decisions to the right person.

Explainability

Every decision includes a human-readable explanation and full audit trail.