Three Agents — One Suite — Live on Shopify

Beyond Automation.

This is Reasoning.

RecoverAIO deploys three specialized Claude Sonnet agents — Cart Recovery, Subscription Save, Win-Back — each reasoning independently, adapting to customer behavior, and explaining every decision. Not a timer. An agent.

// scanCartOpportunity — live trace
AGENT ACTIVE
behavioral signals — emailHash: a7f3c2
signal_typeemail_ignored × 2
prior_offer_pct10%
order_value$340.00
hours_elapsed26h
customer_typefirst_time

claude-sonnet-4-6 — decision output
offer: 10% offer: 15% + urgent

"Two prior ignores at 10%. First-time customer with high cart value. Escalating offer to 15% with urgent subject line — customer shows intent, needs stronger signal to convert."

CONF 0.91
specialized agents

Three Agents.
Three Domains.

Each agent has its own system prompt, reasoning framework, and domain expertise. A subscription dunning agent is not a cart recovery agent. Specialization is in the code.

01 / CART_RECOVERY
Cart Recovery Agent
abandoned_checkout

Evaluates every abandoned Shopify checkout. Reasons about cart value, item count, customer order history, time elapsed, and behavioral signals. Decides urgency, offer size, and whether to act.

cart_value · item_count · hours_elapsed
customer_order_count · is_first_time
behavioral_signal_history
recommended_offer · urgency_level
02 / SUBSCRIPTION_SAVE
Subscription Save Agent
payment_failure

Assesses dunning strategy for failed subscription charges. Knows the difference between a first failure and a fourth. Decides between card update requests, pause offers, discounts, and HITL escalation.

plan_value · retry_count · failure_reason
subscriber_tenure · order_history
dunning_strategy · pause_offer
HITL_escalation_threshold
03 / WIN_BACK
Win-Back Agent
lapsed_customer

Scores lapsed customers by lifetime value, average order size, days inactive, and top product categories. Decides whether re-engagement is worth attempting and what offer will convert.

days_since_last_order · lifetime_value
avg_order_value · top_categories
reengagement_score · suggested_offer
suppress_if_recently_recovered
behavioral signal system

The Agent Remembers.

Every email event is stored as a CustomerBehaviorSignal. Before any agent makes a decision, it reads the customer's full history and adapts its approach. Prior ignores increase offers. Prior bounces suppress sends. Prior recoveries inform confidence.

This is the feedback loop that separates RecoverAIO from every rule-based tool in the category. Rules repeat the same action. Agents learn from outcomes and change.

1 prior ignore → +5% discount boost
2+ prior ignores → +10% boost + urgent tone
2+ bounces → auto-suppress, no send
recently recovered → monitor-only, 90 days
Signal Type Trigger Agent Adaptation
ignored No open after 48h Boost next offer
opened Email opened Increase confidence
clicked Link clicked Mark resolved
bounced Delivery failed Suppress after ×2
recovered Purchase made 90-day monitor
confidence threshold

The Agent Knows When to Stop.

When confidence falls below 70%, RecoverAIO doesn't guess. The action is escalated to your Queue with the agent's full reasoning for your review. You approve, modify, or dismiss it.

HITL escalation is a feature, not a fallback. It means the agents only act when they're sure — and hand off to you when they're not. No tool in this category has this.

confidence ≥ 0.70 → agent acts autonomously
confidence < 0.70 → escalated to HITL
2+ bounces → suppressed, no LLM call
// hitl_queue — awaiting review
action#3f8a2c · cart_recovery CONF 0.61

"New customer, first order $89. Only 2 hours elapsed — insufficient signal history to act confidently. Manual review recommended."

action#9b4e1a · win_back CONF 0.58

"Lapsed 112 days, low LTV ($140). Win-back probability marginal. Agent unsure if offer spend is justified."