The engine
One profile. One decision at a time.
Every module writes to the same customer record, and the retention engine decides what happens next.
Recover
Turn saved items and sold-out stock into orders.
Retain
Keep customers with subscriptions, flows, segments and proof.
Built on Shopify's native primitives. No theme hacks, no ScriptTags.
By outcome
Same engine, same price. Each solution shows the loop pointed at one outcome.
Loyalty & referrals
Subscriptions & pre-order
Dated, sourced comparisons. Free migration from any of these during private beta.
How it works
Ayra is a Shopify retention platform built around one loop: it collects signals from every module and from Shopify, resolves them to a single customer profile, detects retention opportunities, decides the next best action, executes it through the right module and channel, attributes the revenue, and learns from the response. The modules are the hands; the loop is the product.
Today: separate apps Ayra: one loop
One person: Hira
Wishlisted + restocked
expires in 72h
Restock alert
WhatsApp · +200 points
Held · referral → next week
Skipped · email · duplicate
8 of 8 Order #4873 is credited once, to Back in Stock and Loyalty. Her response feeds back into scoring, and the loop starts again.
01 The problem in one picture
Today
Ayra
02 Three layers
Layer 1
Top
What the shopper sees. Loyalty, referrals, reviews, wishlist, back-in-stock, pre-order, subscriptions, flows, account hub.
Layer 2
Middle
Signals, identity resolution, Customer 360, lifecycle stage, segments, timeline, integrations.
Layer 3
Foundation
Opportunities → eligibility → scoring → prioritisation → conflict resolution → next best action → automation.
Most apps are Layer 1 only. Suites bundle several Layer 1 products and call it a platform. Ayra is built from Layer 2 up.
03 The loop, step by step
Every module and Shopify itself emits events: order placed, delivered, item wishlisted, back in stock, review left, referral sent, subscription skipped, support ticket opened. Signals are facts, not decisions.
Example Item wishlisted → viewed again 3 days later → restocked.
Signals are resolved to one person across email, phone, customer ID and device, so a guest wishlist and a logged-in order are the same customer.
Example Guest wishlist merged on login.
One profile: wallet balance, tier, LTV, lifecycle stage, intent score, subscription state, review history, open support issues, timeline.
Example Hira: Gold tier, 9 orders, LTV $1,284, intent high, no open tickets.
Customer 360
The engine watches for patterns worth acting on: a VIP's wishlisted item back in stock, a subscriber who skipped twice, a first-time buyer 20 days post-delivery with no review. Each becomes an Opportunity with a type, a customer, evidence and an expiry.
Example VIP_RESTOCK · Hira · wishlisted + restocked · P1 · expires in 72h.
Open opportunities
| Customer | Type | Evidence | Prio | Expires |
|---|---|---|---|---|
| Hira | VIP_RESTOCK | Wishlisted + restocked | P1 | 72h |
| Daniel | SUB_SKIP_X2 | Skipped 2 cycles | P2 | 5d |
| Mei | FIRST_REVIEW | Delivered 20d, no review | P3 | 10d |
| Omar | REFERRAL_PUSH | NPS 9, 3 orders | P4 | 30d |
Each opportunity is checked for eligibility, scored, prioritised (P0–P4), and run through conflict resolution so one customer never receives competing actions in the same window. The output is a single next best action — which can be NO_ACTION.
Example Win-back and referral push same day → referral scheduled for next week, win-back sent.
The action runs through the right module and channel: a points bonus, a store-credit grant, a back-in-stock alert with an incentive, a review request, a flow. Frequency caps and quiet hours apply.
Example Restock alert for a VIP → sent with a 200-point bonus, outside quiet hours, on her preferred channel.
Revenue from the resulting order is attributed to the opportunity and module that triggered it — once, not by every app that touched the customer. Response and non-response feed back into scoring.
Example Order #4873 → attributed to Back in Stock + Loyalty, not to the welcome flow that ran three weeks earlier.
Attributed revenue
Sums to 100%. Each order credited once.
04 The guards
Every outbound action passes through Journey Guard, which enforces frequency caps, quiet hours and channel limits, and resolves conflicts between modules. 'Do nothing' is a valid outcome.
Example A win-back offer and a referral nudge both qualify on Tuesday. One is sent, the other is scheduled, and the timeline says why.
The single highest-value eligible action for a customer right now, chosen from all open opportunities after scoring and conflict resolution.
Example Three open opportunities for one customer resolve to a single restock alert with a points bonus.
One customer balance that unifies points, store credit, vouchers and deposits, with separate ledgers underneath so every change is reversible and audited.
Example A $10 redemption draws from store credit first, then points, and both ledgers record the split.
Scores purchase intent from wishlist, browse, back-in-stock and price-drop signals so recovery actions go to customers likely to buy.
Example Wishlisted twice and viewed three times this week: the restock alert goes out before the general announcement.
Prevents stacking and over-discounting by checking what a customer has already received before granting a new incentive.
Example A customer who received 15% off last week does not get a second discount on top of it.
Intervenes before a subscription cancellation with a pause, skip or swap offer, based on the customer's history.
Example A subscriber about to cancel after two skips is offered a one-month pause instead.
Detects referral and loyalty fraud (self-referrals, refund farming, duplicate accounts) and claws back rewards on refunded orders.
Example A referral reward is reversed when the referred order is refunded inside the return window.
Suppresses review requests and promotions for customers with open support issues until the issue is resolved.
Example A review request is held while a delivery complaint is open, then sent three days after it is resolved.
Attributes each order to at most one retention action, so module reports never double-count the same revenue.
Example An order touched by a flow, a referral and a loyalty redemption is credited once, to the action that triggered it.
The connected model of customers, signals, opportunities, actions and outcomes that every module reads from and writes to.
Example A review left today is visible to loyalty, referrals and segments the moment it lands.
05 Rules now, prediction later
Opportunity types, eligibility and scoring are explicit, inspectable rules. Every decision has a visible reason in the customer timeline.
Churn, repeat-purchase and LTV prediction, best channel and send time, discount sensitivity, layered onto the same decision infrastructure.
The infrastructure comes first because a prediction with nowhere to act is a report.
Everything above is included in every plan06 A day in the loop
Hira, VIP · 9 orders, wishlists one item on Monday. Here is everything the engine does with that, and why.
Hira
VIP · 9 orders
Signal
Wishlists the Linen Overshirt
Signal recorded on her profile. Intent score moves up.
Opportunity
Item restocks. Opportunity created, P1
Type VIP_RESTOCK, evidence: wishlisted + restocked, expires in 72h.
Decide
Journey Guard checks pass
No sends in the last 48h, no open support ticket, inside send window.
Execute
Alert sent with a 200-point bonus via WhatsApp
Her preferred channel. Email alert suppressed to avoid a duplicate.
Attribute
Order placed, attributed to Back in Stock + Loyalty
Credited once. Points bonus applied from the Smart Wallet.
Decide
Referral opportunity created, P3, scheduled for next week
Frequency cap: one promotional send per 7 days. Nothing goes out today.
Learn
Timeline shows the reason for every step
Each action, hold and skip has a recorded reason you can open.
07 For developers and agencies
Store credit, discounts, selling plans, metaobjects, new customer accounts, Flow. No theme mutation, no ScriptTags.
Developer docsEvery signal and action is a typed event. Flow triggers available.
Developer docsThe ayra.* namespace is readable by themes and other apps.
Developer docsTheme app extensions, checkout, customer account, POS and admin UI extensions.
Developer docs08 FAQ
Not at launch. It is a rules-based decision system with explicit, inspectable logic. Prediction models are added on top of the same infrastructure later.
Yes. Every opportunity and action has a reason recorded in the customer timeline.
Yes. Modules work individually; the engine coordinates them when more than one is active.
Only through flows and actions you've enabled, inside the frequency caps you set.
Journey Guard is the gate every outbound action passes through before it reaches a customer. It applies your frequency caps, quiet hours and channel limits, settles conflicts between modules, and can decide to send nothing. The full definition is in the glossary above.
Next best action is the one highest-value action a customer is eligible for right now, picked from all of their open opportunities once each has been scored and checked for conflicts. The full definition is in the glossary above.
At launch Ayra runs its own flows; integrations that let external tools consume Ayra signals follow after the private beta.