return
Return Intelligence — Documentation
Return & Refund Abuse Intelligence is a Shopify embedded app that scores every return, refund request and new order for abuse risk from 0–100, shows the exact signals behind each score, and recommends an action — while your team makes every final decision.
It estimates risk. It does not determine guilt. There is no automatic decline at any score, at any setting.
- Type: Shopify embedded admin app (Polaris + App Bridge)
- Works alongside: Shopify self-serve returns and third-party return portals — nothing in your customer-facing flow changes
- Shopify API version: 2026-04
- Current version: v1.2
Contents
- Installation and first run
- The free 12-month leakage audit
- How scoring works
- The nine risk signals
- Confidence, and why it is separate
- Dashboard
- Review queue
- Case detail and decisions
- Customer risk profile
- Evidence collection portal
- Chargeback evidence pack
- Pre-ship buyer protection
- Monitor mode and Assisted mode
- Settings reference
- Savings and accuracy reporting
- Shopify integration: tags, metafields, Flow
- CSV exports
- Plans and usage limits
- Privacy, data handling and compliance
- Reliability and data integrity
- Permissions the app requests
- Troubleshooting
- FAQ
- Glossary
1. Installation and first run
The app installs like any Shopify app and starts working immediately. There is nothing to configure before you get value.
- Install the app and approve the requested permissions.
- Historical import starts automatically. Up to 12 months of orders, returns, refunds, fulfilments, delivery events and chargebacks are imported in the background using Shopify's Bulk Operations API. You can close the tab — progress is shown on the Dashboard and the import resumes safely if interrupted.
- Customer profiles are computed. Each customer's abuse-relevant aggregates (orders, spend, returns, refunded total, prior claim types, refunds without return, chargebacks) are built from the imported history.
- Your leakage audit is generated. See section 2.
- Live scoring begins. Every new return request is scored within seconds of arriving; every new order is scored against the buyer's history.
What to expect in week one
- The app starts in Monitor mode. It scores, explains and records, and it does not touch your Shopify returns.
- Confidence will be low on customers with little history — those cases route to a human by design.
- Use the first few weeks to compare the app's recommendations against your own judgement before enabling Assisted mode.
Import window and permissions
The 12-month window requires the read_all_orders permission. With standard read_orders access, Shopify only exposes the last 60 days of orders, and the audit is limited to that window. The app states the window length on the audit report so you always know what the numbers cover.
2. The free 12-month leakage audit
Location: Audit in the app navigation.
The audit is a report on money that already left your business. It is generated from your imported history, is timestamped, states its window length, and can be regenerated at any time with the Refresh button.
What the report contains
- Window totals — order count and value, return count and declared value, refund count and value
- Return rate by value — refunded value as a percentage of order value
- Refunds issued without a completed return — count and total value; in most stores this is the purest form of leakage
- Repeat returners — how many customers have three or more returns
- Highest-exposure repeat returners — a table of customer, orders, returns, total refunded and detected patterns
- Most frequent return reasons, ranked
- Estimated potential exposure, with the formula printed on the page
Pattern flags used in the audit
| Flag | Meaning |
|---|---|
high return ratio | A large share of this customer's orders came back |
refund without return | Refunds were issued where merchandise never physically returned |
repeat damage claims | Multiple "damaged" claims across orders |
repeat not-received claims | Multiple "item not received" claims across orders |
refunds > 50% of spend | Refunded value exceeds half of the customer's lifetime spend |
How the exposure estimate is calculated
Potential exposure, not guaranteed savings. The estimate is 100% of refunds issued without a completed return, plus 30% of refunded value from repeat-return customers whose refunds exceed half their lifetime spend.
The formula is shown on the report itself. It is an estimate of exposure you can investigate, not a promise of recovery.
3. How scoring works
The scoring engine is a deterministic rules engine — a pure function, not a machine-learning model. The same inputs always produce the same output, and every score records the engine version and rule-set version that produced it.
The pipeline
Shopify webhook → signature verified and de-duplicated → event persisted and queued → background worker → customer aggregates recomputed → scoring engine → risk score, signals and recommendation stored → review queue, order tags, metafields and Shopify Flow.
Heavy work never happens inside the webhook request, so Shopify never sees a timeout and your store is never slowed down.
Risk bands
| Band | Score | What it means | Default recommended action |
|---|---|---|---|
| LOW | 0–29 | Nothing unusual | Approve normally |
| MODERATE | 30–59 | Worth a light touch | Require a tracked return, or request photo evidence on damage claims |
| ELEVATED | 60–79 | A human should look at this | Manual review |
| HIGH | 80–100 | Prioritise this one | Urgent review |
Band cutoffs are adjustable in Settings.
What is stored with every score
- The score and band
- Every signal that fired, its point contribution, its plain-English explanation and its raw value
- The confidence score
- The recommended action
- The engine version and the rule-set version in force at the time
Scores are append-only. Re-scoring adds a new score rather than overwriting the old one, so historical decisions always remain explainable against the rules that existed when they were made.
Structural guarantees
- No auto-decline exists. There is no automatic-decline setting at any score. Declines are human decisions citing your published policy.
- Never blocks Shopify. If a score cannot be computed, the case is marked "score unavailable" and your normal Shopify workflow continues untouched.
- No protected characteristics. Scoring uses order, return, refund, delivery and chargeback behaviour only.
4. The nine risk signals
Each signal contributes points to the total. Every signal has an individual weight slider (0–2×) in Settings, and setting a weight to zero disables that signal entirely.
| # | Signal | What it detects | Built-in safeguards |
|---|---|---|---|
| 1 | Return frequency | Share of the customer's orders that came back, with an additional bump for three or more returns in 90 days | Only fires once the customer has at least three orders — first-time buyers are never judged on frequency |
| 2 | Refunds vs lifetime spend | Refunded total as a proportion of everything the customer has ever spent | Banded at >30%, >50% and >80% rather than a single cliff |
| 3 | Repeated claims | Prior "damaged", "item not received" and "wrong item" claims, with an extra escalation when the current request repeats the same claim type | Escalation requires a repeat of the same pattern, not merely several returns |
| 4 | Bracketing | Multiple variants of the same product in one order, most of which came back | Category-aware: heavily discounted for size-driven categories such as footwear and apparel, where buying three sizes is normal shopping |
| 5 | Refund without return | Prior refunds where merchandise never physically came back | Based on completed-return records, not assumptions |
| 6 | Return immediately after delivery | Return opened within hours of the carrier delivery scan | Uses real carrier delivery events; threshold configurable, default 24 hours |
| 7 | Multiple accounts per address | Other customer accounts sharing a hashed address | Deliberately capped low. It can support other signals but can never drive a decision on its own, and it is one-click suppressible per customer |
| 8 | High-value exposure | Request value above your threshold, weighted higher when the buyer has almost no history | Threshold configurable, default equivalent of $300 in your store currency |
| 9 | Chargeback / refund double dip | A chargeback filed on an order that already refunded, and customers with that pattern in their history | Surfaces as a dashboard alert and a case banner as well as a signal |
VIP exemption
Orders carrying any of your configured trusted-customer tags receive a 60% risk discount. It is a discount rather than a blanket zero, so a compromised VIP account still surfaces at extreme totals.
5. Confidence, and why it is separate
A score of 70 on a customer with two orders does not mean the same thing as a 70 on a customer with forty. The app therefore computes confidence (0–100) separately from risk, based on how much evidence exists:
- The customer's order count
- How much of your store's history has been imported
- Whether a carrier delivery scan is on file for the order
Below your configured confidence threshold, every case routes to a human regardless of score. The app is designed to say "there is not enough history here" rather than guess.
6. Dashboard
Location: the app home screen.
KPIs
- Open cases
- Awaiting review — elevated and high band cases with no decision yet
- Refund value at risk
- Recorded savings
Alerts and status cards
- Double-dip alert — open chargebacks on orders that have already been refunded, with the warning not to refund again while the dispute is open
- High-risk alert — links straight into the queue, pre-filtered
- Ingestion health — if any webhook events failed or stalled, you are told in the app rather than discovering it weeks later
- Audit summary — potential exposure, refunds without return, repeat-returner count
- Plan usage banner — appears at 80% of your monthly scored-request quota, with a clear notice if scoring pauses
- Onboarding state — import progress bar, with a retry button if an import fails
All monetary figures throughout the app display in your store's own currency.
7. Review queue
Location: Review queue.
The queue is the day-to-day working screen. Cases are sorted by score, highest first.
Filters
- State — open and unreviewed / reviewed / all
- Risk band — low / moderate / elevated / high
Columns
- Risk badge, with confidence shown alongside
- Order
- Customer, with their lifetime return count
- Value
- Reasons — the top contributing signals
- Recommendation
- Date
Actions
- Bulk mark reviewed — select multiple cases and clear them in one action
- CSV export of the case list
Cases still being scored, and cases where scoring was unavailable, are labelled explicitly rather than shown as low risk.
8. Case detail and decisions
This is the core screen of the product. Everything the app knows about a request is on one page.
Why this score
Every signal that fired is listed with:
- Its point contribution
- A plain-English explanation of what it means
- The raw value behind it, so you can verify the reasoning
The engine version, rule-set version, confidence and recommended action are shown alongside.
Case context
- Returned items — quantities, reasons and customer notes
- Delivery status — including an explicit "no delivery scan recorded" where that is the case
- Chargeback banner — shown when a dispute exists on the order, with the warning not to refund twice
- Duplicate-evidence warning — shown when a submitted photo also appears on another case, naming the other order
- Customer sidebar — orders, spend, returns, refunded total, and a badge if refunds were issued without returns
- Full audit trail — every system and human action with actor and timestamp
The six decisions
| Decision | What it does |
|---|---|
| Approve | Records approval. In Assisted mode, syncs the approval to Shopify's native return workflow |
| Approve as store credit | Records a store-credit outcome; logged in the savings ledger as cash retained |
| Request evidence | Mints a private customer upload link and a copy-ready message — see section 10 |
| Require tracked return | Records the requirement; if merchandise then comes back, it is logged as recovered |
| Mark reviewed | Clears the case from the awaiting-review count with no customer-facing outcome |
| Decline under policy | A human decision, citing your published policy. In Assisted mode, syncs to Shopify. Never automated |
Every decision has a free-text reason field, and the reason is kept permanently in the audit trail.
Other controls
- Re-score now — for when data arrives late, such as a delivery scan, a dispute or updated aggregates
- Open order in Shopify — deep link to the real order
- Evidence pack — printable case report, see section 11
9. Customer risk profile
Reachable from any case or from the customer column in the queue.
- Aggregate KPIs — orders, spend, returns, refunded total with percentage of spend, refunds without return, and a claim breakdown across damage, not-received and wrong-item
- Chargeback and double-dip badges
- Full return history, each row linking to its case
- Watchlist toggle — future orders from this customer are tagged automatically
- Shared-address panel with a one-click "Mark as legitimate shared household" action, which permanently suppresses the shared-address signal for this customer and records the suppression in the audit trail
- Deep link to Shopify for the customer's real contact details
Full personal data is never duplicated here — see section 19.
10. Evidence collection portal
Choosing Request evidence on a case does two things: it mints a private upload link for the customer, and it gives you a ready-to-send message you can paste into your email or helpdesk.
What the customer sees
A calm, neutral page asking for photos of the product, the packaging and the shipping label. It never mentions fraud, risk, scores or investigations. Customers are never shown their risk score or band.
Upload rules
- Up to 6 photos, 5 MB each
- Accepted formats: JPG, PNG, WEBP and HEIC — iPhone photos work without conversion
- An optional free-text note from the customer
- The link is single-purpose and expires after 14 days, showing a clear expiry message afterwards
- Photos attach to the case and are served only to authenticated staff
- Submission is recorded in the audit trail
Duplicate-photo detection
Every uploaded image is fingerprinted. If the same photo appears on a different claim — the classic recycled damage photo, and increasingly the same AI-generated image reused across claims — the case shows a warning naming the other order.
The recommended response is to request fresh, specific evidence, not to accuse the customer.
11. Chargeback evidence pack
Every case can be printed or saved as a PDF containing everything a dispute response needs:
- Order, case status, value and customer
- The full risk assessment — every signal explanation, the confidence score and the engine version
- Returned items with reasons
- A complete chronological timeline: order placed → fulfilment created → every carrier scan → return requested → refunds issued → chargeback filed → every staff decision with actor, timestamp and reason
- Customer-submitted evidence photos, inline, with upload timestamps
When you are contesting an "item not received" claim against a signed delivery, the paper trail is the evidence. This is the paper trail, in one document.
12. Pre-ship buyer protection
The best return to prevent is the one you have not shipped yet. Every new order is scored against the buyer's own history — return ratio, refund-to-spend, prior claims, refunds without returns and chargeback history — as it arrives.
When a buyer looks risky
- The order is tagged
RRAI:risky-buyer, andRRAI:watchlistif you have flagged that customer — visible in Shopify, filterable, and usable in Flow - Risk metafields are written to the order so staff and automations can read them
- An audit-log note records what was tagged and why
- Optional: a releasable hold on fulfilment for HIGH-band buyers, so a human looks before it ships. One click in Shopify Admin releases it. This is off by default and must be enabled deliberately
A hold is a pause, not a cancellation — and it is the only pre-ship action the app can take.
13. Monitor mode and Assisted mode
| Mode | Behaviour | When to use it |
|---|---|---|
| Monitor (default) | The app scores, explains and records. It never writes to Shopify returns. | Your first few weeks — measure the app against your own judgement before trusting it |
| Assisted | Approve and decline decisions sync to Shopify's native return workflow — only on an explicit click, and only for requests still awaiting a decision. | Once you are satisfied the recommendations match how your team already decides |
Assisted mode does not add any automation. It removes the double entry of deciding in the app and then repeating the action in Shopify.
14. Settings reference
Location: Settings. Every control below is per-store.
Scoring controls
| Setting | Description |
|---|---|
| Operating mode | Monitor or Assisted |
| Signal weights | An individual 0–2× slider for each of the nine signals. Zero disables the signal entirely |
| Band cutoffs | Move the LOW / ELEVATED / HIGH boundaries |
| High-value threshold | The request value that counts as high-value exposure, in your store currency |
| Minimum confidence for automation | Below this confidence, every case routes to a human regardless of score |
| Category exceptions | Product types where multi-size buying is normal. Footwear and apparel are pre-filled |
| VIP / trusted order tags | Orders with these tags receive the trusted-customer discount |
Integration toggles
- Push risk metafields to orders and customers
- Tag elevated and high-risk orders
- Pre-ship buyer scoring
- Hold fulfilment for HIGH-risk buyers (off by default)
- Weekly digest email
Preview impact before you save
Adjust any setting and the app dry-runs the new rules against your open cases, then reports exactly how many cases would change band and in which direction — before anything is saved.
Versioned rules and rollback
Saving creates a new rule-set version and re-scores open cases. Old scores keep pointing at the exact rules that produced them. A change you regret is rolled back by re-saving the previous values, and your audit trail stays intact throughout — nothing is edited in place.
15. Savings and accuracy reporting
Location: Savings.
Savings are recorded only from explicit outcomes, and every ledger entry stores the basis of its calculation so you can audit it.
The three savings types
- Prevented refund — a refund declined under your published policy after review
- Store-credit conversion — a cash refund turned into store credit, so the cash is retained
- Recovered merchandise — a return that actually came back after your team required a tracked return
Accuracy metrics, including the unflattering ones
- Approved-after-elevated-flag rate — the false-positive proxy. If the app is crying wolf, this number says so
- Agent override rate — how often your team disagrees with the recommendation, measured against the recommendation captured at decision time
- Average review time — from scoring to decision
- Requests scored, reviewed elevated and high cases, savings by type, monthly savings trend, and recent ledger entries with the basis for each
Per-product return rate
Units returned divided by units sold, per product — so you can find the product that is actually the problem. Sometimes the cause is not abuse at all but a misleading size chart or product photo.
Weekly digest email
Every Monday the store contact receives new cases, high-risk count, value at risk and savings recorded, with a link back into the queue. The digest is silent when there is nothing to report, and can be switched off per store in Settings.
16. Shopify integration: tags, metafields, Flow
Admin order risk card
A native admin block on the order page showing score, band, confidence, recommendation and top signals — right where your team already works.
Order tags
| Tag | Applied when |
|---|---|
RRAI:elevated-risk | The scored request falls in the ELEVATED band |
RRAI:high-risk | The scored request falls in the HIGH band |
RRAI:risky-buyer | Pre-ship scoring flagged the buyer's history on a new order |
RRAI:watchlist | The customer is on your watchlist |
Metafields
Written to orders and customers in an app-owned namespace, readable by staff and by any automation:
risk_scorerisk_bandconfidencerecommended_actiontop_signals
Shopify Flow trigger — "Return risk scored"
Fires after scoring completes, carrying order, customer, score, band, confidence and recommended action. Shopify's own "Return requested" trigger fires before any score exists; this trigger closes that gap.
Example workflows:
- Score above 75 → create a staff task
- Recommendation is request evidence → send the customer a templated email
- Tag is
RRAI:high-risk→ notify a Slack channel
Coexistence with your returns stack
Shopify self-serve returns and third-party return portals keep running exactly as they are. This is an intelligence layer, not a replacement — there is no customer-facing return flow to migrate.
17. CSV exports
- Cases export — one row per scored request with score, band, confidence, recommendation, value and decision
- Customers export — one row per customer with aggregates and pattern flags
Both are one click from the relevant screen, for merchants who want the data in their own reporting stack.
18. Plans and usage limits
| Plan | Price | Scored requests per month |
|---|---|---|
| Audit | Free | Full 12-month leakage audit + 25 scored requests |
| Growth | $99 / month | 250 |
| Pro | $249 / month | 1,500 |
| Plus | $599 / month | Unlimited |
All plans include every feature — audit, explainable scoring, pre-ship protection, evidence portal, Flow integration, evidence packs and savings reporting. Plans differ by scored-request volume only.
How quotas actually behave
- Billing is handled by Shopify — there are no separate payment details to enter
- You get an in-app warning at 80% of your monthly quota
- A 10% grace buffer above your quota means a busy week never hard-stops you mid-flow
- If you do exceed the buffer, cases are marked "not scored" with a logged reason. Nothing is silently dropped, and no request disappears
19. Privacy, data handling and compliance
Data minimisation
- Email, phone and address are never stored in readable form. They are stored only as one-way salted hashes, used purely for matching
- Customer names are stored as first name plus last initial
- When staff need the real details, the app deep-links to the customer record in Shopify rather than duplicating personal data
- Per-store salt — hashes are unique to your store, so nothing can be correlated across merchants
No cross-merchant data sharing
Your customers' behaviour stays yours. The app deliberately does not build or contribute to a shared cross-merchant fraud ledger.
GDPR and CCPA
- Customer data request — produces a report of exactly what is held about that customer
- Customer redaction — wipes identity data while retaining de-identified statistics
- Uninstall — processing stops immediately and your store's data is deleted in line with Shopify's data-deletion flow
Auditability and fairness
- Append-only audit trail of every score, decision, override and signal suppression, with actor and timestamp — including who decided and why
- No protected or demographic characteristics are used in scoring
- Household sharing is capped and one-click suppressible; bracketing is category-aware; no automatic declines exist
- Transparent methodology — the exposure formula, the signals, the point contributions and the rule version are all shown to you. Nothing is hidden behind "proprietary AI"
- Encrypted in transit throughout; the database is not publicly reachable
False accusations against legitimate customers cost more than the refund you are investigating. The product is built around that fact.
20. Reliability and data integrity
- Real-time ingestion of orders, returns, refunds, fulfilments, delivery events, disputes and subscription changes across 15 webhook topics
- Every webhook is signature-verified and de-duplicated — a duplicate delivery can never double-process, and no refund action can fire twice
- Fast acknowledgement, background processing — heavy work never happens in the webhook request, so Shopify never sees a timeout
- Nightly self-healing reconciliation — a sweep re-syncs recent orders, returns, refunds and disputes and scores anything the webhooks missed. Webhook delivery is at-least-once, not guaranteed; the app repairs the gap every night instead of assuming it never happens
- Retries with backoff on every background job, with cost-aware handling of Shopify API rate limits
- Resumable import — interrupt the historical import and it picks up where it left off without duplicating anything
- Ingestion health surfaced in-app — failures appear on the Dashboard, not only in logs
- Multi-currency — every figure displays in your store's own currency
- Multi-tenant isolation — every record is scoped to your store
- Version-pinned Shopify API with a deliberate upgrade cadence
- Tested scoring engine — a pure, deterministic function covered by fixture tests, including explicit tests that it never recommends a decline and that household sharing alone can never raise a case
21. Permissions the app requests
| Permission | Why it is needed |
|---|---|
| Read orders | Order history and values, the basis of every aggregate. read_all_orders extends the audit window from 60 days to 12 months |
| Read customers | Customer identity matching (hashed) and lifetime aggregates |
| Read returns | Return requests, items, reasons and states |
| Read fulfilments | Carrier delivery scans, used by the return-after-delivery signal and the evidence pack timeline |
| Write returns | Assisted-mode approve and decline sync — only on an explicit click |
| Write orders and customers | Risk metafields and order tags for Flow and admin visibility |
| Read Shopify Payments disputes | Chargeback and double-dip detection |
| Read and write fulfilment orders | The optional releasable pre-ship hold on HIGH-risk buyers |
22. Troubleshooting
The audit only covers 60 days
The 12-month window requires read_all_orders access. Without it, Shopify exposes only the last 60 days of orders. The report always states the window it actually covers.
A case shows "score unavailable"
Scoring could not complete for that request. The case is not treated as low risk, and your normal Shopify workflow is unaffected. Use Re-score now on the case; if it recurs, check the ingestion health card on the Dashboard.
A case shows "not scored"
Your monthly scored-request quota and its 10% grace buffer were exceeded. The reason is logged on the case. Upgrading the plan resumes scoring; nothing that arrived while paused is lost from the record.
The score looks wrong for this customer
Open Why this score and read the per-signal raw values — the explanation states which behaviour triggered each contribution. If a signal is systematically wrong for your store, lower or zero its weight in Settings and use Preview impact to see the effect before saving.
A family shares one address and keeps getting flagged
Open the customer's profile and use Mark as legitimate shared household. The shared-address signal is permanently suppressed for that customer, and the suppression is recorded in the audit trail. Note that this signal is capped low and cannot raise a case on its own.
Delivery status says no scan recorded
No carrier delivery event has reached the app for that order. The return-after-delivery signal cannot fire without one, and confidence is reduced accordingly. If the scan arrives later, use Re-score now.
The import failed or stalled
Use the retry button on the Dashboard onboarding card. Imports are resumable and idempotent — retrying never duplicates data.
A decision did not appear in Shopify
Check the operating mode. In Monitor mode — the default — decisions are recorded in the app and deliberately never written to Shopify. Switch to Assisted mode in Settings if you want approve and decline to sync.
23. FAQ
Do you replace my returns portal, or Shopify self-serve returns?
No. This is an intelligence layer that runs alongside whatever you already use. Your customer-facing return flow does not change.
Will this automatically decline my customers?
No, and it cannot. There is no automatic-decline capability in the product at any score or setting. Declines are made by your team, citing your published policy.
What do my customers see?
Nothing, unless you request evidence. If you do, they see a neutral page asking for photos of the product, packaging and label. Scores, risk bands and internal notes are never shown to customers.
Will it flag my honest customers?
It is built specifically to avoid that. Bracketing is discounted for size-driven categories, shared-address matching is capped and one-click suppressible, low-confidence cases always go to a human, and the app reports its own false-positive proxy so you can hold it to account.
How much history do you need?
It works from day one and improves as history accumulates. Until then, confidence stays low and cases route to your team rather than being guessed at.
What data do you store about my customers?
Order, return, refund, delivery and chargeback behaviour. Email, phone and address are stored only as one-way salted hashes; names as first name plus last initial. Full personal details stay in Shopify, where the app deep-links to them.
Is my data shared with other merchants?
No. Never. There is no cross-merchant ledger.
What happens if I uninstall?
Processing stops immediately and your store's data is deleted in line with Shopify's data-deletion flow.
Can I tune the scoring?
Yes — every signal's weight, all band cutoffs, the high-value threshold, the confidence floor, category exceptions and VIP tags. And you can preview the impact of any change before saving it.
Does it use AI or machine learning?
No. It is a deterministic, explainable rules engine. Same inputs, same output, every time — which is why every score can be broken down into the exact points each signal contributed.
Does it work with Shopify Flow?
Yes — a dedicated "Return risk scored" trigger that fires after scoring, plus risk metafields and order tags that any Flow workflow can read.
Does it help with chargebacks?
In two ways: it detects refund-plus-chargeback double dips and alerts you, and it produces a one-click evidence pack with the full order, delivery and decision timeline for your dispute response.
Who is it for?
Stores with meaningful return volume — apparel, footwear, beauty, accessories and consumer electronics are the strongest fit — where a support team handles returns by hand today, and where the goal is to enforce a generous return policy fairly rather than tighten it.
What is it not?
It is not a returns portal, a shipping-label tool, an exchange-automation engine or a warehouse system. It is the intelligence layer that sits behind those tools.
24. Glossary
- Risk score
- A 0–100 number expressing how much abuse-relevant signal exists on a request. Never a verdict.
- Risk band
- LOW, MODERATE, ELEVATED or HIGH — the score bucketed for triage, with merchant-adjustable cutoffs.
- Confidence
- A separate 0–100 number expressing how much history the score is based on. Below your threshold, cases always route to a human.
- Signal
- One of the nine behaviours the engine measures. Each contributes points and carries a plain-English explanation and a raw value.
- Bracketing
- Buying several variants of the same product intending to keep one. Normal in size-driven categories, which is why the signal is category-aware.
- Refund without return
- A refund issued where merchandise never physically came back — usually the largest single source of measurable leakage.
- Double dip
- A chargeback filed on an order that has already been refunded, so the merchant pays twice.
- Potential exposure
- The audit's estimate of money at risk, calculated with a published formula. Not a guarantee of recovery.
- Monitor mode
- The default. The app scores, explains and records, and never writes to Shopify returns.
- Assisted mode
- Approve and decline decisions sync to Shopify's native return workflow, only on an explicit click.
- Rule-set version
- The append-only snapshot of your settings in force when a score was produced. Old scores always point at the rules that created them.
- Evidence pack
- A printable case report containing the risk assessment, the full timeline and customer-submitted photos, built for dispute responses.