Returns Suite
phase-2 · product-strategy · input-artefact
To-Be V1.1 In Review
To-Be Service Blueprint · Future State · North Star Vision

How Returns Should Work
for Mid-Market Apparel Retailers

A full-actor service blueprint mapping the future state returns journey across 8 human actors and 9 system actors. North star vision — unconstrained by current technical feasibility or phasing. Direct before/after companion to the as-is blueprint. Click any node to explore resolved pain points, system transformations, and feedback loop closures.

"A returns experience so effortless for the returner, so intelligent for the merchant, and so invisible in its orchestration that returns become a source of loyalty rather than a source of loss."
SegmentMid-market US apparel, $5M–$100M ARR
ChannelOnline only — no physical retail
PathHappy path · Self-service initiation
OwnerUX Lead
Assumptions12 documented · click to view
Executive Summary — What Changed
01
Returns begin with a conversation, not a form — AI conversation agent captures return intent in natural language; reason codes become AI-verified, not shopper-selected. P1 pain points #1 and #13 resolved.
02
The warehouse knows what is coming before it arrives — advance return notice generated at initiation (Stage 2), not shipment. SKU breakdown, arrival windows, and staffing recommendations delivered before the first item moves. P1 pain point #6 resolved.
03
Stages 4–7 move from Heavy to Low manual load — autonomous vehicle unloading, automated conveyor scanning, computer vision inspection, robotic re-tagging, and programmatic disposition eliminate manual labor at the highest-friction stages. P1 pain points #4, #10, #11, #12 resolved.
04
Both Merchandising feedback loops are closed structurally — inspection findings (Stage 5) and restock confirmations (Stage 7) route automatically to Merchandising. No report written. No email sent. The data flows because the system is designed for it. P1 pain points #2 and #3 resolved.
05
Fraud is eliminated at the source — refunds triggered post-inspection, not post-receipt. Computer vision photographs every item. Permanent fraud evidence chain. P1 pain point #7 resolved.
06
Operations Supervisor sees everything in real time — unified dashboard aggregates receiving, inspection, disposition, restock, refund liability, and true cost of returns in a single view with 4-hour refresh. P1 pain points #5, #8, and #9 resolved.
07
Analytics built on inspection-verified ground truth — NLP classification and computer vision inspection replace shopper-reported reason codes as the primary analytics input. Reports are credible for the first time. P1 pain point #13 resolved.
08
Human actors are decision-makers, not process operators — Marcus handles 3 of 47 receiving exceptions. Elena adjudicates 12 of 127 inspections. Jordan approves recommendations. Diana reviews 3 alerts. Every human spends time on judgment, not data assembly.
Experience Principles — Governing Every Design Decision
P1
Relief through conversation
Every interaction feels like talking to a knowledgeable human
P2
Humans decide, the system does
Orchestration is automated; humans handle decisions only
P3
Proactive intelligence, routed right
Right information, right role, right format, before it's asked for
P4
Predicted truth, not reported opinion
AI derives ground truth; stated reason is one input only
P5
The trust contract is unconditional
Resolution guaranteed once the returner completes her side
P6
The brand is the experience
Returns suite is invisible; retailer brand is what the returner sees
P7
Full cost visibility, always current
Unified dashboard, 4-hour refresh, fully-loaded cost as headline
P8
Closed feedback loops, human-approved
System routes intelligence; humans approve actions
P9
Automate the physical, not just the digital
Robotics and autonomous systems eliminate physical labor wherever possible
Legend
Real-time automated
AI-orchestrated
Pain resolved
New system actor
Feedback loop closed
Human decision point
Actor / Stage →
01
Return Decision
Shopper
02
Initiation & Auth
Both
03
Outbound Shipping
Both
04
Warehouse Receiving
Merchant
05
Inspection & Grading
Merchant
06
Disposition Routing
Merchant
07
Restocking
Merchant
08
Refund Processing
Both
09
Reporting & Analytics
Merchant
Human Actors
👤
Returner
Human
Policy surfaces proactively — no hunting
Natural language conversation — no forms, no dropdowns3
AV arrives; hands over item; no packaging, no label2
Proactive updates; predictable 3–5 day SLA; self-service tracking2
🚐
AV / Return Hub
Human + Autonomous
NEWAV executes home pickup; no human handling; logs transfer to OMS in real time
AV docks; transfers items to conveyor automatically
📬
Warehouse Receiving
Human — Exception Handler
Monitors dashboard; handles exception lane only (~5% of items)
3 of 47 items this morning
🔍
Returns Inspector
Human — Adjudicator
Monitors CV station; adjudicates flagged exceptions only
12 of 127 items today
Routes manual adjudication items to staging
📦
Inventory Mgmt
Human — Approver
Reviews AI disposition recommendations; one-click approval; overrides on judgment
Monitors restock cycle time; handles exceptions only3
💰
Finance Team
Human — Analyst
2-min exception review; BNPL + chargebacks automated; fully-loaded cost model visible5
Interprets numbers, identifies trends — not extracting data2
👁
Ops Supervisor
Human · P&L Owner
Dashboard shows live initiation rate — forward signal
Inbound pipeline updates in real time
Real-time in-transit count on dashboard
Receiving throughput live; staffing alerts proactive
Inspection queue depth + grade distribution live
Disposition split + recovery rate live
Restock cycle time per SKU live
Real-time refund liability by payment method
Reviews 3 dashboard decisions; weekly report auto-generated — 10-min review3
🛍
Merchandising / Buying
Human — Intelligence Consumer
Receives defect alerts automatically — Feedback Loop 1 CLOSED
⬤ Loop 1 Closed
Receives restock notifications automatically — Feedback Loop 2 CLOSED
⬤ Loop 2 Closed
SKU dashboard: AI-verified returns data; defect patterns; buying cycle intelligence3
System Actors — Transformed & New
🤖
AI Conversation Agent
New System Actor
NEWSurfaces return policy proactively; available for pre-return Q&A
NEWNLP maps natural language → structured return codes; validates eligibility; creates RMA; schedules AV pickup; sends confirmation
NEWSends "we've got your return" proactive SMS confirmation
Sends proactive "exchange on its way" notification to returner
NEWProvides NLP-classified reason code analytics — first credible reason code dataset
⚙️
OMS (Transformed)
Shopify / NetSuite — Real-Time
Holds order record; enriches with real-time return eligibility
Real-time RMA creation; immediately sends ARN to WMS; publishes RMA_CREATED event to OI Core
Receives real-time AV pickup confirmation; triggers "en route" notification
Receives real-time receipt confirmation from WMS; starts refund clock immediately
Receives inspection outcome; triggers refund/exchange processing automatically
Records final disposition in real time
Real-time inventory count update; item immediately available-to-sell; triggers Merchandising Feed
Triggers refund to payment processor automatically post-inspection
Source of resolution analytics; real-time data — not batch
🏭
WMS (Transformed)
Deposco / Extensiv — Integrated
Receives ARN at initiation; updates inbound volume forecast immediately with SKU detail
Receives real-time arrival estimate from AV network; refines arrival window
Pre-matched items; automated conveyor scanning; real-time OMS sync on receipt; exception routing to Marcus's terminal
Receives automated grades + photos from CV system; auto-generates disposition recommendation; logs Elena's adjudications
Programmatic disposition routing on Jordan's approval; sends liquidation batch to vendor automatically; routes discard to recycling
Robotic re-tagging with correct SKU/condition/price; automated conveyor; no human handling for clean items
Sends inspection outcome to OMS; triggers refund pipeline
Real-time data feed to Returns Intelligence Dashboard; no CSV export required
📷
Computer Vision System
New System Actor
NEWMulti-angle photography; AI condition grading with confidence score; defect classification; fraud pattern detection; grades + photos auto-logged to WMS
NEWProvides inspection ground truth: condition distribution, defect pattern analysis, fraud signal frequency
📊
OI Core & Agent Mesh
New System Actor
Monitors early return signal indicators; updates inbound volume forecast
Updates inbound forecast on every new RMA; recalculates staffing recommendation
Refines arrival window based on AV tracking data
Updates receiving-vs-forecast in real time; proactive staffing alerts
Updates grade distribution forecast; fraud spike alerting
Liquidation pricing signals; optimal timing recommendation; recovery rate benchmarking
Restock cycle time tracking; alerts when items exceed threshold
Refund liability forecasting for open returns pipeline
NEWGenerates all automated reports; trend detection; anomaly alerting; SKU-level return rate analysis; week-over-week modeling
🚗
Autonomous Logistics
New System Actor
NEWReceives pickup schedule from AI agent; assigns vehicle; confirms 2-hour pickup window
NEWExecutes home pickup; box-free; real-time tracking to OMS and WMS throughout transit
NEWDocks at warehouse; transfers items to receiving conveyor automatically; logs transfer to WMS
💳
Payment & Accounting
Transformed — Automated
Automated refund on OMS trigger; BNPL auto-routing to Klarna/Afterpay/Affirm; real-time QuickBooks/NetSuite sync; chargeback evidence auto-assembled
Real-time accounting data; chargeback rate tracked separately; fully-loaded cost model inputs
📡
Merchandising Feed
New System Actor
Receives structured defect data from CV system → routes to Priya's dashboard
⬤ Feedback Loop 1 CLOSED
Receives restock confirmation from OMS → routes to Priya's dashboard
⬤ Feedback Loop 2 CLOSED
NEWProvides buying cycle intelligence and returns-adjusted analytics to Priya's view
🖥
Intelligence Dashboard
New System Actor
Live return initiation rate; forward pipeline signal
Inbound pipeline count updates in real time on every new RMA
In-transit volume count updates in real time from AV network
Receiving throughput live; exception queue; Marcus's terminal view
Grade distribution, inspection throughput, fraud alert count live
Disposition split and liquidation recovery rate live
Restock queue and cycle time per SKU live
Outstanding refund liability by payment method in real time
NEWUnified cross-system view; 4-hour refresh; role-based views (Ops / Finance / Merch); auto-generated weekly reports; zero manual assembly
Manual
Load
None
Minimal
None
Low
Low
Low
Low
Minimal
None
Broken Feedback Loops — Now Closed

Two structural gaps closed automatically

Loop 1 — Stage 5
Inspection Findings → Merchandising
As-is failure: WMS inspection findings (damage type, quality defects, sizing issues) never reached the Buying team. High-return SKUs were reordered without modification. Quality failures were invisible to the people with the power to fix them.

To-be resolution: Computer Vision Inspection System generates structured defect data at every inspection. This feeds the Merchandising Intelligence Feed automatically. When a defect pattern threshold is breached for a SKU, an alert with inspection photographs is routed to Priya's dashboard. No report written. No email sent. The loop is structural and automatic. P1 pain point #2 resolved.
Loop 2 — Stage 7
Restocked Returns → Merchandising
As-is failure: Buying team was never notified when returned inventory re-entered stock. Could not use restocked returns in promotional planning, markdown strategy, or buying decisions. Unknowingly ordered new units of SKUs that already had hundreds of restocked returns in the warehouse.

To-be resolution: When a returned item re-enters available inventory in the OMS, the Merchandising Intelligence Feed automatically routes a structured notification to Priya's dashboard: SKU, condition, quantity, restock price. This fires on every restock event without Jordan writing an email. P1 pain point #3 resolved.
Friction Heatmap — As-Is vs. To-Be Comparison

The contrast is stark at Stages 4–7

Stage
As-Is Manual Load
To-Be Manual Load
Stage 1 — Return Decision
Low
None
Stage 2 — Initiation & Auth
Partial
Minimal
Stage 3 — Outbound Shipping
Partial
None
Stage 4 — Warehouse Receiving
Heavy
Low
Stage 5 — Inspection & Grading
Heavy
Low
Stage 6 — Disposition Routing
Heavy
Low
Stage 7 — Restocking
Heavy
Low
Stage 8 — Refund Processing
Partial
Minimal
Stage 9 — Reporting & Analytics
Heavy
None
P1 Pain Point Resolution Register

All 14 P1 Pain Points — Resolved

#Pain PointStage ResolvedHow Resolved in To-Be State
P1·01Reason code accuracy structurally low2+5AI Conversation Agent NLP maps natural language at initiation; Computer Vision provides inspection-verified ground truth; shopper-selected reason codes retired
P1·02Broken Feedback Loop 1 — inspection findings never reach Merchandising5CV System → Merchandising Intelligence Feed closes loop automatically; defect pattern alerts with photographs routed to Priya's dashboard
P1·03Broken Feedback Loop 2 — restocked returns never reach Merchandising7OMS → Merchandising Intelligence Feed closes loop on every restock event; structured notification to Priya's dashboard
P1·04Restock cycle time 5–10 business days7Automated robotic re-tagging + real-time OMS sync; target <24 hours from receipt to available-to-sell
P1·05Returns data fragmented across 3–5 systems9Returns Intelligence Dashboard unifies all system data; single source of truth; 4-hour refresh; no CSV export required
P1·06WMS has no advance return notice2ARN generated at initiation (not shipment); WMS receives SKU breakdown and arrival windows before item leaves returner's hands
P1·07Refund-on-receipt fraud window5+8Refund triggered post-inspection confirmation; CV photographs every item; permanent fraud evidence chain
P1·08True cost of returns systematically underestimated9Fully-loaded cost model: direct costs + indirect costs (restock opportunity cost, liquidation gap, LTV impact); Rachel sees complete picture
P1·09Operations Supervisor has no real-time visibility across Stages 3–79Returns Intelligence Dashboard: real-time cross-stage view with receiving throughput, inspection queue, grade distribution, disposition split, restock cycle time, refund liability
P1·10Grading standards informal and unenforced5Computer Vision standardized grading rubric; AI confidence scores; no drift across staff or shifts; Elena adjudicates edge cases only
P1·11No photographic documentation at inspection5Multi-angle photography of every item under standardized lighting; permanent storage against return record; immediately available for dispute resolution
P1·12Disposition rules informal and not programmatic6Disposition Agent reasons from multi-signal inputs (condition grade, OI Core demand signal, liquidation pricing, days-in-staging) → recommendation with reasoning surfaced to Jordan; Jordan approves; consistent outcomes
P1·13OMS reason code data corrupts all downstream analytics2+5+9Analytics rebuilt on NLP classification (Stage 2) + CV inspection ground truth (Stage 5); OMS reason code data retired as primary analytics input
P1·14Returns portal absent for smaller retailers2AI Conversation Agent provides portal-equivalent for all tiers; natural language initiation; no portal software license required
Full Assumptions Register

12 Documented Assumptions

#AssumptionConfidenceImplication if Wrong
A1Autonomous vehicle home pickup is available in major US metro areas at commercial scale within the North Star horizon (3–5 years)MediumIf AV pickup is not commercially deployable, Stage 3 falls back to carrier drop-off with label generation; ARN architecture remains valid regardless
A2Computer vision grading at mid-market price points is commercially accessible within the North Star horizonHighComputer vision in retail growing 25.4% CAGR; enterprise deployment already live. If price remains prohibitive, Stage 5 falls back to structured mobile inspection workflow (rubrics + photo capture without CV automation)
A3Robotic re-tagging systems are deployable at mid-market warehouse scale and price points within the North Star horizonLow–MediumMost capital-intensive physical automation assumption. If cost-prohibitive, Stage 7 returns to manual re-tagging; Feedback Loop 2 closure at Stage 7 remains intact regardless
A4BNPL providers (Klarna, Afterpay, Affirm) offer API-based refund triggering compatible with automated merchant systemsHighKlarna and Afterpay APIs confirmed. If any provider's API scope changes, automation falls back to merchant-initiated portal workflow for that provider only
A5Real-time WMS-OMS API integration is technically achievable across mid-market WMS platforms (ShipBob, Deposco, Extensiv)HighBoth platforms have documented APIs. If a merchant's WMS has no API, a daily batch sync fallback is required; real-time receiving benefits partially eroded
A6Mid-market apparel retailers will accept an AI conversation agent as the primary returns initiation interfaceMediumIf adoption is slower, a hybrid (AI + form-based) initiation path is required; NLP reason code capture still achievable
A7Operations Intelligence Core (OI Core) can forecast inbound return volume with sufficient accuracy (±15%) to be operationally useful for staffing recommendationsMediumIf forecast accuracy is lower at launch (particularly during the OI Core cold start period), staffing recommendations are advisory rather than automated; ops value reduced but not eliminated
A8Computer vision fraud detection achieves false positive rate low enough that merchant trust is maintained (<5%)MediumIf false positive rate is above 5%, Elena must adjudicate more items and efficiency gain is partially eroded; photographic evidence chain remains intact regardless
A9Merchants will grant the Disposition Agent authority to recommend liquidation timing and batching, with human approval gateHighHuman approval gate is a design constraint, not an assumption. Merchants who want tighter control can require approval for every item rather than batch approval
A10The Merchandising / Buying team at mid-market apparel retailers ($20M–$100M ARR) has a separate staffed function that will consume the Merchandising Intelligence FeedMediumAt $20M+ ARR with a separate Buying function, this holds. At $5M–$15M ARR, the buyer may be the founder; the feed is still valuable but consumption workflow is informal
A11QuickBooks and NetSuite real-time sync is achievable through existing API capabilitiesHighBoth platforms have documented real-time APIs for transaction posting. Integration complexity is known and addressable
A12Fully-loaded cost of returns modeling (including LTV impact and restock opportunity cost) can be produced with sufficient accuracy to be credible to a CFOMediumDirect costs are measurable; indirect costs require proxy modeling. Accuracy improves over time as the platform accumulates merchant data