Guide 22 mins

Fashion and Apparel: Returns Intelligence on D23.io

Master returns intelligence for fashion brands using D23.io and AI-driven analytics. Reduce return rates, optimise SKU performance, and unlock profitability.

The PADISO Team ·2026-04-26

Why Returns Intelligence Matters for Fashion Brands
Understanding Returns Data on D23.io
Superset Deployment for AU Fashion Brands
SKU-Level Return Rate Analysis
Fit Signals and Product Design Optimisation
Claude-Driven Returns Reason Classification
Building a Returns-Centric Operating Model
Measuring Impact and ROI
Implementation Roadmap and Next Steps

Why Returns Intelligence Matters for Fashion Brands

Returns are the silent profit killer in fashion and apparel. While most brands obsess over acquisition costs and conversion rates, they ignore the 15–40% of revenue that walks out the door as returned stock. For a mid-market Australian fashion label doing $5M in annual revenue, that’s $750K to $2M in returned goods—before restocking costs, logistics, and markdown losses.

Returns intelligence isn’t about shame or blame. It’s about signal extraction. Every return is a data point that tells you something true about your product, your sizing, your photography, your customer expectations, or your supply chain. The brands winning in 2024 aren’t those with the lowest return rates—they’re the ones who understand their returns deeply enough to act on them.

This is where D23.io and modern returns analytics platforms become strategic tools. They let you move from reactive (“why did this customer return?”) to predictive (“which SKUs will drive returns? Which fit signals are broken? Which product lines are cannibalising each other?”). For founders and operators building fashion tech or modernising apparel operations, this capability is non-negotiable.

When you partner with an AI automation agency Sydney or venture studio to build returns intelligence, you’re not just installing software. You’re embedding a feedback loop that touches product design, inventory planning, supplier relationships, and customer experience. The payoff compounds: lower return costs, better inventory turns, smarter SKU decisions, and—critically—happier customers who feel understood.

Understanding Returns Data on D23.io

D23.io is a data orchestration platform purpose-built for retail and fashion operations. Unlike generic analytics tools, D23.io is built around the operational reality of fashion: SKUs, sizes, colours, fit profiles, inventory locations, and the complex web of supplier, warehouse, and customer data that determines whether a garment makes it to a customer’s wardrobe or back to a distribution centre.

Returns data on D23.io typically ingests from multiple sources:

Point-of-sale systems (Shopify, custom e-commerce platforms, in-store POS)
Returns management platforms (Returnly, Happy Returns, custom RMA systems)
Warehouse management systems (NetSuite, SAP, Cin7, custom WMS)
Customer service tools (Zendesk, Intercom, custom ticketing)
Inventory systems (ERP, merchandising tools)

The platform normalises this data into a unified schema where a return is never just a transaction—it’s a rich event with context: the original order, the SKU, the size ordered, the colour, the customer’s location, the reason given, the reason inferred, the return window, the refund status, and the disposition of the returned item.

For Australian fashion brands, this matters acutely. Australia has unique logistics constraints: longer lead times from overseas suppliers, higher domestic shipping costs, and regional variation in sizing preferences and climate-driven demand. A returns intelligence system that doesn’t account for these factors will give you generic advice. One built on D23.io can surface that, say, your size 8 denim has a 35% return rate in Brisbane due to fit, but only 18% in Melbourne—and that’s because Brisbane customers tend to prefer a slimmer fit, or because you’re sourcing from a different supplier for that region.

Superset Deployment for AU Fashion Brands

Superset is an open-source data visualisation and exploration platform. When deployed on top of D23.io returns data, it becomes a returns intelligence cockpit—a real-time, explorable dashboard that lets product teams, merchandisers, and executives ask questions about returns without needing a data analyst in the room.

Here’s what a Superset deployment for AU fashion brands typically surfaces:

Core Returns Metrics

Overall return rate (by month, by channel, by cohort)
Return rate by SKU (which products are driving disproportionate returns)
Return rate by size (are certain sizes returning more? Are they sizing up or down?)
Return rate by colour (is black returning less than white? Why?)
Return rate by price point (do higher-priced items return more?)
Return window analysis (are returns happening at day 3, day 7, day 14? What does that signal?)

Operational Metrics

Return reason distribution (fit issues, quality defects, changed mind, damage in transit, wrong item shipped)
Refund status and timeline (how long from return received to refund issued?)
Returned item disposition (resold as new, marked down, donated, scrapped)
Return logistics cost (cost per return, by method, by region)

Predictive Signals

Cohort-level return propensity (customers who bought in January have a 22% return rate; those who bought in March have 18%—why?)
Product-level return velocity (new SKU has 8% return rate in week 1, 15% in week 2—early warning of a problem)
Geographic variation (return rates by state, by city, by customer segment)

When you deploy Superset on D23.io returns data, you’re creating a self-service analytics layer. A product manager can drill into “why did our new summer collection return at 28%?” and surface that it’s not the design—it’s that the fabric shrinks 3% in the wash, and your sizing guide didn’t account for that. A merchandiser can see that your supplier’s latest batch of size 10 jeans is running half a size small and flag it before it hits the warehouse.

For AI automation for e-commerce teams, Superset also becomes a feedback loop. If you’re using recommendation engines to suggest complementary items, you can see whether customers who bought the recommended item are returning it at a higher rate—and if so, the recommendation logic needs to evolve.

SKU-Level Return Rate Analysis

SKU-level return analysis is where returns intelligence stops being defensive (“let’s reduce returns”) and becomes strategic (“let’s use returns data to make better product and inventory decisions”).

Identifying Problem SKUs

Start with the basics: rank all active SKUs by return rate. You’ll typically see a power law distribution: 80% of your returns come from 20% of your SKUs. Those outliers are your signal.

A SKU returning at 35% isn’t just a problem—it’s a gift. It’s telling you something concrete:

Fit is broken. The size chart is wrong, the fabric has stretch properties that aren’t documented, or the cut is inconsistent with customer expectations (e.g., “boyfriend fit” that’s actually “oversized”).
Quality is inconsistent. A batch from supplier A has 8% defects (seams failing, dye running, zips breaking); a batch from supplier B has 2%.
Expectations don’t match reality. Your product photography shows a look that the garment doesn’t deliver. Your description says “luxe” but the fabric feels cheap. Your marketing promises durability but the stitching fails after three washes.
The customer base is wrong. You’re selling a $150 minimalist linen shirt to fast-fashion customers who expect something more forgiving and trend-driven. They buy it, try it, and return it.

Cohort-Level Return Patterns

Beyond individual SKUs, look at cohorts:

By collection. Does your winter collection return at 18% but your summer collection at 28%? That suggests a fit or fabric issue specific to seasonal products.
By supplier. Do SKUs from supplier A return at 12% while supplier B’s SKUs return at 22%? That’s a quality, consistency, or specification problem you need to address with supplier B.
By price point. Do your $50 basics return at 15% but your $200 premium items return at 32%? That might mean your premium line has quality issues, or it might mean you’re setting expectations too high in the marketing.
By channel. Do items sold via your own website return at 20% but items sold via a marketplace return at 35%? That could indicate that marketplace customers are less familiar with your brand, or your product descriptions on the marketplace are weaker.

Using SKU Returns to Inform Inventory and Buying Decisions

Here’s where returns intelligence directly impacts the bottom line. If a SKU has a 30% return rate, you should not be buying 10,000 units next season. You should either:

Fix the problem (work with the supplier to improve fit, quality, or specifications) and then buy cautiously (2,000 units to test the fix).
Discontinue the SKU and reallocate budget to better-performing products.
Reposition the product (change the price point, the marketing angle, or the target customer) and test the new positioning with a small batch.

For Australian fashion brands, this is especially important because supplier lead times are long (often 90–120 days from Asia). If you place a large order based on last season’s sales without accounting for returns, and the product has a 25% return rate, you’ve just committed $200K to a SKU that will generate $50K in net returns. By the time you realise the problem, the next order is already in flight.

When you work with a venture studio partner to build returns intelligence, you’re creating a feedback loop that informs buying decisions 3–4 months in advance. That’s the difference between reactive (“we have too many returns”) and proactive (“we’re not going to make that mistake again”).

Fit Signals and Product Design Optimisation

Fit is the single largest driver of returns in fashion. Studies consistently show that 40–50% of fashion returns are fit-related: customer ordered the wrong size, the garment fit differently than expected, or the sizing was inconsistent with their usual size.

Traditional fashion brands handle this with static size charts and hope. Modern brands use fit signals—data points extracted from returns, customer feedback, and product imagery—to continuously refine and improve sizing.

Extracting Fit Signals from Returns Data

When a customer returns a garment due to fit, you capture:

Size ordered (e.g., size 10)
Reason (“too small,” “too large,” “not the right fit”)
Customer profile (if available: height, usual size in other brands, body shape preferences)
Product attributes (fabric, stretch, cut, fit profile)

Now, aggregate this: if 200 customers ordered size 10 in your summer dress, and 85 of them (42%) returned it saying “too small,” that’s a signal. Either:

The size chart is wrong (size 10 should actually be listed as size 8–9).
The fabric shrinks in the wash (you need to add a care label: “size up one size”).
The cut is tighter than customers expect (you need to adjust the pattern or reposition the product as “fitted” rather than “relaxed”).

Building a Fit Model

Advanced returns intelligence systems build a “fit model” for each product. This model captures:

True size (based on actual measurements and customer feedback)
Fit profile (fitted, regular, relaxed, oversized)
Stretch (does the fabric have 2-way or 4-way stretch? Does it relax after wear?)
Shrinkage (does the fabric shrink in the wash? By how much?)
Consistency (does this SKU run true to size, or is there variance between batches?)

For a Sydney fashion brand with multiple suppliers, this is critical. A supplier in Vietnam might cut your size 10 jeans 1cm narrower in the waist than a supplier in Indonesia. Without a fit model, you’re flying blind. With one, you can flag the discrepancy, adjust the pattern, and ensure consistency across suppliers.

Using Fit Signals to Improve Product Descriptions

Fit signals also feed back into marketing and product descriptions. If your data shows that 60% of customers who buy your “oversized” blazer order a size smaller than their usual size, then your product description isn’t communicating the fit correctly. You should update it to say “runs large—size down one” or adjust your photography to show the actual fit more clearly.

This is where AI and ML integration becomes valuable. Computer vision models can analyse product photography and flag when the fit shown in the image doesn’t match the fit described in the text. Natural language models can generate fit descriptions based on returns data: “This style runs large through the shoulders. Most customers order one size down.” These descriptions, grounded in data, reduce returns and improve customer satisfaction.

Claude-Driven Returns Reason Classification

When a customer returns an item, they usually provide a reason. But customer-provided reasons are messy: “didn’t fit right,” “too small,” “not what I expected,” “quality issue,” “changed my mind.” They’re subjective, inconsistent, and often incomplete.

Claude (Anthropic’s large language model) can be used to classify these reasons into structured categories, extract nuance, and flag signals that the customer didn’t explicitly state.

How Claude Classification Works

You send Claude the return reason (plus optional context: the product description, the price, the customer’s purchase history) and ask it to classify the reason into a structured taxonomy:

Fit / Sizing (too small, too large, not the right fit, sizing inconsistent with size chart)
Quality / Defect (seam failed, dye ran, zipper broken, fabric pilled, material felt cheap)
Expectation Mismatch (product didn’t match the description, photography was misleading, colour was different, fabric felt different)
Changed Mind (customer changed their mind about the style, decided they didn’t need it, found a better option)
Damage in Transit (arrived damaged, arrived wet, arrived stained)
Wrong Item Shipped (received wrong size, received wrong colour, received wrong product)
Other

Claude can also extract secondary signals:

Confidence level (is the customer’s reason clear and specific, or vague?)
Severity (is this a minor issue or a deal-breaker?)
Repeatability (is this a one-off or does the return reason suggest a systemic problem?)
Customer sentiment (is the customer frustrated, neutral, or satisfied with the return process?)

Example: Returns Reason Classification in Action

Customer return reason: “Not what I expected.”

Claude analysis:

Primary classification: Expectation Mismatch
Secondary signals:
- The customer’s language is vague; they haven’t specified what they expected vs. what they received.
- Possible sub-reasons: the fabric felt different, the colour was off, the fit was different, the quality was lower than expected.
- Given that this is a $120 linen shirt, the most likely issue is either fabric feel (linen can feel rougher than customers expect) or colour accuracy (linen colours shift depending on lighting and dye lots).
Recommended action: Follow up with the customer to understand the specific issue. If it’s fabric feel, update the product description to clarify that linen has a natural texture. If it’s colour, review your product photography and consider adding a note about colour variation.

Another example:

Customer return reason: “Fit issue—ordered size 10 but it’s too small.”

Claude analysis:

Primary classification: Fit / Sizing
Secondary signals:
- The customer is specific about the problem (too small, not too large).
- The customer ordered size 10 (we can cross-reference this with the SKU, the customer’s usual size in other products, and aggregate data on this SKU).
- This is likely a systemic issue if other customers are reporting the same problem.
Recommended action: Check the aggregate return data for this SKU in size 10. If >20% of size 10 orders are being returned for “too small,” flag this with the product team and consider updating the size chart.

Scaling Claude Classification

For a fashion brand processing 100+ returns per week, manual classification isn’t feasible. But Claude can be integrated into your returns workflow via an API:

Customer submits return reason via your returns portal or email.
Reason is sent to Claude API for classification.
Classification is stored in your returns database (D23.io, Superset, or your data warehouse).
Dashboards and reports automatically aggregate classified reasons.
Alerts trigger if a particular reason or SKU crosses a threshold (e.g., “size 10 in SKU-XYZ is returning at >25% for ‘too small’”).

This automation means you’re not waiting for a quarterly business review to realise you have a sizing problem. You’re catching it in real-time, within days of the pattern emerging.

Building a Returns-Centric Operating Model

Returns intelligence isn’t just a dashboard. It’s a operating model—a set of processes, roles, and incentives that make returns data actionable across your organisation.

Roles and Responsibilities

Define who owns returns intelligence:

Returns Operations Manager: Owns the day-to-day returns process, data quality, and Superset dashboard maintenance.
Product Manager: Uses returns data to inform design, fit, and product positioning decisions.
Merchandiser / Buyer: Uses returns data to inform buying decisions, supplier selection, and inventory allocation.
Quality / Supply Chain Lead: Uses returns data to identify supplier quality issues and work with suppliers to improve.
Marketing / Brand: Uses returns data to improve product descriptions, photography, and customer expectations.
Customer Service: Uses returns data to improve return processes and customer communication.

All of these roles should have access to Superset dashboards relevant to their function. A product manager doesn’t need to see logistics costs, but they do need to see fit signals and SKU-level return rates. A buyer doesn’t need to see customer sentiment, but they do need to see supplier-level quality metrics.

Key Metrics and Cadence

Establish a returns intelligence cadence:

Daily: Monitor real-time return volume and any spikes (e.g., “returns jumped 30% today—why?”).
Weekly: Review returns by SKU, reason, and geography. Flag any new patterns.
Monthly: Analyse returns by collection, supplier, and customer cohort. Review action items from previous month.
Quarterly: Deep-dive analysis. Revisit sizing, fit, and quality issues. Plan next quarter’s buying and product strategy based on returns insights.

Feedback Loops and Action

The critical part: returns data has to feed into decisions. Establish clear feedback loops:

Fit issues → Product team updates size chart, updates product description, works with supplier to adjust pattern.
Quality issues → Quality team investigates supplier, flags batch for rework or return, updates supplier specifications.
Expectation mismatches → Marketing team updates product description, photography, or positioning.
Inventory decisions → Buyer uses returns data to inform next season’s order quantities and supplier selection.

Without these feedback loops, returns intelligence is just data. With them, it’s a profit driver.

Measuring Impact and ROI

When you implement returns intelligence, you should measure impact. Here’s how:

Direct Cost Reduction

Return rate reduction: If you reduce your overall return rate from 25% to 20%, that’s a 5-percentage-point improvement. On $5M in revenue, that’s $250K in reduced returns.
Logistics cost reduction: If you optimise your returns logistics (consolidating shipments, using cheaper return methods), you might reduce cost per return from $15 to $12. On 10,000 returns per year, that’s $30K in savings.
Refurbishment cost reduction: If you reduce the number of returned items that need to be marked down or scrapped (by improving quality or fit), you might recover an additional 5–10% of return value. On $250K in annual returns, that’s $12.5K–$25K in recovered margin.

Indirect Benefits

Faster decision-making: With returns intelligence, you can make buying and product decisions in weeks instead of months. That means you can respond to market signals faster and reduce the risk of over-committing to slow-moving SKUs.
Improved customer satisfaction: Customers who receive products that fit correctly and meet expectations are more likely to repurchase and recommend your brand. Reducing returns by 20% might increase repeat purchase rate by 5–10%, which compounds over time.
Better inventory management: By understanding which SKUs are returning at high rates, you can avoid overstocking them and free up cash to invest in better-performing products.
Reduced supplier risk: By identifying supplier quality issues early, you can address them before they affect thousands of units.

ROI Calculation

Let’s say you invest $50K in building returns intelligence (Superset deployment, data integration, Claude classification setup, and initial analysis). You expect:

Year 1 impact:
- Return rate reduction: 25% → 22% (3-point improvement) = $150K in reduced returns
- Logistics cost reduction: $30K
- Refurbishment cost recovery: $15K
- Total Year 1 benefit: $195K
- Net Year 1 ROI: ($195K - $50K) / $50K = 290%
Year 2+ impact:
- Sustained return rate reduction: $150K/year
- Sustained logistics savings: $30K/year
- Sustained refurbishment recovery: $15K/year
- Additional benefit from improved inventory management and faster decision-making: $50K/year
- Total Year 2+ benefit: $245K/year (payback period: ~2.5 months)

For a mid-market fashion brand, this ROI is compelling. And it doesn’t account for the strategic benefits: better product decisions, happier customers, and a competitive advantage in understanding your own products.

Implementation Roadmap and Next Steps

If you’re a founder or operator in the Australian fashion space looking to build returns intelligence, here’s a practical roadmap:

Phase 1: Data Foundation (Weeks 1–4)

Audit your current data sources. Where does returns data live? Is it in your e-commerce platform, a returns management system, your warehouse, or scattered across multiple tools?
Define your returns schema. What information do you capture about each return? (SKU, size, colour, reason, refund status, disposition, etc.)
Set up D23.io or similar. Ingest your returns data into a centralised platform. Normalise and clean the data.
Establish data quality. How accurate is your returns data? Are return reasons consistently categorised? Are SKUs accurately linked to product metadata?

For this phase, you’ll need data engineering support. Many founders partner with an AI automation agency Sydney or a CTO as a Service provider to handle the technical setup. The investment is typically $10K–$30K depending on data complexity.

Phase 2: Visualisation and Exploration (Weeks 5–8)

Deploy Superset. Set up dashboards for core returns metrics (return rate by SKU, by size, by reason, by geography).
Train your team. Make sure product, merchandising, and operations teams know how to use Superset to answer their own questions.
Establish baseline metrics. What’s your current return rate? What are the top 10 SKUs driving returns? What’s the most common return reason?
Create alerts. Set up automated alerts for anomalies (e.g., “return rate for SKU-XYZ jumped 15% this week”).

For this phase, you’ll need analytics and dashboarding expertise. Investment: $5K–$15K.

Phase 3: Intelligence and Classification (Weeks 9–12)

Implement Claude classification. Set up API integration to automatically classify return reasons.
Build fit models. For your top 20 SKUs, analyse returns data to build a fit model (true size, fit profile, shrinkage, consistency).
Create actionable reports. Move beyond dashboards to reports that tell a story: “Why is our summer collection returning at 28%? Here are the three SKUs driving it, here’s the root cause for each, and here’s what we recommend.”
Establish feedback loops. Define the process for how returns insights feed into product, buying, and marketing decisions.

For this phase, you’ll need AI/ML expertise and domain knowledge (someone who understands fashion and your business). Investment: $15K–$40K.

Phase 4: Optimisation and Scaling (Ongoing)

Refine your models. As you collect more returns data, your fit models and classification accuracy improve. Continuously update them.
Expand to new dimensions. Once you have the basics working, explore new questions: “Which customer segments have the highest return rates? Which marketing channels drive the most returnable orders? Which combinations of product attributes are most returnable?”
Integrate with inventory and buying systems. Make returns data available to your ERP and buying systems so that returns insights automatically inform inventory decisions.
Measure and communicate impact. Track the metrics above (return rate reduction, cost savings, inventory improvements) and share them with your team and stakeholders.

For ongoing optimisation, you might maintain a retainer relationship with an AI agency or hire a full-time analytics engineer. Cost: $3K–$8K/month.

Choosing the Right Partner

If you’re not ready to build this in-house, partner with a venture studio or AI automation agency that has fashion and retail experience. Look for partners who:

Have built returns intelligence systems before (ask for case studies and references).
Understand the Australian fashion and retail landscape (local logistics, supplier relationships, customer behaviour).
Can work with you on a fractional or project basis (you don’t need a full-time team, especially early on).
Have expertise in D23.io, Superset, and Claude (or equivalent tools).
Can help you think strategically about how returns intelligence fits into your overall product and operations strategy.

A strong partner will ask you hard questions: “What decisions are you actually going to make based on this data? What’s the cost of not having this intelligence? What’s your current return rate, and what’s a realistic improvement target?” If they’re just selling you software and dashboards, that’s a red flag.

Conclusion: Returns Intelligence as Competitive Advantage

Returns intelligence isn’t a nice-to-have for fashion brands anymore. It’s table stakes.

Your competitors are already mining their returns data. They’re identifying which SKUs are bleeding margin, which suppliers are cutting corners, which fit signals are broken, and which customer segments are most returnable. They’re using that intelligence to make better product decisions, manage inventory more efficiently, and improve customer satisfaction.

If you’re not doing the same, you’re leaving money on the table.

The good news: you don’t need to build this alone. You don’t need a massive data science team or a six-month implementation project. A focused, well-designed returns intelligence system—built on D23.io, Superset, and Claude—can be up and running in 12 weeks and delivering ROI in months.

Start with the basics: centralise your returns data, visualise it, understand your top problem SKUs, and establish feedback loops so that insights turn into action. From there, you can expand to more sophisticated analysis: fit modelling, predictive classification, inventory optimisation, and strategic decision-making.

The Australian fashion brands that will thrive in the next 3–5 years won’t be the ones with the lowest return rates. They’ll be the ones who understand their returns deeply enough to use that understanding as a competitive advantage: better products, smarter inventory decisions, happier customers, and healthier margins.

If you’re ready to build returns intelligence for your brand, reach out to PADISO. We’ve helped fashion and retail brands across Australia implement AI automation for supply chain optimisation and AI automation for retail operations. We can help you design and implement a returns intelligence system that’s tailored to your business, your data, and your strategic priorities. Whether you need a fractional CTO to guide the technical architecture, a venture studio partner to co-build the platform, or an AI automation agency to handle the full implementation, we’re here to help you ship returns intelligence that moves the needle.

Your returns data is already telling you the truth about your business. The question is: are you listening?