What a useful comparison needs
Separate front, side, detail, and on-model images, then align attributes and sales periods. Review both the similarity score and the products themselves.
AI Demand Forecasting · Product Analysis
Understand each product.
Forecast demand for the next season.
ONDO AI FORECAST helps fashion teams plan the next season through similar-product analysis and sales forecasting. It combines product images, attributes, and past sales to inform merchandising and order-quantity decisions.
Front-view selection · attribute tagging · similar-product search · planner review and adjustment
- CollectPrepare products and imagesConnect source data and order sheets
- SelectIdentify front viewsReview candidates and originals
- TagSuggest attributesPlanners refine length, fit, fabric
- MatchFind similar productsImage vectors and attribute weights
- ForecastEstimate sales volumePast sales and trend signals
- ApproveAdjust order quantitiesTarget sell-through and change history
The Problem
Which products can help forecast
demand for a new style?
When product images and sales results are scattered, finding a meaningful comparison is difficult. ONDO AI FORECAST uses images and attributes to find similar products and past results to inform a forecast.
Matching Design
Set the comparison baseline first.
Then consider similarity and sales history together.
The same product can score differently depending on the camera angle and comparison criteria. Select front views and organize attributes by category before comparing.
Start with front views · ONDO AI FORECAST
Select front-view images for visual similarity. Attribute comparisons and past sales then support the quantity forecast, with a planner reviewing each result.
- Front-view selection
- Weighted attribute similarity
- Sales pattern analysis
- Target sell-through calculation
- Confidence scores
- Final decisions by people
One Question, Two Engines
Compare by both image and attributes
Visual similarity finds products that look alike; attribute similarity finds shared length, fabric, and fit. Planners review both alongside past sales.
Visual Match
Image similarity
Vectorize front-view images to find visually similar products. Compare the silhouettes and colors yourself to choose useful candidates.
- Confidence-based front-view selection
- Vector-distance image search API
Attribute Match
Attribute similarity
Weight AI-tagged length, fabric, and fit by category to find comparable products, then connect them with sales history.
- Category-weighted matching scores
- Sales forecasts informed by product performance patterns
Shared design principle
Review AI suggestions and keep a record of decisions
Store AI suggestions separately from planner-approved values and retain change history. Lock approved values; revisions use a cancellation procedure.
Capabilities
From product analysis to sales forecasts
Automate front-view selection, attribute tagging, and similar-product search, then analyze sales patterns and trends. Planners review and adjust the forecast before using it.
Image similarity
Select front-view images
Automatically select the most likely front-view image from your own and competitors' product photos, based on a confidence score.
Image vectorization API
Send an image file and receive a vector representation for similarity search.
Similar image search API
Send an image file and receive similar products with similarity scores.
Sales forecasting
AI attribute tagging by category
Suggest product attributes such as length, material, and fit from images and product information.
Attribute-weighted product matching
Compare tagged attributes with category-specific weights to rank similar products from past seasons.
Sales pattern reports
Summarize shared traits of strong and weak sellers in natural language to explain the patterns.
Trend-fit analysis
Score each product against seasonal trend research by silhouette, material, color, and other attributes.
Order quantity from target sell-through
Calculate a suggested order quantity from forecast sales and target sell-through, then let a planner adjust it.
Stage controls and change history
Track uploads, tagging, and quantity approval. Lock approved values to prevent accidental changes.
TOOLSAI tools by role
Upload product data to begin demand forecasting
See how AI uses product details and images for automatic tagging and demand forecasting. Products and figures shown are examples.
01
Import team
AI sales volume forecasting
Upload a product image to let AI tag its attributes and forecast seasonal sales volume from similar products' sales history. The team can use the result when reviewing quantities.
1Upload dataOrder sheet · product images
2Automatic attribute taggingAI attribute extraction
3Demand forecastSales volume · trends
Upload source dataUpload an Excel order sheet and a ZIP of product images to match and import them automatically.
📊
Order sheet (XLSX)
Click or drag · XLSX up to 50 MB
XLSXBLANC_27E_order.xlsx6 rows
🗂️
Product images (ZIP)
Click or drag · ZIP/JPG up to 500 MB
ZIPBLANC_27E_images.zip12 images
📋 Combined product list Order sheet ↔ product images matched automatically · 6 SKUs
ImageV-STYLE · colorCategory · itemSizeMaterial compositionWHLS($)RETAIL($)Status
V136584815001 BLKKnitwear · crewneck pulloverXS–XLWOOL 70 / NYLON 30128295Matched
V136684798137 OWTKnitwear · round neckS–XLWOOL 100135310Matched
V136684799944 TURKnitwear · cable pulloverXS–LCASHMERE 30 / WOOL 70142330Matched
V132713134242 COSCoat · wrap coatS–LWOOL 80 / NYLON 20268620Matched
V137084800466 AMAKnitwear · waffle textureXS–XLCOTTON 60 / WOOL 40119275Matched
V137084801946 SIEKnitwear · Fair Isle patternS–LWOOL 100156360Matched
Filename rules V-STYLE_COLOR automatically match each image to its order-sheet row.
Automatic attribute taggingAI extracts and tags attributes from product images automatically.
Product image
From the image, AI extracted these attributes
ItemCoat96%
LengthMidi92%
Collar typeNotched lapel89%
ClosureWrap · belt87%
SleeveSet-in sleeve90%
FitRelaxed84%
ColorCamel95%
MaterialWool blend88%
Demand forecastForecast sales volume, selling period, and trends from tagged attributes.
Forecast by size SKU allocation
XS8%
S24%
M38%
L22%
XL8%
Weekly sales trend Peak: second week of March
2027 SS · peak sales in the second week of March
⚠Compared with similar camel midi coats, sales rise sharply in the first four weeks, creating a risk of early stockout. — Consider increasing the order by 12% (to 1,400 units) and allocating more units to sizes M and L.
Automated workflow demo
ANALYTICSData analytics · DASHBOARD
Turn data into decisions
Explore customer purchasing behavior, reviews, and competitor products in one dashboard. This is an example interface; the figures are not live operating data.
Company-wide Reference: Jun 2026 · current month
TOTALOfflineOnlineCross-channel
Purchasing customers
81K people
▼ 32.5% · −39K
Repeat customers
53K people
▼ 26.2% · −19K
First-time buyers
8,110 people
▼ 27.4% · -3,060
Churned customers
444K people
▲ 13.4%
Member sales
₩29.29B
▼ 24.9% · −₩9.7B
Purchase identification rate
58.1%
▼ 0.4%
Gender and age distribution Men 23,133 (28.5%) · women 58,060 (71.5%)
~29
30–34
35–39
40–44
45–49
50–54
55–59
60–64
MenWomen
Customer journey First purchase → repeat purchase → churn
First-time customers
11K people → 53K repeat customers
Reactivated customers
17K people
Value per purchase
₩232K
Top stores By purchasing customers
StoreCustomers · sales
1NORDÉ online store39,793₩18.2B
2NORDÉ Gangnam5,894₩3.1B
3NORDÉ Hyundai Pangyo1,634₩900M
4Atelier M Myeongdong1,210₩600M
5Blanc Lotte Jamsil982₩500M
Top products By sales
Style code · productSales · customers
1NDPK5A201IV · Cotton piqué polo₩88M586
2NDSW5B118NV · Signature knit₩79M468
3ATJU5B062BK · Tailored jacket₩70M124
4BLPA5A235BG · Wide-leg slacks₩68M326
5NDDR5B405WT · Pleated dress₩65M274
Customer analytics · nine reports including brand summary, cross-purchases, buying patterns, and top customers
NORDÉ 4,325 reviews · STCL 290 · since SS24
By customerBy style code
Score · overall rating
4.7
★★★★★
Reviews · weekly trend
W-5 → W · rising over the last four weeks
Sentiment · analysis
4,325REVIEWS
Positive 2,812 · 65%
Negative 575 · 13.3%
Neutral 938 · 21.7%
[NORDÉ] Signature cotton piqué polo
NDPK5A201IV · ONDO AI review summary
Key purchase driversFabric feel · fit · color — “a weight that works year-round”
What customers likedSoft feel, easy everyday fit, and little change after washing
What customers dislikedRuns smaller than standard sizing — consider one size up
AI-predicted score88.4 / 100 · high likelihood of repeat purchase
ONDO AI summarizes and classifies 4,325 reviews · supports top-review views and reruns
Competitor · URBNOur brand · NORDÉCategory · shirtsMaterial · all⬇ EXCEL DOWN
URBN-10%
URBN-15%
URBN-20%
URBN-10%
Automatically collect competitor product listings and organize them by category, list and sale price, fit, material, and country of origin.
How It Works
From product data to a demand decision
Organize product details and images, then let AI tag attributes. Compare products by appearance and attributes, use past sales and trends to forecast demand, and have a planner review and adjust the result.
Collect
Automatically gather product details and images from the company and its parent organization.
Select and tag
Rank front-view candidates and tag category-specific attributes. A planner reviews the originals and suggestions.
Vectorize and match
Turn front-view images into vectors and tagged attributes into weighted measures for separate similarity calculations.
Find patterns
AI summarizes shared traits of strong and weak sellers in natural language.
Forecast quantity
Combine similar products and trend fit to estimate sales volume and a likely range. Check forecast errors against historical sales.
Review and decide
A planner checks and adjusts the suggested order quantity calculated from forecast sales and target sell-through.
Correct
Planners can revise automatic selections and forecasts before approval. Later changes follow a cancellation procedure and remain in the history.
Front-view selection and attribute tagging happen before similarity calculations. A planner reviews originals and suggestions to set the comparison baseline.
Adoption Criteria
Assess search and forecasting
with your own products
Check whether search results fit the comparison task and how forecasts differ from actual sales. Include the planner's review and revision steps when defining adoption scope.
Catalog
Prepare product data
Connect product IDs, categories, images, attributes, and sales history. Confirm access rights and rules for missing or duplicate data.
Image Selection
Review front-view selection
Check whether selected images are truly front views and fit the purpose. Assess candidate scores separately from measured classification accuracy.
Search Relevance
Assess similar-product search
Compare top search results with products chosen by planners and see how image and attribute weights change the ranking.
Operation
Set processing volume and cadence
Design batches and search around product and image counts and refresh cycles. Measure processing time, errors, and retries on real data.
Forecast Validation
Measure forecast error
Compare forecasts with sales after the forecast date, considering seasonality, stockouts, discounts, sales periods, and the number of comparable items.
Data Boundary
Define internal data connections
Plan around managing images, attributes, and sales history internally. Check transmission scope and access rights for any external API or data.
Decision
Keep the planner's final review
Review matches, sales forecasts, and target sell-through together before adjusting a suggested order. Record approvals and cancellations separately.
Scope & Validation
Define the data and conditions
behind a demand forecast
Prepare product images, attributes, and sales history, then agree on planner review and approval rules. Validate error and scope with real data.
| Inputs | Product details and IDs, images, category attributes, and period-by-period sales history |
|---|---|
| Operating scope | Set collection, selection, and tagging batches based on product and image counts and refresh cadence |
| Candidate score | Review front-view candidates and match scores; decimal precision is not an accuracy measure |
| Tagged attributes | AI suggests category-specific length, fit, material, and pattern; a planner can revise them |
| Order calculation | Suggested order quantity = forecast sales ÷ target sell-through (greater than 0, up to 1); apply order units and rounding separately |
| Workflow state | Track upload → tagging → quantity approval; lock approved values for viewing only |
| Change history | Link AI suggestions and planner-approved values by product ID |
| Adoption scope | Confirm brands, categories, sales channels, and available data during adoption |
- Configure AI front-view selection and URL-pattern parsing for the source data, then check misses and incorrect selections.
- Agree on duplicate and representative image rules so comparisons share a consistent baseline.
- Forecasts depend on the number of comparable products, sales periods, stockouts, and discount history. A statistically validated confidence interval requires separate evaluation.
- Lock approved order quantities against accidental overwrites and require an explicit cancellation to revise them.
- A planner reviews and adjusts automated selections and forecasts before approval; later changes follow cancellation and history rules.
A high matching score does not guarantee sales. Forecast sales and suggested order quantities support decisions; a planner makes the final call.
Related work — Customer analytics dashboard
Our customer analytics work helped shape how analysis is brought into everyday workflows.
Roadmap
Link images and attributes
to widen the evidence behind forecasts
Visual similarity search and attribute-based sales forecasting run in separate workspaces today. Next comes combined matching and integration with internal systems.
- STAGE 0 · Complete
Design image and attribute processing
Designed front-view selection, category tagging, and storage; registered product details and images from the company and its parent organization.
- STAGE 1 · Complete
Introduce tagging and forecasting screens
Teams now use screens for uploading data, tagging attributes, setting sales periods, and reviewing quantities.
- STAGE 2 · Current
Run image and attribute matching
Visual similarity search and attribute-based forecasting each run automatically every week. Planners review and refine the results.
- STAGE 3 · Next
Integrate the two engines
We plan combined image-and-attribute matching, better duplicate handling, and broader links to internal search, forecasting, and order workflows.
Could AI demand forecasting work for your products?
Let's explore the path from tagging image attributes to forecasting with sales history.