01
Unified customer signal
CRM, commerce, service and behaviour data resolve into one usable profile.
Less wasted acquisition spend and cleaner attribution.
Micro-segmentation for a fashion house
120 micro-clusters instead of one generic audience.
FirstVal customer — fashion brand, Indonesia
Fashion retail, marketing and merchandising
- 53%
- Increase in up-sells
- 36%
- Improvement in ROAS
- 16 weeks
- Data to deployment
01 — The intro
Where the story starts
A large fashion brand wanted to optimise marketing spend by building targeted campaigns off its own customer data.
The catalogue held 1,400+ seasonal styles across 12+ categories in nine cities — millions of possible mix-and-match recommendations.
Existing campaigns were generic. Return on ad spend was running 2.4x higher than target, trimming margin and leaking the sales funnel because messages were not personalised across channels.
- Catalogue
- 1,400+ styles
- Market footprint
- 9 cities
- Spend gap
- 2.4x
Seasonal products and accessories spread across more than 12 categories.
Customer and inventory signals fragmented across locations and channels.
Return-on-ad-spend level reported above the desired target before personalisation.
02 — The challenge
Customer data was fragmented across inventory, CRM, support, billing, surveys, social listening, app data and session streams — multiple formats, all in silos.
- —No unified customer view across platforms or channels.
- —Styles and SKUs managed separately with only basic demand forecasting.
- —Overlapping style preferences made hard clustering a poor fit.
- —Cost of acquisition rising while margin fell.
Improve
Sales and return on ad spend
Reduce
Cost of acquisition
03 — The solution
A unified data pipeline feeding two clustering systems — style and customer — connected so each explains the other.
- 01Unified extraction and transformation pipeline for real-time and batch machine learning.
- 02Trend model using CNN encoders to read influencer imagery and find SKU gaps.
- 03Style clusters built on dress type, fit, fabric, colour, occasion, price, historical sales, seasonal trends, up-sells and bargains — soft clustering with fuzzy c-means and graph methods because styles overlap.
- 04Customer segmentation on location, age, life stage, frequency, recency, basket size, total spend, prior style purchases, colour and fabric preference, complaints, returns, app usage, discount sensitivity, loyalty and brand affinity.
- 05120+ micro-clusters identified and matched style-to-customer.
- 06In-app ads and recommendations tuned in real time from click-stream intent: search type, filters, wishlist, saved items, cart and dwell time.
Impact
Up-sells up 53%, ROAS up 36%.
04 — Value logic
How operating change reaches financial value
02
Soft micro-segments
Customers and styles can belong to overlapping preference groups.
More relevant targeting without forcing buyers into one rigid segment.
03
Intent-aware recommendations
Search, wishlist, cart and dwell signals tune the next offer.
Higher up-sell rate and return on advertising spend.
05 — Feature delight
What people actually felt
01
Style creator model
Drove recommendations and up-sells, and fed back into segmentation as a feature in its own right.
02
Segmentation clusters
120 micro-clusters turned generic targeting into messages people recognised as theirs.
06 — Ground impact
What moved, and by how much
53%
SKU up-sells
Micro-segments and the style model working together.
16 weeks
End to end
Data collection through model creation, test and deployment.
36%
ROAS increase
Spend moved to segments that convert.
Trend spotting
Production planning
Social listening and influencer imagery informed SKU planning.
07 — Measurement
How the result stays accountable
Commercial lift
Compare up-sell rate and ROAS by exposed micro-segment against the prior generic campaigns.
Efficiency
Track acquisition cost, margin contribution and media allocation by cluster.
Learning loop
Monitor segment drift and feed campaign response back into style and customer models.
08 — CSAT score
Scored by the customer, A+ to D
Cost reduction, higher sales and customer satisfaction — the brand's three stated priorities.
- A+
- Customer service and customisation
- A
- Segmentation AI
- B+
- Style creator
- A
- Overall impact
09 — More customer stories
Keep reading
Transport & Logistics
Automating the Bill of Lading
90,000 bills a day, read in seconds instead of minutes.
Health & Life Sciences
Voice Rx — the 100 second prescription
Give the doctor back the minutes that belong to the patient.
Retail & Consumer
Contact-less skincare analysis
Dermatology-grade advice without a sales rep in the way.