Orkestra by Data-Hat AI
Demand, meet supply.
The demand signal retail never captures, and the layer that reasons and acts on it. Orkestra builds a demand profile for every store and turns it into allocation and replenishment decisions, continuously.
Store 22 · demand profile
liveWould sell 302 · sold 205 · gap on size 38
one demand truth
Live with
350+ stores
Fashion brand
Global maison
Watchmaking
Multi-geo network
Beauty retail
The blind spot
Demand is everywhere. Almost none of it reaches your decisions.
60–80%
of people who walk into a store buy nothing. The reason is never captured, so it never reaches planning. It isn't in your POS, because nothing was sold.
~20%
of all the data you already collect is ever actually used. The other 80% sits idle across ERP, POS, BI and WMS.
So supply is planned half-blind, wrong assortment, wrong allocation, wrong replenishment.
The cost
Overstocking and understocking aren't opposites.
Every missing unit is demand you paid to acquire. Every excess unit is capital that could have gone elsewhere. Optimize one side and you worsen the other, the real cost is the sum of both.
Inventory balance · one SKU, one store
Empty facings, the full-price sale you never see, and never record.
Units above capacity, markdowns, dead stock, cash locked on the shelf.
$50M
Understock 8%
$4.0M
Overstock 15%
$7.5M
Total exposure 23%
$11.5M
Based on published industry benchmarks for retail inventory loss.
Why it persists
You've bought a tool for every piece. None of them close the loop.
Planning executes a plan set a year ago. ERP tracks stock. BI reports what already happened. POS and WMS see transactions, not intent. Each owns a slice, none of them share demand, and none of them act.
The problem isn't a missing tool. It's a missing layer.
“40 looked for WOOL-COAT-BLK-38. 6 sales, 10 left.”
POS
saw the six that sold
ERP
stock was accurate
BI
reports it on Monday
WMS
nothing was requested
Planning
set last season
Five systems answered. Not one of them acted.
Missing. No system owns the call.
Nothing here turns what the floor saw into a decision, an owner and a deadline. This is the layer Orkestra is.
Coat re-sized, moved and re-priced by Monday
The system
Orkestra. Three layers. One system. Demand becomes action.
Not another tool on your stack, the intelligence and action layer that sits above it.
Pulse
Captures the demand no system sees, the moment it happens.
Store 22 · 14:07 · Signal
A demand signal you've never had, captured live, not reconstructed later.
StoreDNA
Unifies every signal into one demand profile per store.
Store 22 · demand profile
Size curve
Would sell vs did sell
Silos end. Every store gets one demand truth, what it would sell, not just what it did.
Cockpit
Reasons across the network, and acts.
Rebalance: store 22 → store 9
acted, not suggested
Replenish: store 14: +120 units
acted, not suggested
Decisions get made and executed continuously, not observed on a dashboard and forgotten.
Pulse
StoreDNA
Cockpit
StoreDNA and Cockpit overlap at intelligence, that overlap is the handoff, not a seam.
Together, they are the layer your stack never had.
The impact
The right stock, everywhere. And the revenue you were losing, back.
Specialised agents do the work, assortment and allocation at store-SKU level, autonomous replenishment, multi-horizon forecasting, and substitution when the exact reference can't be supplied.
Before · one plan, every store
Same assortment everywhere: uneven results.
After · every store its own demand
Different assortment per store: consistent results.
Full-price sell-through
Net working capital, as % of sales
EBITDA margin
Right stock·Right place·Right time·Right price
Industry DNA
Orkestra isn't generic AI. It's trained on the constraints of your category.
The team
Built by people who ran this at Farfetch and Zalando.
Data-Hat AI was founded by Kshitij Kumar (KK), former Chief Data Officer at Farfetch and Zalando, alongside retail and AI veterans. Orkestra is the system we wished existed when we were on the other side of the table.
About Data-Hat
Kshitij Kumar
Founder · Data-Hat AI
Chief Data Officer
Chief Data Officer
The pilot
Live in 48 hours.
Your first agent is running within two days of kickoff, no re-platforming, no data migration project.
Hour 0
Connect your existing feeds, ERP, POS, WMS as they are.
Hour 24
Pulse goes live on the store floor; advisors start capturing.
Hour 48
Your first allocation and replenishment decisions land.