Meridian Twin

A digital twin of a simulated airline. Real airports and distances, a delay model trained on 7 million real DOT flights and tested on July 2026, planning assumptions for everything else.

Deloitte AI Innovation · Pratyush Mudgal

Meridian Twin

An AI digital twin that shows an airline what disruption will cost, finds the cheapest way out, and leaves every decision to a person.

$97ksaved on an average storm day, 28% less than waiting it out
77%of the flights it ranks riskiest ran late, against 28% of all flights in July 2026
$5.7Ma year from storm recovery alone for a 300 aircraft airline (base case)
36 simulated storms · 612,038 real July 2026 flights · EIA fuel prices to Oct 6, 2026github.com/bypxr/meridian-twin
The problem, in dollars

Airlines lose money in three places

Storms

$347k

what an average storm day costs when the airline simply waits it out

With the twin: $251k. A saving of $97k.

Delays

$418k

a day in routine delay costs: about 59 late flights, 72 minutes each, at $98.41 a minute

Ranks the riskiest flights 2 hours ahead.

Fuel

−$300k

a day at October 6 jet fuel of $4.34 a gallon, with fares unchanged. 42 of 49 routes lose money

Fares must rise 14.4% to earn 6% again.

One simulated airline: 61 aircraft, 178 flights a day, hubs at ATL, ORD and DFW2 / 5
Show, not tell

The product is live, right here

  1. Fuel. Set the October 6 fuel price and watch the routes that lose money turn red.
  2. Risk. A model tested on 612,038 real July 2026 flights ranks every departure two hours ahead.
  3. Storm. Close O'Hare for four hours. The twin tests thousands of recoveries and shows the trade between money and passengers.
  4. Ask. Claude explains the plan and drafts the passenger notice. It cannot approve a plan. A person does.
Everything runs in this page. Press Esc to come back to the slides.
Simulator, delay model, fuel layer and Claude advisor in one page3 / 5
Impact in dollars

What it could save an airline each year

Airline sizeStorm recoveryRangeEach 1% cut in late minutes
61 aircraft (this twin)$1.2M$0.3M to $2.9M$1.5M
150 aircraft$2.9M$0.7M to $7.1M$3.8M
300 aircraft$5.7M$1.4M to $14.3M$7.5M
800 aircraft$15.2M$3.8M to $38.0M$20.0M
How the storm number is built30 storm days a year, times the $97k saving per storm day, times the 40% a real operations team captures. The range uses 15 to 50 days and 20% to 60%.
Fuel is a risk, not a savingEvery $10 on oil moves a 300 aircraft airline's yearly profit by $159M. The twin shows which routes lose money first and how far fares must rise.
The simulated airline scaled by fleet size. A pilot replaces these estimates with measured figures4 / 5
The ask

One hub. 26 weeks.
A saving finance can sign off.

DiscoverWeeks 1 to 3
ConnectWeeks 4 to 9, the client's operations, crew and cost data
ModelWeeks 8 to 15, rebuild the twin and retrain on their flights
ShadowWeeks 16 to 21, run beside real controllers
AssistWeeks 22 to 26, a person approves every plan

The same pattern then carries to other sectors: simulate the operation, find the best response, let AI explain it, keep a person in charge.

Energy: grid outages and fuel costHealthcare: bed and theatre recoveryGovernment: transit and emergency responseTelecom: outage triage
Pratyush Mudgal · Questions welcome5 / 5
1 / 5