ELARA·CORTEXTHE SAVINGS REPORT · POWERED BY THE .LEKOLA CORTEX ENGINE
Hard numbers, computed from public cost data and your own published figures, against the tools in your stack right now. Every engine claim has a measured receipt. Every assumption is stated so your CFO can re-run the arithmetic.
Run on the public CVRPLIB benchmark (100 to 512 customers) against the published best-known solutions, equal conditions, receipts on disk. We publish the losses too. That is why you can trust the wins.
| Tool (in your stack today) | Who runs it | Measured result | Verdict |
|---|---|---|---|
| Google OR-Tools (guided local search) | Embedded in countless fleet and TMS stacks; the default solver | OR-Tools 3.6 to 16 percent gaps at 15 to 20 s vs Elara 0.7 to 4.6 percent at 4 to 5 s; on time-window tests OR-Tools drops customers on 4 of 5 where Elara serves all | Elara wins 9/9 |
| Z3 exact optimisation | The mathematical ground truth (small instances) | Elara matched the proven global optimum 5 of 5 | Elara = optimal where provable |
| PyVRP (academic HGS library) | The leading research library; a library, not a deployed service | PyVRP 1.6 percent mean vs Elara about 2.7 to 3.2 percent at equal 5 s | PyVRP leads on raw quality. Named next target |
| VROOM (open-source dispatch) | Many open TMS deployments | Windows binding failed in our harness; not yet measured | honest gap, queued |
| Google Route Optimization / Maps APIs | The metered incumbents | $5 to $30 per 1 000 requests, metered, batch-oriented; no millisecond replan, no certificates | Elara: flat price + ~1 ms replan + verifiable answers |
The sentence a CTO can check: Elara beats Google OR-Tools, the default solver deployed in fleet and TMS stacks today, matches the proven optimum wherever optimality is provable, and is the only one with millisecond re-planning and certificate-verified answers, published, reproducible, in an afternoon (PyVRP, a research library, currently leads on raw quality, our named next target).
Today your recovery lives in Sabre Ops or Recovery Manager, Amadeus (Optym) or Lufthansa Systems NetLine, plus controllers rebuilding by hand under pressure. The leak is the reactionary delay that spreads while the day is rebuilt.
Money. Shave 10 minutes of reactionary delay off just 50 disruption-affected flights a day: 500 min x $100.76 per block-minute x 365. One conservative scenario, one carrier.
Fuel. The fuel share of those same minutes: 500 min/day x $33.06 fuel-per-block-minute x 365. Roughly 2 million gallons of jet fuel a year not burned in taxi queues and holds.
Customers. Delay-minutes that never reach a passenger: fewer missed connections, fewer EU261 and DOT compensation events, on-time percentage up. The metric your NPS follows.
Mechanism: recovery in ~1 s vs minutes, measured on 10 real route networks (/proof)
Scale it yourself: your disrupted flights/day x minutes shaved x $100.76 x 365. European carriers: use EUROCONTROL's ~EUR 100/min. The engine's part, valid rotations back in about one second, is measured and certificate-verified.
Pick a carrier (real route-map sizes). Ground an aircraft. Watch the re-flow.
The precedent here is not ours. It is UPS's own published result. ORION, their in-house route optimiser, is the most famous proof in logistics that this exact mathematics pays.
The precedent (UPS's own number). ORION's published savings from route optimisation: 100 million miles and 10 million gallons of fuel a year. Published by the buyer, not the vendor.
Money (Amazon-scale arithmetic). ORION saved 6 to 8 miles per driver per day. Take a conservative 2 miles/day across ~390 000 delivery drivers x $2.26/mile x 312 days. At ORION's own 6 to 8 miles the figure is $1.6 to 2.2 bn. We state the conservative end.
Anchors: ATRI 2024 $2.26/mile + ORION's published per-driver miles
Fuel. The same conservative 2 mi/driver/day is about 243 million miles a year not driven. At ATRI's $0.48/mile fuel line that is about $117 m of fuel, and the emissions line your ESG report wants.
Anchor: ATRI 2024 fuel $0.48/mile (per-mile proxy; substitute your fleet's burn)
What we add beyond ORION-class: ORION took a decade to build and computed routes overnight. The .LEKOLA CORTEX engine is an API that beats Google's deployed solver 9 of 9 on the public benchmark, and re-plans a disrupted day in about 1 millisecond, all day long. Plus a flat fee instead of the metered per-request bill ($5 to $30 per 1 000 requests).
Your dispatch is in-house; your map bill is not. Uber's own S-1 disclosed roughly $58 m paid to Google Maps over 2016 to 2018, and deadheading (driving without a passenger) is the utilisation leak every platform fights.
Money (the bill). The migration offer: bring the invoice, pay a quarter. Routing, matrices and re-plan on a flat engine instead of metered calls.
Anchor: Uber S-1 Google Maps disclosure; metered pricing public
Money (the street). Re-matching in about 1 ms instead of on a timer recovers idle and deadhead minutes. One utilisation point across a million driver-hours a day = your revenue-per-driver-minute x 600 000 minutes. Your number, our speed.
Mechanism measured: ~1 ms re-plan on a live city fleet feed, 49 percent shorter than as-dispatched (/proof)
Customers (and engineers). Every answer is a certificate your team verifies. The difference between a black box and a system you can put in the dispatch path.
Certificate format public at /solve-api
Yes, to Google itself. The most interesting reader of this page works there.