Six systems in production at large companies

Challenge

Orders from the app, the website and the call center lived in three separate systems, and store stock levels updated once a day. Pickers got orders for items that were no longer on the shelf, and a noticeable share of orders went out with substitutions.

Solution

We built one order platform: a single source of stock for every channel, event-driven updates from point-of-sale systems and picking routes that follow each store’s planogram. The picker app and the customer app run on the same API.

We rolled it out region by region, 100-200 stores per wave.

Outcome

Order picking is three times faster, and all 1,800 stores run on one platform. The customer app holds a 4.8 rating in the app stores.

Challenge

Equipment data went into a separate SCADA system at each plant. Analysts saw the picture a day late and learned about most failures only after the line had stopped.

Solution

We brought telemetry from 40,000 sensors into a single stream and built a digital model of the equipment with early-warning rules.

Failure prediction models are trained on maintenance history and run inside the plant network, with no internet access.

Outcome

Data travels from sensor to model in 0.8 seconds. Unplanned downtime is down 17%, and maintenance crews get a warning before a failure, not after it.

Challenge

Planners built courier routes in spreadsheets the evening before. Once volume reached 120,000 deliveries a day, the plan could no longer keep up with cancellations, traffic and new orders.

Solution

We built a routing service that recalculates routes throughout the day, taking delivery windows, vehicle load and road conditions into account.

The courier app gets changes instantly, and planners handle exceptions instead of every single route.

Outcome

The system plans 120,000 deliveries a day, and fleet mileage is down 12%. Service availability is 99.97%.

Challenge

The monthly bill run for 9 million subscribers took more than a day, and every new tariff plan meant changes to the core and months of testing.

Solution

We split billing into rating, calculation and invoicing services connected by a shared event bus. Tariffs are now defined in configuration, not in code.

Before cutover, we ran three billing cycles in parallel on the old and new systems and reconciled every invoice.

Outcome

Bill run time is down 46%. There hasn’t been a single billing error since launch, and new tariffs ship without touching the core.

Challenge

Settling a claim took five days on average: customers brought documents to an office, and an adjuster checked every case by hand, even the simplest ones.

Solution

We launched online claims. The customer files a claim with photos in the app, the system checks the policy, the documents and fraud signals, and settles simple cases on its own.

Complex cases reach an adjuster with the data already collected and a risk score attached.

Outcome

For simple claims, the payout arrives in 15 minutes instead of 5 days. 64% of claims are settled with no human involved, and NPS is up 21 points.