Logistics and e-commerce
Systems, integrations and AI for networks where every hour in the sorting hub and every delivery attempt counts.
Context
A parcel passes through the sender's warehouse, the sorting hub, linehaul, the last mile and a locker or pickup point. Data is generated at each of these stages, but the systems of the sender, the operator and their partners rarely see it at the same time, and customer service often learns about a problem from a phone call. At volumes counted in millions of parcels, each such gap carries an operational cost.
We build applications for couriers and sorting hubs, integrations between operator, sender and partner systems, and AI models for forecasting and ticket handling. We start from the indicators that operations teams already report: on-time and first-attempt delivery, cost per stop, locker utilisation and the number of status queries. We connect to existing systems through APIs and events, without replacing what works.
Challenges in the industry
A peak planned months in advance
Black Friday and the pre-Christmas period put pressure on sorting hubs, linehaul and lockers at the same time. Decisions on temporary agency staffing, subcontractor fleets and extra shifts are made months ahead, and a forecast based on last year's average does not see sender campaigns, shifts between channels or the wave of returns after the holidays.
Last mile cost
A growing share of parcels goes to lockers, so last mile cost depends on stop density and the number of failed attempts. Each failed attempt means a repeat trip, an additional customer contact and the risk of a complaint. Routes planned once in the morning do not reflect changes during the day.
Occupancy of lockers and pickup points
Compartment occupancy changes throughout the day and depends on compartment size, dwell time and returns sent from the same locker. A full locker means a parcel has to be redirected, while an underused location still generates cost. Without a forecast for each location, it is hard to plan the servicing sequence and network expansion.
Status queries and returns
A large share of customer service contacts are questions about parcel status, even though the information is already in the systems. Returns create a separate, reverse flow: label, transport, inspection and a decision on what happens to the goods, with a different owner at each stage.
Where we deploy software and AI
Volume forecasting for sorting hubs and lockers
The model accounts for seasonality, campaigns reported by senders, the promotion calendar and weather. The forecast covers daily and weekly horizons as well as the peak period, broken down by sorting hub and locker. At the start it runs alongside the planner's own forecast, so the results can be compared.
Staffing demand planning
From the volume forecast we calculate the demand for workers, lines and docks by the hour, taking working time regulations into account. The shift supervisor approves or adjusts the proposal. The system does not assess employees and does not allocate tasks based on their performance.
Routing and ETA during the day
The route is recalculated when conditions change: traffic, a failed attempt, a new parcel in the area. The ETA is based on the history of the route, the area and the parcel type, and the delivery window narrows during the day and feeds into recipient notifications.
Applications for couriers and sorting hubs
Applications for scanners and phones covering scanning, proof of delivery by code or photo, and exception handling, integrated with operator systems. They work with poor coverage and synchronise data once the connection is restored. Location is collected only during the shift.
AI agents for tickets, complaints and returns
The agent checks the status, the scan history and the order data, then prepares a reply. The customer knows they are talking to an AI assistant. The agent closes simple cases on its own and passes disputed ones to a customer service agent together with the full data, and every decision is logged.
Integrations with senders, partners and KSeF
Data exchange with e-commerce platforms, marketplaces, carriers and pickup point operators through APIs instead of files sent by email. Parcel data feeds settlements with senders and invoices in KSeF (Poland's National e-Invoicing System). Every integration has data validation and monitoring.
Regulatory requirements are taken into account from the analysis stage
NIS2 and the amended KSC Act
The amendment to the Polish National Cybersecurity System Act (KSC) has applied since 3 April 2026. Postal operators, including couriers, are important entities if they are at least medium-sized enterprises. The requirements on risk management, business continuity, ICT supply chain security and incident reporting must be met by 3 April 2027. As a provider we are part of that chain, and we supply the client with the documentation it needs.
GDPR, location data and employee monitoring
Courier location and recipient addresses are personal data processed at large scale. Processing employee location is on the UODO (the Polish data protection authority) list of operations that require a data protection impact assessment (DPIA), and monitoring is subject to Article 22(3) of the Labour Code, including the obligation to inform employees two weeks in advance. We train models on aggregated or pseudonymised data.
AI Act and systems concerning employees
AI systems that allocate tasks based on employee behaviour or assess their performance fall into the high-risk category, with obligations applying from 2 December 2027. Volume forecasting and routing are not in that category. Assistants that talk to customers must disclose that they are AI from August 2026.
KSeF and transport documents
The largest taxpayers have issued invoices in KSeF since 1 February 2026, and penalties for violations apply from 2027. From 9 July 2027, authorities across the EU must accept transport data from certified eFTI platforms. We design integrations so that this flow runs as a single data stream.
How we start
Workshop with operations and IT
We agree on the indicators that determine the result and on the owner on the operations side. We identify data sources in the systems of the operator, senders and partners, and the places where the data diverges.
Data review and choice of the first area
We assess the completeness and latency of data from scanners, sorters and courier applications. We select an area with a clear owner and a measurable outcome, most often volume forecasting or ETA.
Pilot in shadow mode
The solution first runs on a selected part of the network, alongside the current process, without affecting decisions. The assessment of dispatchers and shift supervisors is a condition for further rollout.
Scaling before the peak
We extend the solution to the whole network, run load tests against peak volumes and freeze changes several weeks before Black Friday. We provide maintenance under an SLA or transfer it to the client's team.
Frequently asked questions
With a single process, for example volume forecasting. We connect only the sources that process needs, and the shared data layer grows with each subsequent deployment.