A Control Tower for a Tin Corridor

An airport does not run on the honour system. It does not ask each pilot to self-report position and hope the reports add up. It runs a control tower: several independent channels — radar, transponder, radio, a controller at the window — resolved continuously into one picture, with automation for the routine and human attention reserved for exceptions.

A single export consignment out of the eastern DRC passes through nine or ten counterparties on its way to a vessel at an Indian Ocean port: mine logistics, road carriers, clearing agents, customs on both sides of at least one border, port agents, shipping lines. Each reports in its own format, on its own timeline, in its own interest. The operator — the only actor with an end-to-end stake — is asked to resolve all of it into a spreadsheet that is obsolete by Friday. The question we kept putting to our client, a publicly listed tin producer operating in the eastern DRC, was: what would it take to run this like a control tower?

01Challenge

The status quo has three symptoms any operator will recognise.

The reconciliation tax. Every week, somebody spends hours merging spreadsheets and rebuilding a status board that is stale by the time it circulates. The work is skilled, invisible, and thrown away weekly.

Silent failures. A truck breaks down where there is no signal. A document is rejected for a missing stamp. Caught in hours, these are survivable; caught in days, they compound into demurrage and missed vessels. The cost of a silent failure is rarely the failure — it is the distance between the event and its discovery.

The ambiguity of "where is it?" When the answer depends on which of five people you ask, you are not operating a supply chain. You are running a rumour network.

No single provider is at fault. Collectively, they produce a view of the cargo that the operator does not own and cannot query. Full accountability, partial picture — that is the problem.

02Diagnosis

The natural response is to buy a tool. We have watched that fail, because the problem is not tooling — it is information architecture. A control tower is four things in a specific order: multiple independent channels, none trusted alone; one canonical picture, continuously reconciled, where a contradiction between channels is an alert, not a data artefact; automatic cascades for routine events; and human authority at the exception layer.

Every failure above maps to a missing property. The design principle follows: cargo visibility is not a data-collection problem. It is a reconciliation problem. Collection is the first fifth of the work; resolving partial, self-interested reports into one operational truth is the rest.

03Approach

Absorb every provider in its native format. Truckers send spreadsheets; clearing agents send PDFs; dispatchers use WhatsApp; drivers send voice notes. The system takes all of it into one data model — shipment, legs, events — hourly, with a change log. The invariant: it fills gaps, it never overwrites. Provider data-entry errors are common enough that an aggressive overwrite policy spreads mistakes faster than they can be caught. The unstructured half — a six-month comment blob, an "arrived border 07:30, departed 09:15" message — is turned into dated, structured events by language-model extraction. Anything ambiguous is flagged for a human.

Capture at the source. Providers report against their own interests; the report you most need is the one you are least likely to receive. So the operator runs its own channel: mine staff, drivers and agents at the border posts and ports record events directly, through a mobile app that works offline and syncs when a signal returns. When the two channels agree, confidence rises quietly. When they disagree, an alert fires. And because the people who witness an event write its record, the old version drift — what the driver said, what the agent filed, what finance saw — collapses into one version.

Documents travel with the cargo. Customs declarations, certificates, waybills — each filed against its shipment, findable in seconds. The two most expensive moments in the corridor — a truck held for a missing paper, and a month-end reconstruction of a shipment's file — largely stop happening.

One live picture, cascades underneath. The ops team reads a single view: where every lot is, what just happened, what is flagged — in plain English, refreshed every 30 to 60 minutes, around the clock. Any truck, container, licence or lot can be pulled up by reference with its full history. Every event triggers the next steps automatically — recalculated ETAs, updated commercial views, an exception when a status sits still too long. The team's attention goes to exceptions, not to chasing.

04Outcome

The numbers read as a ledger, not a brochure: seven years of continuous corridor data. More than 7,000 export lots tracked lot-by-lot since 2019. Over 6,800 outbound truck journeys analysed and 1,700 trucks profiled across dozens of transporter companies. 45,000-plus documents indexed and searchable. Close to 50 live report views across 17 pages.

The more important outcome is what the data revealed once it was clean enough to question. The median dwell at the corridor's main border crossing more than doubled between 2022 and 2026. A structural degradation, not an incident — and invisible before the system existed, because nobody could prove it. Once the lot-level baseline existed, the trend surfaced on its own. An operator negotiates differently when the conversation is "our median dwell has more than doubled in four years, here is the record" rather than "we feel it is getting slower." Visibility is not the dashboard. It is the argument the dashboard lets you make.

Day to day: the weekly reconciliation meeting became an exception-handling meeting, and the gap between a cargo event and the operator knowing about it collapsed from days to hours.

05Lessons

1. A control tower is an information architecture, not a screen. The view is the tip of the iceberg; the reconciliation underneath is the work.
2. Triangulate, and capture at source. One channel is a rumour; two is a cross-check. The cheapest second channel is your own people, on a phone that works offline, writing the record as they witness it.
3. The baseline is the product. A multi-year view of border dwell told the operator something no single cargo could. The systems that earn their keep are the ones that let you ask a question you did not know you wanted to ask.

Next case study: The Road In — Every Shipment from Supplier to Mine →