African Ports Dominate Global Congestion Rankings: Bejaia, Conakry Hit 11+ Day Delays

The Analysis

Spread the map out and read the list as geography rather than as a league table, because a league table tells you who is worst and geography tells you why. Portcast's weekly congestion snapshot for 24 to 30 May puts six African ports inside the ten most congested on earth. Bejaia, on Algeria's Mediterranean coast, leads with a median wait of 11.89 days. Conakry, Guinea's Atlantic gate, follows at 11.68. Beira, on the Mozambique channel, is at 9.49. The number I keep coming back to is not the level, it is the one attached to it: a deterioration of 5.45 days inside a single week, which means that seven days earlier this port was sitting at 6.44 days. That is not a port having a bad week. That is a port that moved.

Be precise about what is being counted, because the unit decides the whole argument. Median vessel waiting time, as this snapshot defines it, runs from the moment a vessel reaches anchorage to the moment it is successfully berthed. So 11.89 days is a queue that exists before a crane touches your box. Discharge, yard time, gate-out and the haul inland all sit on top of it, and not one of them is inside the number. There is a second caveat in the same methodology worth underlining twice: vessels can be held offshore instead of queuing at anchorage, which means a port can show a low reading while its ships are still waiting, just not on the anchorage the model is watching. The quiet end of this list is therefore quieter on paper than it is on the water.

And the quiet end is where the surprise sits. Qingdao at 2.06 days, Ningbo at 1.50 and Shanghai at 1.30 are filed under medium congestion with wait times described as rising, while the major American and European ports show minimal delays. Turn that around and look at it from the other side. The ports carrying the overwhelming share of the world's container volume are, this week, the ones with nothing to report. The ports carrying a rounding error of it are the ones where a ship waits close to twelve days for a berth. Ten years ago I would not have expected a list shaped like this, and the reason it deserves ten minutes of your attention is that the shape has been drifting this way for years rather than arriving in May.

Six of ten on one continent is not bad luck, and it is worth saying why before anybody reaches for an explanation about a particular country. Ports on this coast share a structure rather than a set of problems. A market served by one or two container gates has concentrated its exposure whether or not anybody ever decided to. Those gates have a small number of deepwater berths, a handful of services calling, and a hinterland reached by one road and one rail line, and when a queue forms there is nowhere for it to drain to, because the next ship with space on it is a week away rather than a day away.

At a hub with fifteen services a week, the same number of delayed ships clears in days, because there is always another berth window and another booking. The recovery time after a shock is set by how many services call, not by how good the terminal is, and that is a fact about the network rather than about the week.

Ten years ago this was a different place, and I do not say that nostalgically. Ask me then where congestion cost importers the most money and I would have pointed at the high-volume coast of China, at the United States West Coast, at the big European gateways, which is to say I would have pointed at the places where your boxes actually were. That is where the investment went, and it worked. Berths got deeper, cranes got taller, yards got automated, and the terminals at Qingdao, Ningbo and Shanghai today would not be recognisable to somebody who last walked them in that decade. The proof is sitting in this snapshot: 1.30 days at Shanghai. The work was done, and the constraint moved.

What has really changed is the network, not those ports. More of the world's import demand now sits behind gateways served by four or five sailings a month instead of four or five a week, with one road inland and one realistic alternative port, and congestion does not follow volume at all. It follows the ratio between what arrives and what the place can clear, and that ratio is worst precisely where the slack to absorb a bad fortnight is thinnest. This is why a list of the world's ten most congested ports can be dominated by places that between them handle a small share of global throughput. Throughput tells you where the cranes are. It does not tell you where your ship waits.

Convert the number into the unit your planning runs on, because hours are more honest than days. Bejaia at 11.89 days is 285 hours. Conakry at 11.68 is 280. Beira at 9.49 is 228. Shanghai at 1.30 is 31. Bejaia is a little over nine times Shanghai, measured the same way, in the same week, by the same method. A ship that arrives off Bejaia on a Tuesday morning is, on the median, still off Bejaia late the following weekend, and it has not yet started discharging. If your production plan assumes anything like the Chinese number for a port on this list, the gap is not a few days of lateness. It is a different lane.

The deterioration is the part I would act on and it deserves its own paragraph. A level tells you what to plan against; a deterioration tells you the plan is now wrong. Bejaia moved 5.45 days in seven days, from 6.44 to 11.89. Divide the change by the starting point and the queue grew by about 85 per cent inside one week. Nothing about a port's physical capacity moves 85 per cent in a week, so what moved that fast was the arrival pattern, the service pattern, or something upstream that pushed a batch of ships into the same few days. If you were planning against 6.44, you are planning against a figure that no longer exists, and the very next snapshot, not the one after it, is the one that tells you whether this was a step change or a spike.

Then think about the difference between a level and an event, because importers are good at events and bad at levels. An event has a name, a start and an end, so somebody briefs it, a decision gets taken and the plan reopens. A level has none of those things, so nothing triggers, and instead the extra days quietly become how long the lane takes and get absorbed into buffer stock permanently. Eleven point eight nine days is not an event. It is the ordinary experience of the median ship at that port in that week, and if you are buying cover against it as though it were an exceptional seven days, you are funding a standing condition with a one-off justification, and you will still be funding it next year.

Look now at the other end of the list, where the level is low and moving. Qingdao 2.06, Ningbo 1.50, Shanghai 1.30, all medium, all described as rising. A low number that is rising is a different planning problem from a high number that is stable, precisely because your plan is built on the level and the level is shifting underneath it. Two days is something you can live with. Two days that was 1.6 last quarter and is heading somewhere is a figure you have to re-plan against every month, and the cost of that is not the two days. It is the re-planning, the arguments and the cover you keep adding because you have never quite agreed what the number is. At the ports you depend on, watch the change as well as the state.

The median is the ship in the middle, and the ship in the middle is never the one that ruins your month. Your pain comes from the ship at the back. The same methodology behind this snapshot describes seventy-fifth and ninetieth percentile waits alongside the median, and the honest way to plan is against the ninetieth, because the tenth ship is the one that misses the production window and the customer date. If all you carry is a median, you are planning on the vessel that behaved. Ask whoever sells you visibility for the tail, and if they will not give it to you, keep it yourself, because berthing timestamps are yours in a way no report will ever be.

Here is the one piece of good news in all of this, and it is worth taking. This queue is on the water. A ship at anchorage can be counted, and an eleven-day queue gives you eleven days of warning that your boxes are late, if you happen to be watching the anchorage. Most importers are not watching the anchorage. They are watching the carrier's schedule, which is built around the berth window the ship was supposed to get, and those two things diverge by exactly the depth of the queue. That divergence is visible, free and public, and it is the cheapest early warning on this entire list.

Now the arithmetic, with every assumption on the table so you can argue with the method rather than with me. Suppose you import through Beira. Suppose you move thirty forty-foot containers a month, which is a flow of one box a day. Suppose your planning sheet and your contract both assume a port wait of two days, which is roughly what the Chinese end of this list delivers. Suppose ocean transit to the anchorage is twenty-eight days and the haul inland to your distribution centre is five. Your planned door-to-door is thirty-five days. The observed median wait at Beira is 9.49 days, so the same lane actually runs forty-two and a half days. The lane is about 21 per cent longer than the plan, and it is longer for the median ship, not for the unlucky one.

The carrying cost of that delay is the small number, and I want to show you why before anybody quotes it at you. The difference between plan and reality is 7.49 days. At one box a day that is seven and a half boxes of extra cover. Suppose each box carries 40,000 of goods value. That is roughly 300,000 of working capital you did not plan to hold, and at an assumed 8 per cent cost of capital it costs about 24,000 a year to carry, every year, for as long as the level holds. If the level is standing rather than temporary, that is not an event cost to be booked against a bad quarter. It is a permanent line in your cost to serve, and the honest place for it is in the price of the product rather than in the supply chain team's variance column.

So the arithmetic worth running is the alternative, and it needs its own assumption. Suppose the second gate you are considering, further along the same coast, adds four days of ocean transit but has a median wait of three days instead of 9.49, and suppose the inland leg from it is the same five days. Its door-to-door is forty days against forty-two and a half, so it is two and a half days faster on the median before you count any benefit from the two waits not being correlated. Two and a half days at one box a day is two and a half boxes, about 100,000 of working capital released and about 8,000 a year of carrying cost. Against that, price the qualification: one project, one inland tender, a quarter of somebody's time. One of those two costs you money once. The other costs you money every year.

Note which assumption in that sum is the fragile one. It is the inland leg. If the road from the second gate is eight days rather than five, the arithmetic reverses and the second gate is slower door-to-door despite the shorter queue at the anchorage. This is the mistake I have watched people make for years. They change the port and keep the corridor, and then they wonder why nothing improved. Before you call any second gate a backup, make your inland carrier draw both routes, and check how much of the distance the two share. If they converge for more than half the way to your distribution centre, you have bought a different port and the same risk.

Redundancy on a coast like this is genuinely available, and that is unusual, so use it. Six of the ten places on this list are congested in the same week, but the reasons are largely local to each of them: a berth, a yard, a stretch of road, a service pattern. Local causes are not correlated with each other, which is exactly the property that makes a second gate real insurance rather than decoration. Compare that with a situation where two gates share one cause, and you will find that the second gate fails on the same morning as the first. Ask what would have to be true for both of your gates to close at once. If the honest answer is two unrelated things, the second gate is worth qualifying. If it is one thing, it is not.

The decade question is the one that decides whether you are managing this or paying for it. If you are signing a three-year supply agreement into a market served by one of these gates, the congestion level is a term of the deal, not an item on a risk register. A contract priced on a two-day port wait at a gate with a twelve-day median is mispriced on the day it is signed, and the cost of that arrives later as buffer, as missed windows and as arguments about who should have known. Put the observed median into the price, put the review into the calendar, and the level stops being a surprise you absorb and becomes a number somebody owns.

And then the thing that does not fix it, which is where a lot of money gets spent. A berth is roughly a three-year object: design, tender, build, commission, and at the end there is something you can stand on and photograph. The network behind a port is a fifteen-year object made of decisions that do not belong to the port at all: how many services call, who operates the rail path, whether the road was upgraded in the same budget cycle, whether the shift roster covers the peak. A gateway can add a berth, finish it on time and still hand you the same door-to-door service three years later, because what constrains you is the number of sailings and the speed of the box leaving. When you read an expansion announcement, ask how many services a week are expected and what happens to the container after the quay.

One more thing about measuring it, because it is the difference between arguing and knowing. Keep your own series. Your boxes, your gate-out dates, your berthing dates, one column in a spreadsheet and ten minutes a week, and you will have something no report will ever give you, which is a tail measured on your own volume. A provider that reports an average is telling you about the container that behaved. The one that costs you money is the tenth one. Nobody will hand you that distribution unless you ask for it by name, and plenty of providers will quietly substitute an average when you do.

The traps are almost all variations on measuring the wrong end. Planning against the carrier's berth window rather than observed anchorage-to-berth means an eleven-day queue never appears in your numbers at all. Working from the median rather than the tail means your plan describes the ship that behaved. Treating a low reading at a big hub as proof of no delay ignores the vessels held offshore instead of at the anchorage. Assuming a deterioration is a spike without waiting for the next Monday means you re-plan twice instead of once. Buying cover against a level and calling it event cover means you will still be paying for it in two years. And redrawing the port while keeping the corridor means the project delivers a different name on the bill of lading and nothing else.

Put the map away. What this snapshot shows, and what I would rather remember than any single number on it, is that the geography of waiting has quietly inverted against the geography of volume. The terminals that carried the world's growth spent fifteen years getting better at the water side and largely succeeded, and the queue went to look for somewhere else to live. It found the places where volume is thin, services are few and the road inland is one road. Ten years ago here was another matter, and the map I was taught to read would not have shown me this. What has really changed is the network. The port is only the part of it you can see from the water.

↑ Back to the key points

— By Tomás Ferreira