Everything Worked, Except The Bit In Between

On 27 March 2008, Heathrow Terminal 5 opened for business. £4.2bn, the largest free-standing building in the country at the time, and six months of trials behind it before a single fare-paying passenger walked through the door. By the end of that first weekend around 28,000 bags were sitting in temporary storage, more than 500 flights had been cancelled across three days, and one aircraft had left for Paris with nothing in the hold at all.

What interests me about T5, eighteen years on, isn't that it went wrong. It's what specifically went wrong.

The terminal worked. The baggage system worked, and BAA said so at the time. They had a point, it was rated at 12,000 bags an hour. The staff turned up. What failed was everything in between. Handlers couldn't log into their baggage profiles. Some couldn't park in the staff car parks. Others couldn't get through security screening to reach the equipment they were meant to be operating. Willie Walsh, British Airways' chief executive at the time, later told MPs that four days of training hadn't been enough, and that because the building wasn't finished, staff had been trained in a different building to the one they'd end up working in.

So they hadn't skipped testing. They'd tested for six months. What they hadn't done was run the whole thing as a full dress rehearsal, with the actual people, in the actual building, under actual load, and watch what happened at the joins.

Anyone who has delivered a large programme will recognise that shape immediately. The build slips, scope grows because it always does, and the go-live date is the one thing that can't move. So the testing window gets compressed and a risk-based call is taken about what to cut. That call is usually made by sensible people with a fair grasp of the risks, and it's usually the least-bad option on the day. It's also, almost invariably, the integration testing that goes first. It's the hardest thing to schedule, it needs the most people in one room at once, and it's the only bit that no single team can just get on with alone.

In the aftermath of T5, the Commons Transport Committee took evidence from both organisations and reported that November. It found serious failings on both sides, concluding the whole thing "could, and should, have been avoided through better preparation and more effective joint working". It specifically named insufficient communication between BA and BAA as a root cause. Walsh and BAA's chief executive Colin Matthews gave the committee materially different accounts of the decision to open at all. The detail that has always stayed with me is that MPs recorded themselves as astonished that joint meetings between the two organisations only started happening as an afterthought, once things had already gone wrong. Anyone who has worked on enterprise scale change won't be astonished at all, which is rather the point.

Nobody properly owned the space between BA and BAA's accountabilities. So when it failed there was nobody to fix it, just two chief executives giving Parliament different versions of the same week. Ownership stopped being a job and became an argument.

I've spent twenty-six years in payments and commerce and I've ended up with a fairly unglamorous view of where the money goes. The expensive failures almost never start inside a system. They start at the seams between systems. And seams are exactly the thing nobody owns, nobody budgets for, and nobody puts on an org chart.

Individual systems mostly work. They were bought or built by capable people solving a real problem and they behave the way you'd expect. What doesn't behave is the handoff, the point where an order or a transaction or a customer record crosses from one system into another, and it turns out the two sides don't quite agree on what's just been passed between them.

This isn't a new observation. Mel Conway made the point in 1967 that organisations build systems mirroring their own communication structures: if two teams don't talk, the things they build won't either. It's been quoted at me in enough architecture reviews that I will never forget the principle. What I've learned to add over the years is practical rather than theoretical. Conway explains why seams are where they are. He doesn't tell you which one to go and look at on Monday morning, or what to check when you get there.

When I ran group payments technology at Argos and Homebase, then part of Home Retail Group, payment processing had historically been built into each order management system separately, shaped around the batch process that settled it. The group started as a high street retailer, so it sourced POS equipment to take card payments. Then it had the idea to sell over the phone, so it built something new and bolted on a way to take card details over the phone. Then the internet became interesting and yet another set of systems arrived to take payments through a browser. Every one of those calls was sensible on the day it was made. But together, and compounded over time, they meant that changing a single payment type reached into more or less every fulfilment system we had.

The failures I remember weren't systems breaking. They were two systems, each configured correctly, by different people, at different times, quietly holding different definitions of the same thing, with nothing running between them to spot the difference. That sort of gap doesn't announce itself. It sits there until somebody reconciles it by hand, or until it turns up in a settlement report a few weeks later.

Which brings me to something I wrote a few weeks ago about fintech filling the hole gambling left in Premier League sponsorship, and a point in it I don't think I gave enough room to breathe.

Revolut are now Manchester City's official back of shirt partner, men's and women's. But the shirt is the visible part. Revolut Business is being integrated into the club's financial infrastructure, with Revolut Pay to follow as a checkout tool. Revolut's own line was that they aren't just putting a logo on a shirt, they're integrating into the heart of the fan experience. Arsenal and Deel is the same shape, and the sequence is worth noting: Deel became Arsenal's official HR platform partner in December 2025, and the sleeve deal followed the May after. Vendor first, sponsor second.

I should be straight here. I've no idea whether either club has priced what it would cost to unwind those arrangements, and it would be lazy to assume they haven't, they employ people for exactly that. I'm a supporter looking in from the outside, not an adviser to either. And none of this is a football problem anyway. I've watched the same shape play out in retail, banking and transport, and I was responsible for a version of it myself for the best part of a decade. If anything football clubs have a better excuse than most of us, because their entire institutional instinct is pointed at eliminating seams on the pitch: the understanding between a full back and the winger, the ten yards where midfield and forwards are supposed to join up and usually don't. Nobody is asking a director of football to lie awake over the join between ticketing and CRM.

What interests me is simply that the pattern I spend my working life chasing almost certainly exists in the sport I love, whether anyone inside it has ever given it a name. Look at what gets created the moment those contracts are signed: a commercial relationship and an operational dependency, tied together in one agreement, with a boundary between "sponsor" and "supplier" that doesn't obviously belong to either function. It works fine while everyone's happy. It only becomes interesting if the relationship ever needs to end.

The connected fan work I described in that piece is the same problem with more moving parts. A credential linking your ticket to what you buy in the ground, to how you got there, to what the club learns about your matchday. Genuinely valuable, but hardly any of that value sits inside any one system. It's in the handoffs between ticketing, concessions, transport and CRM. Four systems, usually four owners, and a set of boundaries between them that generally has none. Which is why, when I asked how well clubs really know their own cost base, I wasn't being rhetorical. It's hard to see your cost base clearly when the data describing it has to cross four boundaries to assemble itself and nobody owns the crossings.

None of this is an argument against good integration engineering, by the way. We're excellent at the technical half of a handshake. Field mappings, payload translation, schema versioning, error codes, retries, idempotency. A well-built interface is a genuinely rigorous piece of work and the people who do it are very good at it.

But a handshake isn't only a data contract. There's a second half that rarely gets the same attention. Who owns this boundary once it's live. Who gets called when it drifts. Who arbitrates when the two sides turn out to mean different things by "settled", or "complete", or "active customer". We specify the payload down to the individual field, leave the ownership undefined, then act surprised when the failure shows up somewhere nobody was watching. Give the ownership half even a fraction of the rigour we already give the mapping half and you'd end up with reliable data transfer and a secure seam, rather than reliable data transfer and a boundary nobody's minding.

I'm not suggesting anyone assigns a named owner to every boundary in a large estate. The coordination cost would be absurd and most of them genuinely don't matter. The same risk-based judgement that compressed the testing at T5 applies here, and it's the right instinct. But there's a real difference between a seam you've looked at and consciously decided is low risk, and a seam you don't know exists. The first is a decision. The second is luck. And the boundaries that cause genuine damage are almost never the ones somebody assessed and parked, they're the ones that were never on anybody's list to assess.

The thing that's stayed with me from the Home Retail Group years is that we weren't blind to any of this. When we looked at those payment initiatives, the organisational side of the change tended to come out as the smaller part of the problem next to the technology. And yet the thing we kept writing down, in one form or another, was that we needed to agree who owned end-to-end payments across channels and systems.

We'd spotted it. We just filed it as a smaller job than it turned out to be. That's not a criticism of anyone I worked with, I probably called it the same way at the time. It's more that a programme treating the technology as the hard part and the ownership as an afterthought has quietly decided what kind of problem it's got before it's finished looking. Seams fail because nobody owns them, and no amount of platform spend assigns an owner.

The reason I've been thinking about this more lately is AI. Every board wants an AI story and I understand why. There's real capability there and I use these tools daily, including whilst researching and writing this article. I'm not sniping from the sidelines.

But the risk everyone names isn't the one that worries me. "Rubbish in, rubbish out" is true, and it's also a conversation that ends the moment it starts, everyone nods and nobody changes anything. The sharper problem is that an agent doesn't simply need clean data, it needs systems that agree with each other. It will act across a boundary where two systems hold different definitions of "settled" or "complete", and it will do it at machine speed, thousands of times, with nobody owning the crossing and nobody watching it happen.

A person working that seam builds up a feel for it. They notice the thing that looks slightly wrong and they go and check. That instinct is doing far more load-bearing work in most organisations than anyone has written down, and it's precisely what we're automating away. We're removing the last person who understood the join and replacing them with something considerably faster that has never been told the join is there.

I'm working that into a proper paper at the minute, on the risk of going straight to AI before the fundamentals underneath it are stable. Skipping pre-season and expecting to win the league. T5 is in it, for what I hope are obvious reasons.

There's a live example close to home, and I mention it with some affection rather than any criticism. Argos soft-launched its marketplace this month, running on Mirakl, with around eighty sellers trading. The trade coverage is quite open about where the joins currently sit: items bought from a third-party seller can't be returned to an Argos store, they can't be paid for with Argos Pay, gift cards or Nectar points, and the payment goes to the seller rather than to Argos. One write-up put it neatly, the store network is the big opportunity but it isn't fully connected yet.

Every marketplace launches with a boundary somewhere, and you have to start. But that store network is the single biggest structural advantage Argos has over a pure-play competitor, and at the moment it doesn't reach across the marketplace seam. Neither does the payment estate that I spent nine years building. Those are precisely the joins where value either flows or leaks, and if agentic commerce is the next thing anyone bolts on, it will be making decisions across those same boundaries at speed. Whoever owns them over the next couple of years decides which way it goes.

In the meantime, if you want to find where value is leaking in your business, I wouldn't start by auditing your systems. I'd start by asking who owns the boundaries between them. If it takes more than a few seconds to get an answer, or two people give you different ones, you've probably found it.

So, where's the seam in your organisation that everybody works around and nobody owns?

SOURCES

1.   Heathrow Terminal 5 opening. House of Commons Transport Committee, “The opening of Heathrow Terminal 5”, Twelfth Report of Session 2007–08, HC 543, published 3 November 2008. Available at publications.parliament.uk. Source for the committee’s findings, the “better preparation and more effective joint working” conclusion, the differing accounts given by Willie Walsh and Colin Matthews, the training evidence, and the finding on joint meetings.

2.   T5 baggage and flight figures. Hansard, House of Commons debate, 31 March 2008, for the bags held in temporary storage and the £4.2bn investment figure. Flight cancellation totals and the Paris departure from contemporaneous reporting of the opening week.

3.   Conway’s Law. Melvin E. Conway, “How Do Committees Invent?”, Datamation, April 1967.

4.   Manchester City and Revolut. Manchester City and Revolut partnership announcement, February 2026, covering the back of shirt partnership across the men’s and women’s first teams and the integration of Revolut Business and Revolut Pay.

5.   Arsenal and Deel. Arsenal announcement of Deel as Official HR Platform Partner, December 2025, and as Official Sleeve Partner from the 2026/27 season, announced May 2026.

6.   Argos Marketplace. Trade coverage of the Argos Marketplace soft launch, August 2026, including ChannelX and specialist marketplace commentary, for the Mirakl platform, the approximate number of Trusted Sellers, and the current position on in-store returns, Argos Pay, gift cards and Nectar points.

7.   Argos ownership. Sainsbury’s announcement of the sale of Argos to Swift Partners, July 2026.

This article draws on publicly available information and the author’s own professional experience. It contains no confidential information from any former employer.