The Churchill Arms in Kensington
Lab: Observed behaviour over 82 cycles in London
When American Express replaced the Marriott Bonvoy Brilliant Card’s annual dining benefit with a $25 monthly restaurant credit in September 2022, the structure created an obvious question for anyone spending substantial time outside the U.S.: how reliably would it work abroad?
We didn’t set out to test it. We just used the cards.
From November 2022 through March 2026, we ran two Brilliant cards against normal spending in London. Across 82 monthly credit cycles, that meant $2050 of dining credits were available. We captured $2039.
The shortfall over 41 months was $10.94.
That result required almost no change in where or how we spent money. The cards went to ordinary cafés, restaurants and pubs we were already using. Over time, the record became a useful test of something broader: the difference between the stated value of a credit-card benefit and the value you can actually realize in normal life.
The benefit worked without creating the spending
The underlying transactions are unremarkable. Pret, Itsu, Gail’s, Joe & The Juice and Nando’s appear repeatedly, alongside pubs including The Bolingbroke, the Ship in Mortlake, White Horse on Uxbridge Road, Churchill Arms and the Waterfront by the river.
There are some less obvious merchants too, including a school-adjacent café and a hockey club bar/café. Across the full dataset, roughly 15–20 venues recur with any regularity on top of conventional restaurant spending.
We didn’t choose them because of the Brilliant credit. They were places where money was going to be spent anyway. A $25 monthly credit can easily produce $25 of apparent value if you deliberately find something to buy before it expires. That isn’t the same as receiving $25 of value from spending that would have happened without the benefit. Here, the latter is overwhelmingly what happened.
Pret was one of the London merchants that repeatedly triggered the Brilliant dining credit.
What 82 credit cycles showed
The credits don’t necessarily arrive as neat $25 entries. They often post in pieces corresponding to individual transactions: $3.58, $7.62, $10.22 and similar amounts.
Matching those credits back to the underlying charges revealed a remarkably stable pattern. A relatively small group of cafés, restaurants and pubs accounted for much of the credit usage, with other merchants appearing as our normal spending changed.
Most of the familiar merchants worked repeatedly. Some didn’t.
Wetherspoons was the clearest example of why the physical appearance of a merchant tells you relatively little about how a transaction will behave. Sometimes it triggered the credit and often it didn’t, while other pubs that looked no more like conventional restaurants cleared reliably. A small number of other venues were similarly inconsistent. Repeated use didn’t turn them into reliable Brilliant-credit merchants.
The useful information therefore came from the transaction history rather than from trying to decide whether somewhere looked sufficiently restaurant-like. Once a merchant had repeatedly generated credits, we knew what had happened there before.
There was still a small behavioral effect
Saying we made no effort at all would now be slightly too strong. We didn’t change where we went in order to use the credit, and we didn’t create transactions to exhaust it. But experience did occasionally influence which card came out at the point of payment.
If a place had reliably triggered the Brilliant credit before, the Brilliant card was the obvious card to use. If it hadn’t, we might use something else. That is a much smaller behavioral change than going somewhere because a credit is about to expire. The underlying purchase already existed. We were choosing the payment instrument after learning how particular merchants had behaved.
It also explains why realization became so consistent without requiring any meaningful monthly management.
The credits usually appeared within a few days
The statement credits didn’t post immediately. Across the record, most appeared a few days after the underlying purchase, with a noticeable clustering around day four.
Early on, that delay made individual transactions slightly harder to interpret because we didn’t yet know which purchases had qualified. After the same merchants had appeared several times, the uncertainty largely disappeared.
By mid-2023, most months were following a familiar pattern. Small transactions at many of the same cafés and pubs would gradually consume the two $25 monthly credits. Other restaurants came and went according to where we happened to eat.
There was no monthly ritual of checking how much remained and finding somewhere to spend it. Normal activity was doing almost all of the work.
$2050 available, $2039 actually used
The complete record is more useful to me than saying that the Brilliant card offers “up to $300” of annual dining credits. Across 41 months and two cards, the theoretical maximum was $2050. Actual statement credits totaled $2039, leaving $10.94 unused.
That means we realized almost all of the nominal value while largely allowing the benefit to attach itself to spending that was going to happen anyway. That is about as close as I can get to treating the dining credit at face value.
It doesn’t follow that every Brilliant cardholder should value the benefit at $300 a year. Someone who rarely eats at qualifying restaurants, or whose normal merchants don’t trigger the credit reliably, could reach a very different number. The point is that the value of the benefit is ultimately determined by what happens when it meets your existing spending.
In our case, 82 monthly cycles provide a fairly robust record.
The contrast with other Amex credits
This is also why I don’t value every American Express credit the same way. Some require deliberate intervention. A monthly U.S.-only credit can be difficult to use while outside the country, forcing you either to find a workaround or let it expire. I’ve done both the analysis and the work where the underlying benefit was valuable enough to justify it.
The Brilliant dining credit has been almost the opposite. Ordinary spending in London consumed $2039 of a possible $2050 without us having to manufacture purchases or substantially change where we ate. Once we learned which merchants behaved reliably, even the small uncertainty around merchant coding largely disappeared.
That makes the Brilliant dining credit unusually easy for me to value. It isn’t worth close to face value because American Express says it is. It’s worth close to face value because, over 82 opportunities to use it, that’s what actually happened.