In 2008 three researchers followed 49 people through a holiday of about a week and sent each of them a message every single day, asking how happy they had been in the previous 24 hours.
When the holidays were over, they asked one more question, this time about the whole thing.
The answers had a shape that is worth knowing if you run a property, because they say something uncomfortable about which parts of a stay survive in the head of the person who paid for it, and which parts quietly disappear.
A holiday nobody could recall day by day
The study is Kemp, Burt and Furneaux, published in Memory & Cognition, and the design was about as close to real life as this kind of research gets: real holidays, daily self reports written while it was happening, one retrospective judgement at the end.
Two things came out of it.
The length of the holiday had no effect on how it was judged afterwards.
The participants could not reconstruct their own day by day pattern either, even though they were the ones who had reported it, one evening at a time.
The first result has a name, duration neglect, and it had already appeared in a much colder setting: in 1996 Redelmeier and Kahneman found that retrospective evaluations of an unpleasant experience correlated almost zero with how long that experience had lasted.
It showed up again in 2025, in a study published in the Journal of Hospitality & Tourism Research that tracked tourists in real time during actual trips and then measured how they judged the trip afterwards, where duration once more had no significant effect.
For a hotel that is worth sitting with for a second, because the night is the unit you sell, price, forecast and get judged on internally, and it is close to invisible in the memory of the guest.
Take two guests checking out on the same morning, one who stayed two nights on a city break and one who stayed nine with two children and a washing machine problem somewhere in the middle.
The second guest lived through more than four times as much of your hotel, and when they sit down to write about it they will both produce a few lines built out of two or three moments each.
Seven nights do not buy seven nights of remembered experience, and a fourth night at a discount does not add a quarter more goodwill.
If duration does not make it into the final judgement, something else has to.
What the peak and end rule actually says
The peak and end rule says that when somebody looks back on an experience they lived through, the judgement they produce is built almost entirely out of two moments, the most intense one and the last one, while most of what happened in between gets flattened into a rough average or dropped altogether.
The name comes from work by Daniel Kahneman and Barbara Fredrickson with colleagues at the start of the nineties, and it took its familiar shape in what Kahneman did with Donald Redelmeier in 1996, on patients going through an unpleasant medical procedure who reported their pain in real time and then rated the whole thing afterwards.
Those retrospective ratings tracked the worst moment and the final moments, and ignored almost everything else, including how long it had all taken.
Two words are doing a lot of work there, so they are worth taking apart on the timeline of your own property rather than on a laboratory chart.
What a peak looks like in a hotel
The peak is the single most emotionally intense moment of the stay, in either direction, and it is rarely the thing you spent the most money on.
It can be the waiter who noticed that a child had left a toy in the restaurant and brought it up to the room before anybody had asked for it.
It can just as easily be the August night when the air conditioning stopped and the phone at reception rang out.
Both of those are peaks in the technical sense, because both stand a long way above the flat line of everything else that happened that week, and the flat line is what the guest will struggle to remember at all.
What counts as the end
The end is the last stretch of the experience, the part the guest is still living through as the whole thing closes.
Where that actually falls inside a hotel is a lot less obvious than the schedule suggests, and it is the piece most properties get wrong on paper, so I have given it a section of its own further down.
Why memory works like this
Memory does not keep a recording of the stay running from arrival to departure.
It keeps a small number of samples and rebuilds the rest around them when somebody asks, which is exactly why the participants in the holiday study could not reproduce their own day by day curve even though they had written it themselves, one message at a time.
That sampling is also where the duration goes.
If an experience gets stored as a handful of snapshots plus a rough sense of how it felt overall, the number of hours it took to produce those snapshots never gets written down anywhere, so a long stay and a short stay turn up at the review form carrying about the same amount of luggage.
What the rule does not say
Two readings of it are wrong, and both are expensive.
The first is that the middle of the stay stops counting once you have a peak and a decent goodbye.
It counts, and on long mixed experiences the average quality of the whole thing climbs back into the equation with some force, which is the single most useful correction in this literature and the subject of the last section here.
The second is that one spectacular gesture repairs a weak base.
A peak is a multiplier on a decent base, never a substitute for one, and a wow moment bolted onto a mediocre operation is a wow moment sitting next to a mediocre average.
Nothing in the rule says the earlier parts of a stay are wasted effort either.
It describes which moments the memory keeps, not which moments deserve your budget.
What the evidence actually supports
Enough of this holds outside a laboratory to be worth managing, and not enough of it holds to be treated as a law.
The early demonstrations were all on unpleasant experiences, which is a genuine limitation, and Do, Rupert and Wolford extended the finding to pleasant ones in 2008 with support.
In tourism it held up too: a 2019 study in the Journal of Heritage Tourism looked at 187 visitors on guided historical walks and found peak and final moments carrying more weight than the episodes the visitors themselves described as the most memorable.
The 2025 tracking study found the same two positions doing the work, with peaks, lows and endings of emotional valence significantly influencing loyalty after the trip, while physiological arousal did nothing of the sort.
The restriction sits at the other end of the evidence, in the studies almost nobody quotes, and rather than bury it here I have put it in the last section where it can be read properly.
What the rest of it leaves you with is a practical problem, because a peak is not something you can simply buy more of.
A peak is a gap, not a level
A peak only exists because of what surrounds it.
Raise the quality of everything by the same amount and you get a better hotel with no peak in it, since the distance between the best moment and all the others has not moved at all.
Picture two properties in the same town.
The first one is genuinely well run, the rooms are in good order, nobody at the desk is ever short with anybody, the buffet has everything on it and gets refilled properly, and the guest leaves after three nights with nothing in particular to tell anyone.
The second has smaller rooms and a thinner breakfast, and on the first evening somebody at the desk hears a guest mention a birthday in passing and does something about it two mornings later without being asked.
On any checklist you care to build, the first property wins every line.
The second one is the property that gets described out loud at a dinner table three months later.
Improving the average and building a peak are two different investments, and the second does not arrive as a byproduct of the first, which is inconvenient, because the average is the part you can move with process, training and money.
A peak needs a decision instead: which single moment of this stay gets to be louder than the rest.
The slot exists whether or not anybody fills it.
If you don’t design the peak, something else takes it
The moment that ends up in memory is the most emotionally intense one available, and intensity is easier to reach going down than going up.
The 2025 tracking study measured the lows explicitly and concluded that extreme negative experiences deserve more consideration than positive peaks, which lines up with the wider work on negativity bias.
What counts in that moment is intensity, not how long it lasted and not how many times it happened, which is duration neglect showing up again on the other side of the scale.
That has a consequence I find genuinely counterintuitive: a long problem that somebody keeps attending to can weigh less than a short problem that nobody shows up for.
A three hour outage where a member of staff comes back to the guest every twenty minutes keeps the emotional low shallow, while a thirty minute outage where nobody appears at all digs it deep.
That second part is an inference from the mechanism rather than something measured in a hotel, so take it for what it is, but it cuts against the instinct to measure how fast a problem got closed, since speed is exactly the variable memory throws away.
It also means being strict about the difference between a moment that is merely mediocre and a moment that is extreme, because only one of the two survives the week.
A buffet where the fruit salad comes out of a tin and the pastries are the same three every morning drags your average down without ever producing an extreme, and no guest is going to build a review around it.
A fifteen minute queue at 8:30, in a room with four uncleared tables in plain sight, hit by a guest with a train at 9:12, is an extreme, and it will show up in the review with the time of the train written into it.
If the peak is worth designing, the next question is where to put it.
Where a designed peak works best
Dixon, Victorino, Kwortnik and Verma ran two experiments on the design of service sequences, published in Production and Operations Management in 2017, and it is the most directly usable piece of work in this whole area, because it does not stop at “look after the ending”.
A strong peak placed at the end had a positive effect on the overall evaluation.
A peak that arrived as a surprise produced more overall utility than one participants had been told about in advance, and the strongest combination of the two was a surprise peak sitting at the end.
The surprise effect was still measurable a week later.
Which makes the sequence of a stay something you design rather than something you inherit from the property management system, and the order of the pieces changes what the guest walks out with.
That second finding has a cost attached, though, because the same gesture at the same price is worth less to the guest once it has been announced.
A bottle waiting in the room because the confirmation email promised it is an expectation being met, and the guest ticks it off the list they were already holding.
The same bottle, at the same cost to you, appearing on the last morning without anybody having mentioned it, is a peak.
Your accounting cannot tell those two apart: one line, one supplier, one margin hit, and only one of the two buys something the guest repeats to somebody else.
This is the part where the small unadvertised things do the heavy lifting, and I have written before about the kind of gestures that cost close to nothing and get talked about anyway.
Which leaves the other half of the instruction, the ending, and in a hotel the ending is less obvious than it looks.
The ending carries more than it should
The strongest evidence on endings comes from a randomized trial on 682 patients undergoing an unpleasant medical procedure.
Half of them received the standard version.
For the other half the procedure was deliberately extended, with a final stretch that was still unpleasant but noticeably less so than what had come before.
The extended group went through more total discomfort, rated the whole experience as significantly less unpleasant, and were more likely to come back when they were called in again.
A smaller version of the same thing had appeared in 1993, when participants had one hand in water at 14 degrees for 60 seconds, and the other hand in water at 14 degrees for 60 seconds plus another 30 while the temperature crept up to 15.
They preferred to repeat the longer trial, the one that contained more discomfort in total, because it finished slightly better than it had been going.
Applying that to a stay means asking what the ending of a stay actually is, and the answer written in the schedule is the wrong one.
Checkout is an administrative procedure: a folio, a signature, a card, a key handed back.
The last moment of service a guest genuinely receives, in most properties, is breakfast on the morning of departure.
Picture that guest coming down at 9:40 on the day they leave.
The eggs have a crust on them and the tray underneath is nearly empty.
Two people are stacking plates at the next table, and one of them asks, politely enough, whether they can take the spare chair.
The coffee in the jug has been sitting there since a quarter past eight.
That is an ending, it is the last thing the property does to that guest, and it lasted ten minutes out of a stay of five days.
That breakfast occupies the end position is reasoning rather than a measured finding, since nobody has tested which episode sits in the end slot of a guest’s memory of a hotel, but the mechanism underneath it is not in dispute.
Breakfast also happens to be the thing guests write about most, which deserves a caveat of its own before anyone builds a strategy on it.
What reviews say about breakfast, and what the form says
An analysis of online reviews found breakfast mentioned in more than half of all comments, with two thirds of those mentions positive.
The share moved with category, 63% in four star properties against 50% in five star ones, and women mentioned it, and praised it, more often than men.
I would treat all of that as indicative and nothing more, because the version I could find is a PDF whose peer reviewed publication I have not managed to verify, and the composition of the sample is not clear.
There is also a mechanical reason behind part of that frequency, and it has nothing to do with memory.
On Booking.com the guest review score is built from six categories (staff, facilities, cleanliness, comfort, location, value for money), and breakfast is not one of them: it sits outside the average, as an optional secondary score the guest can fill in as well.
Some of how often breakfast comes up in reviews is produced by the form asking about it, so using that frequency as proof of a psychological effect means counting the questionnaire as if it were the guest.
Worth knowing for a second reason too, since the visible score is the one that feeds ranking, and the relationship between what guests write and where you end up in a list is looser than it sounds, which is roughly why a lower rated property can still outrank you.
Platform artefacts aside, the position stands, and it means the departure morning is doing two jobs at once.
The one window where the two positions overlap
If breakfast is the last service moment, and a peak works best when it sits at the end, then the morning of departure is the only point in the whole stay where the end position and the best position for a peak are the same position.
Everything below is that mechanism translated into a shift, not a set of measured results, and as far as I could find nobody has run the experiment of building a deliberate peak into departure morning and then reading what happens to the reviews.
The main argument for trying any of it is that it costs almost nothing.
- The last thirty minutes of service. Guests who come down late find a picked over buffet, coffee that has been sitting in the jug for forty minutes, tables nobody has cleared. The quality of your breakfast gets judged on that final half hour rather than on the average of the morning, because that is where an avoidable extreme lives.
- The departures list. The PMS knows exactly who is leaving today, and that list rarely reaches the people working the room. With it in hand the service stops being uniform: you know who has a train at 9:40 and who is trying to feed two children before a flight.
- The queue. In a lot of properties the worst moment of breakfast falls somewhere between 8:15 and 8:45. When the person hitting that queue is leaving that morning, the wait lands in the position that weighs most, and suggesting a slightly earlier or later time to the day’s departures is free.
- The guests who never get a final service moment at all. Anyone leaving before breakfast opens walks out with the end position empty, or filled by paperwork. Coffee ready to take away, or something left in the room the night before, is the whole fix.
- The order of the last interaction. Settle the bill, then say goodbye, in that order. Closing on the transaction puts the transaction in the end position.
- The folio. An unexpected line on the bill is an emotional low placed at the far edge of the stay, which is the worst possible place to put one. Pricing decisions carry a memory cost as well as a revenue.
None of this appears on a rate sheet, and all of it sits upstream of what any direct booking strategy is trying to achieve, since a guest who remembers the stay accurately and well is the only acquisition channel a hotel owns outright.
There is one more variable worth borrowing, from a 2024 field study I will come back to in a second: retrospective delay, the time between the experience and the evaluation, which moderates how much individual moments weigh.
Meaning the moment you ask matters as much as what happened, and that decision lives inside your post stay email flow rather than in the operation.
Where the rule stops working
The studies that map the limits of the peak and end rule are the ones I found referenced least, so they are worth putting on the table before anyone treats this as a law.
Start with Kemp 2008, the holiday study this piece opened with, because the authors’ own conclusion was not a clean endorsement.
Peak and end did not turn out to be a particularly good predictor of remembered holiday happiness, several summary measures predicted it reasonably well, and the researchers concluded that a lot of reconstruction was going on.
Strijbosch and colleagues, writing in Frontiers in Psychology in 2019, tested a complex and heterogeneous experience and measured recall immediately and a week later, finding peak and end inferior to average valence and average arousal at predicting what people remembered.
Three experiments with 988 participants aged from 2 to 97 failed to observe the rule at all on simple positive experiences.
Then there is McCullough and colleagues, 2024, in the Journal of Business Research, a field study of more than 5,000 hotel stays collected over five years, looking at guest interactions with front line staff.
First impressions, peak impressions and last impressions all had a significant impact on overall satisfaction, and the first impression came out as the largest single contributor, on the biggest hotel sample anyone has run on this question.
Put together, the direction is fairly clear: the more an experience resembles an actual stay, long and mixed and made of dozens of separate episodes, the more the average climbs back into the equation.
Which is probably the most useful thing here, because it means the departure window is worth working on precisely when the rest of the stay is already fine.
The question this leaves
In most properties you can name the person responsible for rate strategy, and the person who writes the review replies.
The last thirty minutes of breakfast on a departure day rarely has anybody’s name attached to it, and it is one of the two positions the memory of the whole stay gets built from.
If you have tried something deliberate in that window, a coffee for the early leavers, a different order at the desk, a staggered time for the day’s departures, I would like to know what actually happened, because as far as I can tell nobody has measured this properly yet.
