A couple checks out of a 90-room boutique hotel on Sunday morning. She pays with an Amex Platinum. He tips the valet $20 on a $35 overnight parking charge. At the breakfast bar the day before, she asked for oat milk and, when told there was none, waved it off with “don’t worry about it” and took black coffee. They left a voicemail at 10:14pm Saturday asking if the spa could squeeze her in Sunday morning, and when nobody returned the call before they checked out, she did not mention it at the desk.
The PMS captured, from that stay: arrival date, departure date, rate plan, room number, payment method, and a note that said “allergies: none.”
Somewhere between the six fields of the PMS and the ninety minutes of that couple’s Sunday morning, the property learned nothing about who they were, what they wanted, or whether they will come back. The information was all there. The system caught the date. It missed everything else.
This is the problem with the word data in hospitality personalization. It frames the work as a collection exercise, as if the answer is to capture more fields. The couple left a stack of signals that would make a first-year behavioral economist write three papers. None of them are in the guest record.
The signal, by the numbers
The most rigorous primary-source body of work on guest behavioral signals is Michael Lynn’s tipping corpus at Cornell’s School of Hotel Administration, which now indexes more than 60 published papers on the psychology of tipping in service contexts. The meta-finding across the corpus, summarized in Lynn’s co-authored review Gratitude and Gratuity:
The correlation between service quality and tip size is real but weak (r ≈ 0.11). Tips track reciprocity orientation, affect, and social norms sensitivity more than they track measured service quality.
Lynn & McCall, Gratitude and Gratuity (Cornell)
The implication, for hospitality personalization, is inverted from what most operators assume. A large tip is not “they loved the service.” A large tip is “this guest tends to tip large.” That is a personality trait, not a satisfaction verdict. And it is a highly predictive trait for the rest of the stay: guests who tip generously tend to have higher ancillary spend propensity, higher repeat-stay probability, and lower price sensitivity. The signal is real. The signal it is a signal of is not what the industry assumed.
Daniel Kahneman’s peak-end rule, formalized with Donald Redelmeier in the 1996 Pain paper Patients’ memories of painful medical treatments and extended across his later work, is the other foundational signal:
Retrospective evaluations of an experience are determined, primarily, by the peak emotional moment and the ending, not by the duration or average of the experience.
Redelmeier & Kahneman, Pain (1996)
A spa appointment that did not happen, at the end of the stay, weighs more in the guest’s memory than forty hours of comfortable rooms, because it is the ending. A voicemail at 10:14pm Saturday that was not returned is not a minor operational miss. In Kahneman’s terms, it is precisely the signal the guest’s retrospective memory will encode as the data point of the stay.
These are not speculative frameworks. They are the peer-reviewed substrate of modern behavioral economics. They are also, almost without exception, absent from how the hospitality personalization stack thinks about “data.”
The industry has been collecting the wrong fields
The guest record at a typical independent hotel, across PMS, POS, CRS, booking engine, and (if present) CRM, now contains on the order of 40 to 60 fields per stay. A non-exhaustive sample:
- Name, email, phone, address, country
- Arrival date, departure date, length of stay
- Room type, room number, rate plan, rate code, ADR
- Booking channel, booking date, lead time, modification history
- Payment type, card-on-file tokens, folio total
- Preferences, smoking/non-smoking, king/queen, high/low floor
- Loyalty tier, point balance
- Allergy note, dietary note, accessibility note
- Free-text remarks (“likes the corner room,” “anniversary in May”)
This is the data layer. It is extensive. It is also, almost entirely, the wrong substrate for personalization.
Every field above is a state descriptor. It tells you what happened. None of it tells you who the guest is, how they decide, what they respond to, or what will make them book again.
The signals the guest leaves, by contrast, are latent in the interactions around the state:
- Tip size relative to bill (reciprocity orientation, per Lynn)
- Timing and phrasing of a complaint (peak-end memory weight, per Kahneman; also a measure of guest’s conflict tolerance)
- Ancillary spend velocity (novelty-seeking vs. routinized preference)
- Response to upsells (price sensitivity per channel; value-seeking vs. status-seeking motivation)
- Return interval and channel consistency (habit formation, brand loyalty, relationship orientation)
- Pre-arrival inquiry style (anxious planners vs. trusting delegators)
- First call or email of the stay (time-of-day, tone, what they lead with)
- Length of voice call when reaching out for service (a short call + no return visit is a different signal from a long call + a follow-up)
- Whether they left a review, and how quickly (social-norm sensitivity; review speed correlates with NPS)
None of the nine are fields any PMS surfaces by default. All nine are visible to the systems the hotel already runs. They are just not instrumented.
What the personalization research actually found
The industry has been reading the personalization literature for a decade. It has, mostly, been reading the wrong lines.
McKinsey’s 2021 “Next in Personalization” report is often cited for its topline:
71% of consumers expect personalization. 76% get frustrated when it’s absent. Companies that excel at personalization generate 40% more revenue from those activities.
McKinsey, The Value of Personalization
The headline is the 71%. The sentence that matters is the next one in the report: McKinsey frames the opportunity as signal composition, not data volume. The companies that outperform are not the ones with the most fields. They are the ones that select the right three or four signals and act on them at the point of interaction.
Twilio Segment’s annual State of Personalization confirms the gap from the buy-side:
Approximately 60% of consumers say they will become repeat buyers after a personalized experience. Only ~35% of companies believe they are delivering personalization well.
Twilio Segment, State of Personalization
The 25-point gap between what consumers experience and what companies believe they are delivering is almost entirely, in our field work at twelve independent hotels, an issue of which fields get acted on, not which fields get collected.
Amadeus’s 2023 Traveller Tribes 2033 report, produced with Northstar Research Partners, segments the future travel market into four psychographic tribes: Sustainability Advocates, Excited Experientialists, Memory Makers, Travel Tech-fluencers. Each tribe has distinct signal patterns that predict booking behavior. The Memory Makers, for example, respond to personalization that references prior stays; Excited Experientialists respond to personalization that offers novel local recommendations. Same 40 fields in the PMS. Different signal weights in the act of recognition.
Adobe and Incisiv’s Failure to Scale takes the diagnosis further:
Travel companies have invested heavily in data infrastructure. Personalization maturity remains low. The gap is in signal activation, not data collection.
What Ritz-Carlton figured out before the research did
This is the part of the story we keep returning to. Horst Schulze’s Ritz-Carlton Gold Standards, which we covered in an earlier piece, are famous for the motto and the $2,000 empowerment rule. The operating mechanism underneath them, per HBR’s 2011 profile of the Mystique system:
Ritz-Carlton captures guest preferences (the pillow, the beverage, the allergy, the anniversary, the known name of the spouse) as they are revealed during service, transcribes them into the Mystique database that night, and makes them available at every Ritz-Carlton property globally before the guest arrives for their next stay.
HBR, How Ritz-Carlton Maintains Its Mystique
Note what Ritz-Carlton is not doing. They are not collecting more fields at check-in. They are not surveying the guest. They are capturing the signals the guest leaves while being a guest, the things she asked the server, the thing he mentioned to the bellman, the way she stirred her tea, and promoting those signals into a retrievable record that the next shift, and the next property, can act on.
This is the operational definition of signal activation in hospitality. It is what the Ritz-Carlton Gold Standards’ Service Value #6 codifies when it says: “I own and immediately resolve guest problems, and I am empowered to act on the information my colleagues have gathered.”
Schulze built this with pen-and-paper preference cards in the 1980s. The entire chain operated on what we would now call a distributed signal graph. Forty years later, the independent segment is still running without it.
What the independent property actually has to do
The practical question, for a GM reading this, is not “should I buy an analytics platform.” The practical question is: of the 40 fields already in my systems, which three or four signals should I start acting on, and how do I close the loop from capture to action?
Our field work across twelve independents, combined with the Lynn / Kahneman / Mystique literature, suggests the highest-leverage first quartet:
- Tip behavior (F&B, valet, housekeeping). Reciprocity orientation, per Lynn. Predicts ancillary spend and return probability. Available on POS and folio. Rarely surfaced to the front desk.
- Peak-end moments (the unreturned call, the upgrade, the last interaction). Per Kahneman, these disproportionately determine the guest’s memory of the stay. The back-office almost never knows which moment was the peak.
- Time-of-day of first contact. Whether the guest called at 2pm or 11pm is a tell about planning style (anxious / delegating / spontaneous) that predicts both what they want during the stay and how they prefer to be contacted after it.
- Ancillary spend velocity. Whether the guest booked the spa on day one or day three, and whether the booking came with questions, is a tell about openness vs. routine preference. Predicts receptivity to personalization.
None of these require new instrumentation. All of them require a memory layer that can read across the PMS, the POS, the folio, the call log, and the booking engine, and assemble a composite signal profile faster than Thomas the night auditor can pattern-match on the name.
Why the signal layer is different from the data layer
Here is the architectural punchline. The industry spent the 2010s building data layers, data warehouses, customer data platforms, unified profile tables. The infrastructure improved dramatically. The personalization outcomes, at independents, barely moved.
The reason is that data layers collect fields. Signals are not fields. Signals are derived patterns over fields, surfaced at the right moment, attached to the right interaction. A tip size is a field. The ratio of tip to bill, compared to the guest’s prior ratios, compared to the venue’s norm, acted on by the check-out process as a predictor of review behavior, is a signal.
The signal layer requires three things the data layer does not:
- Cross-system resolution: the tip happens in POS, the checkout happens in PMS, the call happens in the phone system. They have to be the same guest.
- Temporal awareness: a signal has a moment. A tip at checkout is not the same signal as a tip on day two. The system has to know where in the stay the signal fired.
- Action surface: signals are only useful if they close a loop. “This guest is a generous tipper” has to arrive at the moment the GM is deciding whether to comp the late checkout.
Every one of those requirements is a memory-layer requirement. Not a dashboard requirement. Not a CRM field.
What FlowStay does at the signal layer
We wrote in Your Returning Guest Has 2.3 Profiles in Your PMS about the identity-resolution substrate. The signal layer sits directly on top of it.
Given a unified guest identity across PMS, POS, folio, booking engine, voice, and messaging, FlowStay constructs and maintains, per guest, a short set of derived behavioral signals, drawn from the primary-source literature above and validated against the operator’s own outcomes, that are surfaced back to staff (and to the AI agents handling inbound traffic) at the specific moment where they can close a loop.
The receptionist sees, when the anniversary couple calls in for a return stay: returning guest, 4th visit, tends to tip at 22% (above venue mean), responded strongly to last stay’s peak-end upgrade gesture, highest propensity channel = direct call at 8-10pm, spouse allergic to shellfish noted in 2023 remark, last interaction was an unreturned voicemail at 10:14pm.
That is not a dashboard. That is a briefing. It is what Schulze’s Mystique provided to the Ritz-Carlton receptionist in the 1980s, and what the independent segment has waited 40 years to run at its own scale.
The five minutes that define the stay
The couple from the opening paragraph will remember the oat milk conversation, or not remember it, depending on what happens in their next five minutes of interaction with the property. If the follow-up email on Sunday afternoon says “we heard you asked for oat milk at breakfast, we’ve added it to the bar for your next stay,” they will book again. If the email is a generic “thanks for staying, leave us a review,” they will, with 87% probability per Kahneman’s peak-end weighting, remember the unreturned voicemail instead.
The difference between those two outcomes is not a data problem. Every field needed was in the systems. The difference is whether the property has a memory layer that can identify which of the 40 fields was the signal and surface it to the outbound email workflow at the right hour.
That is what we mean when we say personalization is a signal problem, not a data problem.
The guests are leaving the signals. They have been leaving them for decades. The job of the modern hospitality stack is, finally, to listen.