Recognizing a Returning Guest Across Properties, Channels, and Name Variants
Two years ago she booked direct. This time it was an OTA, a different email, a slightly different spelling of her name. Every personalization promise a hotel makes rests on knowing she's the same person.
A woman walks up to your front desk. Your system sees a first-time guest. Your reservations team sees a booking made through an OTA with an email address that ends in a relay domain. Your spa system holds a profile from last spring under "Katherine," your loyalty platform holds one under "Kate," and your sister property across town has a third under her married name. Nobody in the building knows she has stayed with you four times. She is a returning guest in every way that matters and a stranger in every way your technology can see.
That gap is where a lot of hotel personalization quietly dies. Recommendation engines and generative messaging get the attention; the layer beneath them gets very little. A recommendation built on a fragment of someone's history is a confident guess about a guest you've half forgotten. What follows covers why recognition fails at independents and small groups, how big the gap is, and a working method: matching logic, OTA-masked emails, confidence thresholds, false merges, and the consent line when a profile crosses properties.
The Problem: One Guest, Five Records
Guest identity fragments for structural reasons, not because of careless staff. A single guest can create records in the booking engine, the property management system, the spa and restaurant systems, the Wi-Fi portal, the loyalty platform, and the email service provider, and each system captures a different subset of fields with a different level of care. Platform vendors describe the same pattern: TrustYou's documentation on guest profile unification lists booking engines, property management systems, CRMs, Wi-Fi sign-ups, restaurant systems, and loyalty programs as the sources that have to be reconciled into a single golden profile.
The second driver is human. A Hotel Online guide to clean guest data describes the classic failure: a guest books online, the front desk cannot find the reservation profile because of a misspelling or an incomplete field, and a second profile is created at arrival. The same piece says many hoteliers find 10 to 50 percent of their database incomplete before a cleanse, and it lists the fallout plainly: emails sent to inboxes that don't exist, repeat guests who go unappreciated, targeting that misses. Each one is a recognition failure.
The third driver is distribution. When a guest books through an online travel agency, the property usually receives a relay email address and often a masked phone number rather than the guest's own contact details. Mews explains the mechanism directly: OTAs generate temporary emails so that guests and hoteliers can communicate, but only temporarily, and once the temporary address expires there is no other way to reach the guest by email. The most reliable key a hotel normally has for matching a person, a stable email address, is exactly the field the OTA channel withholds.
A guest who has stayed with you four times and arrives through an OTA is a first-time guest to your systems. The loyalty you earned is sitting in a record you cannot find.
The commercial stakes follow from what guests expect. McKinsey research summarized on its Brazilian site reports that 71 percent of consumers expect personalized interactions and 76 percent are disappointed when they do not get them. In hospitality specifically, Medallia's research, which surveyed 1,749 hotel guests and 1,905 retail consumers in November 2023, found that just 23 percent of consumers reported high levels of personalization after a recent hotel stay. Earlier Medallia research, cited in the same release, put the share of consumers willing to pay more for a customized experience at 61 percent. A property can't close that gap with better offers if it cannot first establish who is standing in front of it. Our earlier work on loyalty personalization assumes recognition is solved. This article is about solving it.
The Data: How Much of Your Database You Cannot See
Start with scale. Revinate's 2025 benchmark data shows that, on average, 21 percent of database records around the world contained a masked email. That is one record in five that cannot be reached by email and cannot be matched on email. The exposure tracks channel mix. Cloudbeds' 2026 report, compiled from 90 million bookings across 180 countries, found that OTAs accounted for 63.4 percent of independent hotel bookings in 2025, nearly 80 percent in some markets, and that OTA cancellation rates reached 21.8 percent against 10.6 percent for direct bookings. A property in a market near the top of that range is not looking at a 21 percent problem. It is looking at a larger one.
The distribution picture has also proved stubborn. Skift's May 2026 analysis concludes that over the past ten years hotels narrowed OTA commissions and built loyalty programs, but OTAs kept roughly the same share of room nights. Meanwhile, a first-half 2026 industry review on Hospitality Net puts OTA commissions at 15 to 25 percent against direct acquisition costs of roughly 5 to 12 percent, and reports that 18 percent of travelers who begin a search on an OTA ultimately book directly with the hotel. That last number matters for this topic. Nearly one in five OTA-originated searchers becomes a direct guest eventually, which means the person who arrives through a relay email today may well be the same person who books direct next year. Only a system that can link the two will notice.
The window for capturing a real contact is short and varies by channel. The table below sets out what the major OTAs make available and for how long.
| Data element | Channel | How long it lasts | Implication |
|---|---|---|---|
| Masked guest email | Booking.com | 60 days after departure, per SiteMinder | Capture a real email during the stay or lose the contact route |
| Messaging thread | Booking.com | Until 7 days after checkout; 14 days to reply if the guest writes first | Post-stay outreach through the platform is a one-week opportunity |
| Message thread visibility | Expedia | 45 days after checkout | Any identity clue in a thread must be written to the profile before it expires |
| Phone number | Expedia | Delivered masked | Phone cannot be used as a match key unless the guest supplies a real number |
| Guest card details | Booking.com | Viewable up to 3 times, for at most 10 days after checkout | Do not rely on payment data as a long-term identity anchor |
Capture is possible, and the reported rates are encouraging when the process is built into the stay. Mews describes a guest portal flow that works for guests who booked through Booking.com, Airbnb, and Expedia, which it says equates to about 96 percent of masked emails across its properties, while noting that Agoda strips out the portal link so those guests cannot supply a real address that way. Revinate's chief marketing officer, citing the company's database health data, reports properties converting roughly 12 percent of masked profiles into verified, contactable leads. That's one vendor's number. Still, masked profiles aren't a lost cause, and a property that does nothing converts none of them.
Deduplication has measurable payoff as well. An Infarsight case study of a global hotel chain reports a 35 percent reduction in duplicate profiles after the chain standardized records and established a central golden record across its property management, CRM, and loyalty systems, with better personalization and improved campaign conversion as the stated outcomes. The case study does not disclose its matching method, so read it as evidence that the work pays. It isn't a recipe.
| Benchmark | Figure | Who reported it | Read it as |
|---|---|---|---|
| Database records with a masked email | 21% global average | Revinate | Your starting exposure before channel mix adjustment |
| Independent hotel bookings via OTA | 63.4% in 2025 | Cloudbeds | Why masked records keep arriving |
| Masked profiles converted to contactable leads | About 12% | Revinate | One vendor's reported result, not a benchmark |
| Reduction in duplicate profiles after golden record build | 35% | Infarsight case study | Typical scale of cleanup at a multi-system chain |
| Database incomplete before cleansing | 10% to 50% | Hotel Online | The size of the mess before any cleanup starts |
The Framework: Normalize, Match, Score, Then Decide
Identity resolution is four steps. First, normalize the fields so equal things look equal. Second, apply deterministic rules for the cases where a hard key exists. Third, apply weighted probabilistic scoring for everything else. Fourth, and this is the step that gets skipped, turn the score into an action sized to the risk of being wrong.
Step one: normalize before you match
A lot of failed matches are formatting problems. Lowercase and trim email addresses and correct common domain typos such as a missing letter in a major mail provider. Strip punctuation and country codes from phone numbers before comparing them. Transliterate names so that accented and non-Latin spellings meet their Latin equivalents, and add a phonetic fallback so that "Catherine," "Katherine," and "Kathryn" are candidates for one another. TrustYou's documentation describes this approach: names lowercased, transliterated and given a phonetic fallback; emails lowercased, typo-corrected, and stripped of temporary OTA aliases. Filtering out relay domains is the single most valuable normalization rule for an OTA-heavy property, because a masked address that sits in the email field will otherwise produce false matches between unrelated guests or, worse, a false sense of a verified key.
Step two: deterministic keys for the easy cases
When a verified, stable key exists, use it and move on. A loyalty number, a verified personal email confirmed through a guest portal or a double opt-in, and a government-verified identifier collected lawfully at check-in are all keys that can anchor a merge. Be strict about what counts as verified. An email captured at the front desk and typed by an associate is better than a relay address and worse than one the guest confirmed herself. Record how each key was obtained so the confidence of a match can reflect it later.
Step three: weighted scoring for everything else
Where no hard key exists, score the evidence. A published example is useful here because it shows how to build in restraint. TrustYou's documentation assigns each attribute a weighted contribution: name 30, phonetic name 15, email 20, phone 25, birthday 15, and country 4, with phone compared on its last five digits and requiring an exact match, and names needing a similarity of at least 0.80. The design choice that deserves copying is the explicit statement that a shared email and phone, worth 20 plus 25 for a total of 45, does not by itself merge family members. Households share contact details constantly, and a model that treats a shared phone as proof of identity will fuse a husband and wife, a parent and a grown child, or an assistant and an executive into one record.
| Attribute | Weight | Matching rule | Practical note for a hotel |
|---|---|---|---|
| Name | 30 | Normalized, similarity of at least 0.80 | Handles spelling variants and married names only partly, so pair it with other evidence |
| Phone | 25 | Exact match on the compared digits (last five) | Worthless when the OTA masks it, so capture a real number at pre-arrival |
| 20 | Normalized, OTA aliases filtered out | Count verified personal addresses only | |
| Phonetic name | 15 | Fallback when spelling differs | Useful for transliterated and dictated names |
| Birthday | 15 | Exact | Collect only where lawful and useful, and store minimally |
| Country | 4 | Exact | A tiebreaker, never a driver |
| Shared email and phone only | 45 combined | Does not trigger a merge | Link as a household, do not fuse |
Step four: confidence tiers that match the cost of error
A score is only useful once it is attached to an action. The two errors have very different costs. A missed match leaves a returning guest unrecognized, which costs a little warmth and a little revenue. A false merge fuses two people, which can expose one guest's stay history, preferences, or contact details to another, and that is a privacy incident as well as a service failure. The framework below treats the two errors asymmetrically: be generous about suggesting and cautious about merging.
| Tier | Evidence | System action | Risk if wrong |
|---|---|---|---|
| 1. Verified key | Loyalty ID or guest-confirmed email, plus name similarity of at least 0.80 | Auto-merge and log the key source | Low |
| 2. Strong composite | Exact phone, name match, and one more attribute such as birthday or a prior stay overlap | Auto-link for recognition; merge after one more confirming signal | Medium |
| 3. Probable | Name and country only, or phonetic name with one weak attribute | Show the front desk a "may have stayed before" prompt; do not merge | Medium to high if acted on blindly |
| 4. Shared contact | Same phone or email, different names | Link as household; never merge | High, because it exposes one person's data to another |
| 5. No signal | Masked email, masked phone, common name | Create a new profile and run capture workflow | Low, and it is recoverable later |
The middle tiers carry the discipline. For a probable match, the correct move is to surface a prompt to a human, not to act. A front desk associate who sees "possible returning guest, last stay in March, confirm" and simply asks the question has resolved the ambiguity at almost no cost, and the guest has been made to feel known without a single assumption being baked into the record. Our work on disclosure and guest trust argues the same principle from the other side: guests accept personalization they can see and understand, and they punish personalization that feels like surveillance. Vibhu Gaind, CIO of RBH Hospitality Management, put the line plainly in a Hospitality Investor feature on whether personalization has gone too far: "Knowing a guest's birthday might be okay but anything more than that may be too much information." Recognition that is confirmed with a question stays on the right side of that line.
Be generous in suggesting a match and conservative in making one. A missed match costs you a little warmth. A false merge costs you a guest's trust.
Handling the OTA-masked record specifically
Treat every masked record as a capture opportunity with a deadline, not as a dead profile. The deadline is the shortest applicable window in the first table above. The workflow has four parts. At booking, flag the record as masked so downstream systems do not treat the relay address as a verified key. Between booking and arrival, send a pre-arrival message through the permitted channel inviting the guest to confirm arrival details and supply a personal email and phone, which also serves the broader goals set out in our piece on the AI-orchestrated pre-arrival experience. At check-in, ask for the real contact once, with a reason the guest values, such as sending a digital folio or arrival instructions. After departure, if the real email was captured, promote the profile and run the match against history. The Revinate-reported three-touch sequence at 48 hours, 7 to 10 days, and 30 to 45 days after the stay gives a reasonable cadence for post-stay conversion once a real address exists.
Cross-Property Recognition and the Consent Boundary
Recognizing a guest across a group of properties is where recognition shifts from an operations question to a legal one. A profile that lives inside one property's system is processed under that property's notice. A profile that is merged across properties may involve different legal entities, different owners, and different privacy notices, and the guest may have agreed to none of that. Decide the boundary before you build the merge.
Three principles apply under the European GDPR, and comparable logic runs through most modern privacy regimes. Accuracy: Article 5(1)(d) requires personal data to be accurate and kept up to date, with inaccurate data erased or rectified without delay, and a false merge is by definition inaccurate data about two people. Storage limitation: Article 5(1)(e) requires that data be kept in an identifiable form for no longer than necessary, which constrains how long a dormant profile can be held to enable future recognition. Penalties: infringements of the basic principles in Articles 5, 6, 7, and 9 sit in the top fine tier, which Article 83(5) sets at up to 20 million euros or 4 percent of worldwide annual turnover, whichever is higher. Enforcement in the sector is real. The CMS GDPR Enforcement Tracker Report for accommodation and hospitality counts 91 fines from 15 data protection authorities totaling about 22.7 million euros, most of it one Marriott fine; the typical fine is far smaller. It also flags a newer category: Spanish and Czech authorities sanctioning hotels that collect identity documents to verify guests, including a 42,000 euro Spanish fine in 2025.
| Requirement | What the source says | Design consequence for identity resolution |
|---|---|---|
| Accuracy, GDPR Art. 5(1)(d) | Data accurate and up to date; inaccuracies rectified without delay | Build an unmerge function and log every merge so errors can be reversed |
| Storage limitation, Art. 5(1)(e) | Identifiable data kept no longer than necessary | Set retention clocks per profile; one guide's example uses 36 months for active guests and deletion at 60 months for lapsed ones |
| Top fine tier, Art. 83(5) | Up to 20 million euros or 4% of worldwide turnover | Treat cross-entity merging as a board-level risk, not a database setting |
| Data subject requests | Generally one month under GDPR; 45 days under CCPA, extendable | A merged profile must be searchable and erasable as one unit |
| Opt-out signals, CCPA | Honor Global Privacy Control for California residents | Consent flags must travel with the merged record, not stay behind in the old ones |
The practical rule for groups is to separate recognition from data sharing. A property can often use cross-property history to recognize a guest, for example by flagging "returning guest to the group" to the front desk, without copying folio details, preferences, or contact data into the receiving property's record. Where the guest has been told, in plain language and at the point of collection, that preferences travel across the brand, you can share more. Where the notice is silent, share the minimum. The guest data consent and preference center we describe elsewhere is the control surface that makes this workable, because a merge that respects each flag is only possible if the flags exist and travel with the profile. The SendSquared guide to hotel guest data adds the operational detail that retention should be enforced by automatic deletion at policy boundaries, not merely promised in a notice.
Implementation: A 90-Day Path for an Independent or Small Group
You don't need an enterprise customer data platform to start. An audit, a ruleset and a review queue will do.
In the first thirty days, measure. Export the guest table from the property management system and count three things: the share of records with a relay-domain email, the share with a missing or masked phone, and the apparent duplicate rate on a normalized name plus country pairing. Pull a random sample of fifty suspected duplicate pairs and have a front desk lead classify each as the same person, a household, or a different person. That sample is your calibration set, and it is the only honest basis for setting your own tier thresholds.
In days thirty to sixty, fix capture and normalization. Add the masked-record flag, apply the normalization rules above, and add a single pre-arrival request for a personal email and phone. Make sure the front desk script asks for the real contact once, with a reason. None of it costs much, and the masked population starts shrinking right away.
In days sixty to ninety, turn on tiered matching with a human review queue. Auto-merge only Tier 1. Send Tiers 2 and 3 to a queue that one person clears daily, and record every decision so the thresholds can be tightened against evidence. Add the unmerge function before you add scale. Hotels beginning this work often benefit from a structured design of the guest data model and the recognition workflow before any platform is selected, and our AI-Powered Guest Experience Systems service is built for exactly that scoping step.
Measuring It: Metrics That Show Recognition Is Working
Track recognition as an operating metric. The core set is small. Masked-profile share, tracked monthly against your 21 percent reference point and your own baseline. Capture rate, meaning the share of masked records that gain a verified personal email within the platform's contact window. Duplicate rate on a fixed sample, so that you can see whether cleanup is holding. Recognition rate at arrival, meaning the share of returning guests flagged as returning before check-in completes. False merge rate, measured through complaints and audit samples, with a target as close to zero as you can hold. Finally, the revenue marker that justifies the effort: repeat-stay conversion among recognized guests compared with unrecognized ones. That comparison is where the return shows up, or doesn't.
What Owners and GMs Get Wrong
The first mistake is buying a personalization tool before fixing identity. The tool will run, the dashboards will fill, and the offers will be addressed to half-remembered guests. The second is treating a shared phone number or email as proof of one person, which fuses households and creates privacy incidents. The third is merging aggressively to make a duplicate count look better, trading a visible metric for an invisible risk. The fourth is ignoring the clock on OTA data: a property that waits until it wants to market to a masked guest has usually waited past the window. The fifth is assuming the group can share everything across properties because the logo is the same. Different legal entities and different notices make that assumption expensive. None of these needs new technology to fix. They need rules, a review habit, and a decision about where the consent boundary sits.
Frequently Asked Questions
Why can't I just match guests on email address?
Because the field is unreliable in exactly the channels that matter most. Revinate reports that 21 percent of database records worldwide contain a masked email, and OTA relay addresses expire. SiteMinder puts the usable life of a Booking.com masked email at 60 days after departure, while Booking.com's messaging closes seven days after checkout. A match on a relay address identifies a booking, not a person. Use email as a key only when it is a verified personal address, and filter out relay domains before any comparison runs.
How do I avoid merging two different guests by mistake?
Never merge on a single weak signal, and never treat a shared phone or email as proof of identity, since households share contact details routinely. Reserve automatic merging for verified keys, send mid-confidence matches to a human review queue, and log every merge with an unmerge function so errors can be reversed. Measure false merges through audit samples and complaints, and tune thresholds against your own reviewed pairs rather than a vendor default.
Can I get real emails from OTA guests?
Often, yes, but only if you ask inside the contact window and give the guest a reason. A pre-arrival message asking for a personal email and phone, a guest portal step in online check-in, and a single well-scripted question at the front desk are the main routes. Mews reports that its portal approach covers Booking.com, Airbnb, and Expedia guests, about 96 percent of masked emails across its properties, while Agoda strips the portal link. Always check each OTA's current terms on how contact data may be used.
Is it legal to recognize a guest across properties in a group?
It depends on the entities involved, the notice the guest saw, and the jurisdiction, so confirm with counsel. As a design principle, separate recognition from data sharing: flagging a returning guest to the group is a lighter step than copying folio history and contact details into another property's record. Under GDPR, accuracy and storage limitation apply to every merged profile, and the top fine tier reaches 20 million euros or 4 percent of worldwide turnover. Make consent flags travel with the profile.
What is the first step for a hotel with a small team and no data platform?
Audit your own database and add the masked-record flag. Count the relay-domain emails, draw a sample of fifty suspected duplicate pairs, and have a front desk lead classify them. Then add one question to the pre-arrival message asking for a personal email and phone. Those steps need no new platform, they shrink the masked population from the first week, and they give you the calibration set you need before you turn on any automated matching.
Peter Mack is a hospitality technology strategist and founder of HospitalityOS, helping independent hotels and resorts implement AI systems that drive revenue and reduce operational costs. With 25 years in hospitality operations and technology, he has worked with properties of all types and in every region as both a General Manager, Founder, Operator, Asset Manager, and Owner.