Lifetime value for guests who vanish for years

A couple books a boutique hotel on the Amalfi Coast for their honeymoon in 2022. Then nothing for three and a half years: no email opens, no app logins, no site sessions. In late 2025 they book again, a different property in the same small collection, longer stay, this time with a toddler. Any subscription-style model would have marked them churned in month three. They were worth two four-figure bookings plus a referral to two other couples.

The awkward part is that the model was not wrong about the silence. It was wrong about what silence means. Lifetime valueLifetime valueLifetime Value: the total revenue (or profit) a customer generates throughout their entire relationship with your business.View full definition → (LTV, sometimes CLV) formulas were built for businesses with weekly or monthly purchase cycles, not for businesses where the most valuable guests disappear for 18 to 48 months and then return at full price.

Why the SaaS-style formula breaks

The textbook version:

LTV = Average Order Value × Purchase Frequency × Customer Lifespan

For monthly billing, frequency is stable and lifespan is observable. For a leisure hotel group, frequency might be "one stay every 2.3 years" and lifespan is a guess. A guest who has not booked in 14 months is not gone; a guest who has not booked in 14 months and whose youngest child just turned 14 probably is.

Applied naively, the formula fails in two directions:

• Overstating value: calculate frequency from your most recently booked cohort and you extrapolate a rhythm most guests never sustain.
• Understating value: apply a 12-month active window borrowed from retail churn conventions and you declare guests dead who are mid-cycle, then cut spend on them in the year before they were going to return.

There is a third, quieter failure: right-censoring. If your booking data only goes back three years, you cannot observe four-year returners, so your measured return probability is biased downward by construction. Survival methods handle censored observations explicitly; a simple "% who came back" query does not.

A model that fits infrequent, high-ticket travel

Step 1: Segment by trip type, not by recency. Honeymooners (very low frequency, high spend, high referral value), repeat family leisure (medium frequency, strongly seasonal), corporate bleisure (higher frequency, lower spend per stay). Averaging them destroys the signal in every direction at once.

Step 2: Extend the observation window to match the category. Instead of a 12-month churn clock, use survival analysis, a statistical method borrowed from medical research that estimates the probability a guest returns *at all* over a multi-year window, rather than assuming a fixed lifespan. Three to five years before classifying a guest as dormant is a defensible working range for leisure hospitality, because reactivation past year one is common rather than exceptional (background on guest retention curves via Cornell's Center for Hospitality Research).

Step 3: Use a probabilistic repeat estimate, not a flat frequency. Ask what share of this segment ever returns, and what a returning guest is worth, discounted for time.

**Step 4: Model *when*, not only *whether*.** Return hazard in travel is lumpy, not smooth. Center Parcs demand sits on the school holiday calendar: February half-term, Easter, late summer, October. A family that skipped one half-term has told you almost nothing. A family that skipped two consecutive equivalent weeks has told you a great deal. For that kind of guest, dormancy is better counted in seasons missed than in months elapsed.

Simplified formula for high-ticket, infrequent travel LTV:

LTV = (P_return × Avg_Revenue_per_Return_Visit × Expected_Visits_over_Horizon)
      + Referral_Value
      - Acquisition_and_Servicing_Cost

Where P_return is the observed probability a first-time guest books again within your chosen horizon (say five years), not within 12 months.

Worked example

A boutique chain with these inputs (illustrative, patterned on small luxury groups, not a specific company's disclosed data):

• Average first booking value: $1,800 (3 nights)
• Probability of a second stay within 5 years, from cohort data: 35%
• Average value of second stay (longer trip or higher room category): $2,400
• Probability of a third stay given a second occurred: 45%, average value $2,600
• Referral value: a returning guest historically brings ~0.4 new bookings by word of mouth, each worth ~$1,800, so $720

Second stay contribution: 0.35 × $2,400 = $840
Third stay contribution:  0.35 × 0.45 × $2,600 ≈ $409
Referral contribution:    $720

5-year LTV beyond first booking ≈ $1,969. Total ≈ $3,769.

Set that against the fully costed acquisition figure the true-CACCACCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition → lesson builds, commissions and cancellations included. At $300 to $600 per new guest, plausible for boutique performance marketing in the US and Europe in 2025 to 2026, the ratio lands somewhere between 6:1 and 12:1, well clear of the 3:1 rule of thumb used across marketing.

A naive 12-month model counts almost none of that, because almost none of it arrives inside year one. It shows a thin or negative return and pushes the team to cut spend on the highest-value guests in the file.

What this changes about retention marketing

Don't judge win-back on a 90-day clock. A "we miss you" email sent eight months after checkout can fail while the same guest books direct, unprompted, in month 30. AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → windows in Google Ads and Meta default to 30 or 90 days, so they will structurally undercount travel retention ROIROIReturn on Investment: the ratio of net profit to the cost of an investment. A 300% ROI means each dollar invested returns $3.View full definition →. Treat that as a measurement gap to work around, not a reason to stop the programme.

Time the contact to the return season, not to the silence. TUI's UK summer bookings cluster heavily in the turn-of-year window, when families commit to next August. A win-back sent in April is talking to someone who has already booked or already decided not to. Same budget, same creative, half the yield, purely because of the month on the send calendar.

Loyalty mechanics have to survive the gap. Marriott Bonvoy keeps a relationship alive across multi-year silences because status and points accrue outside the hotel stay, through co-brand card spend and partners, with well over 100 million members enrolled. The tier and points mechanics themselves belong to the loyalty lesson; what matters here is that they turn an invisible dormant guest into an observable, still-earning one. A small group without that machinery can approximate it with a returning-guest rate lock, a priority booking window before public release, or a perk attached specifically to the second stay, which is the highest-leverage transition in the model above.

Identity fragmentation quietly deflates P_return. If the same guest arrives via a different OTA, a different email address and a married surname, your data says two first-timers and your measured return probability drops. A meaningful share of real repeat behaviour is invisible for this reason alone, which means most operators' P_return is a floor rather than an estimate.

Watch the cohort age out. Center Parcs runs UK occupancy in the high nineties on families with young children. Once the children hit their teens, that household's return probability collapses regardless of how satisfied they were. A survival curve fitted on guests whose kids were five will overstate value for guests whose kids are eleven. The counter-example is the annual package buyer: where the cycle is already close to yearly, a survival model adds cost and very little information, and a simple year-on-year repeat rate does the job.

Discounting and the payback gap

Undiscounted multi-year LTV is the number that gets marketing teams into arguments with finance. In the example above, second-stay revenue lands around year 2.5 and third-stay revenue around year 4.5. At a 10% discount rate, $2,600 arriving in year four is worth about $1,776 today, and the whole post-first-stay block of $1,969 shrinks to roughly $1,470, a haircut of about a quarter. The ratio still clears the hurdle, which is the point: run the discount before someone else does it for you.

The harder constraint is cash, not ratio. Acquisition is paid this quarter; the second stay funds it in year three. A group growing its new-guest count fast enough will be cash-negative on a healthy LTV:CAC, and the annual budget cycle has no natural place to hold a three-year promise. Two decisions follow. First, publish a months-to-payback figure alongside LTV, so the finance conversation is about timing rather than credibility. Second, decide explicitly whether OTA-sourced guests carry the same expected P_return as direct guests; commission at the 15% to 25% range typical of the sector already lowers the margin on the first stay, and if those guests also return less often, funding them out of the same budget line is a slow, invisible mistake.

🎬 [VIDEO: "Customer Lifetime Value Explained" - youtube.com/results?search_query=customer+lifetime+value+explained+marketing - search for concise CLV walkthroughs from marketing analytics educators to see the standard formula before applying the travel-specific adjustments in this lesson]

Key takeaways

• Standard LTV formulas assume continuous purchase cycles and break for infrequent, high-ticket travel where the gap between stays runs years.
• Estimate the probability of return over a three to five year horizon times expected revenue per return, and use a method that handles censored data, since short history biases measured return rates downward.
• Model when guests come back, not only whether: return hazard clusters on school holidays and annual booking seasons, so dormancy is often better counted in seasons missed than months elapsed.
• Include referral value explicitly, and treat identity fragmentation across OTAs and email addresses as a reason your measured repeat rate is a floor.
• Discount the future stays before finance does, and report months-to-payback next to the ratio; a 6:1 undiscounted case can still be a cash problem in a fast-growing year.