Replacing Government and Inbound International Demand: A Segment Rebuild Playbook
Per diem room nights fell, Canadian visits dropped by a fifth, and U.S. Travel expects international visitor spending to fall again in 2026. Hotels that lost those guests can win replacement demand without buying it back on the OTAs: a segment audit built from PMS data, demand sensing from flight search, event and forward booking signals, CRM mining of past guests and accounts, and segment-specific offers pushed through the booking engine and CRM.
Every Monday segment report I read at Starwood had the same shape. Market codes down the left side, this year and last year across the top, and a variance column that went red before anyone in the room said a word. Across more than 800 hotels, the hotels that recovered fastest from a lost segment were rarely the ones with the best sales team. They were the ones that knew, by Wednesday, which nights the hole was on and who else wanted those nights.
That's the problem a lot of select-service and midscale hotels have right now. Two segments that used to show up without much effort have thinned at the same time.
Government per diem travel is the first. Inbound international, Canada most of all, is the second. Neither is likely to come back on its own schedule, and the reflex fix, opening the floodgates on the OTAs at a lower rate, refills the rooms at the worst cost the hotel can pay.
This piece is about doing it differently. The machinery is a segment rebuild run on data: an audit of where the room nights went, demand sensing that finds who wants those dates, CRM mining that finds those people in the hotel's own records, and segment-specific offers delivered through the booking engine and CRM. I owned ten hotels and built Collective Retreats to ten locations before it was acquired in 2025, and the most expensive room night I ever sold was always the one I bought back after losing it.
The Problem: Two Segments Left at Once
Kalibri Labs tracked government per diem room nights down 9 percent year to date in early 2025, with 76 percent of 334 U.S. markets declining and 30-day forward government bookings down 20 percent. Washington was down 20 percent year to date and 44 percent on the forward view. Government-reliant markets took it hardest: Santa Fe down 34 percent, Huntsville down 28. Economy and midscale hotels saw total demand fall 1 to 3 percent in the same data.
A parallel AAHOA survey with Kalibri found nearly half of members reporting fewer government per diem bookings, and fewer than 5 percent reporting growth in government or general business. Then came the fall shutdown. CoStar data reported by Skift showed Washington metro occupancy down 6.2 percent for October 1 to 25, after declines of 3.9, 5.2 and 5.8 percent in the prior three months.
The per diem itself didn't move. GSA held the FY2026 standard lodging rate at $110, the same as the year before, while CoStar put the 2025 U.S. ADR at $160.54. The government guest who does still travel fits fewer hotels.
International is the second leg. U.S. Travel counted Canadian visits falling from more than 20 million in 2024 to about 16 million by the end of 2025, and Cirium seat data in the same report had Canadian capacity to Las Vegas down 27.6 percent year over year. Inntopia measured Canadian bookings to mountain resorts down about 40 percent. Total inbound visits fell 5.5 percent in 2025, and U.S. Travel now forecasts international visitor spending down 6.3 percent in 2026, with arrivals not back to 2019 levels until 2030.
| Segment signal | Measure | Change |
|---|---|---|
| Government per diem, U.S. | Room nights year to date, early 2025 | -9% |
| Government per diem, U.S. | 30-day forward bookings vs 2024 | -20% |
| Government per diem, Washington | 30-day forward bookings vs 2024 | -44% |
| Washington metro, shutdown | Occupancy, October 1-25, 2025 | -6.2% |
| Canadian visitors | Visits, 2025 vs 2024 | Above 20M to about 16M |
| Canadian visitors, mountain resorts | Bookings, 2025 vs 2024 | About -40% |
| International visitors, U.S. | Spending forecast, 2026 | -6.3% |
The Data: Why the OTA Refill Costs So Much
When a segment disappears, the fastest dial to turn is OTA visibility. It works. It's also the most expensive way to fill a room the hotel used to fill for almost nothing.
Robert Rauch's 2026 math in Hotel-Online puts OTA commissions at 15 to 25 percent of the room rate against $3 to $7 a night for a direct booking once card fees and loyalty perks are counted, and cites Kalibri Labs on OTA business requiring up to 6.5 times the acquisition investment of a direct booking. A government guest booked on a negotiated or per diem rate code usually cost the hotel close to the direct number. Replace that guest through an OTA and the hotel is paying four to seven times as much for the same night.
Run it on a composite I'll use through the rest of this piece: a 140-room select-service hotel near a federal facility, 70 percent occupancy, $142 ADR, with government at 15 percent of room nights and inbound international at 6 percent. Lose a quarter of the government business and a fifth of the international and the hotel is down about 1,770 room nights a year, roughly $251,000 in rooms revenue.
Refill all of it through OTAs at 20 percent and the hotel hands back about $50,300 in commission. Refill it direct at $5 a night and the bill is about $8,900. The gap is $41,400 a year on one hotel, before counting that OTA guests are harder to bring back.
| Segment | Before (room nights, share) | After the loss | After the rebuild (target) |
|---|---|---|---|
| Government (per diem) | 5,366 (15.0%) | 4,024 | 4,024 (11.2%) |
| Inbound international | 2,146 (6.0%) | 1,717 | 1,717 (4.8%) |
| Corporate transient, local accounts | 7,869 (22.0%) | 7,869 | 8,169 (22.8%) |
| Domestic leisure, drive market | 13,593 (38.0%) | 13,593 | 14,293 (39.9%) |
| Group and SMERF | 5,008 (14.0%) | 5,008 | 5,458 (15.2%) |
| Extended stay and project crews | 1,789 (5.0%) | 1,789 | 2,139 (6.0%) |
| Total (occupancy) | 35,771 (70.0%) | 34,000 (66.5%) | 35,800 (70.1%) |
Notice what the rebuild doesn't do. It doesn't try to win back the government business. It adds about 1,800 room nights spread across four segments the hotel already serves, and it adds them through channels the hotel controls.
The Framework: Find the Hole, Find the Buyer, Reach Them Direct
The rebuild has four moving parts, and each one is software doing work that a director of sales used to do by feel.
First, a segment audit that shows exactly which nights the lost demand used to occupy. Second, demand sensing that reads forward signals for those nights by segment and origin market. Third, CRM mining that turns the hotel's own history into a list of people and accounts likely to want those nights. Fourth, offers built for each segment and delivered through the booking engine and CRM, with OTAs left as the backstop.
The order matters. A hotel that starts with offers ends up discounting nights that would have sold anyway.
Part 1: A Segment Audit Built From the PMS
Most hotels think they know their segment mix. Then they pull the reservation data and find a third of the stays coded to a catch-all like "transient retail."
The audit starts with a nightly export from the PMS or CRS at the reservation level: stay dates, rate code, market code, source, company or travel agent profile, guest address country, and email domain. Two years is enough. That export lands in a small database or a cloud spreadsheet the revenue manager can query, and it refreshes every night without anyone touching it.
Then a classifier cleans the coding. Simple rules catch most of it: a GOV or per diem rate code, a .gov or .mil email domain, a booking country of Canada or Mexico, a company profile tied to a federal contractor. A language model can read the free-text fields that rules miss, such as a comment that says "TDY orders attached" or a group name that's obviously a contractor. The output is a corrected segment tag on every stay, with the rule or model that assigned it.
With clean tags, the hole stops being a percentage and becomes a calendar. Government demand at a hotel near a base or an agency office usually sits Sunday through Wednesday, outside federal holidays. Canadian drive demand sits on winter weekends and long stays in Sun Belt markets. The audit shows, night by night, how many rooms each segment used to fill and how many it fills now.
That calendar is the brief for everything that follows.
Part 2: Demand Sensing by Segment
Historical data tells a hotel what used to happen. Forward signals tell it whether the nights in the hole are likely to fill some other way, and from where.
Lighthouse has published the clearest research on how early the signals arrive. In its original study, demand for one market showed up in flight searches about 145 days before arrival and hotel searches about 80 days out. Its 2024 to 2026 retest across 40 destinations found the same order, with flight searches correlating best with occupancy nine to ten weeks ahead. The link is strongest in leisure destinations and weakest in big business hubs, which matters for a hotel trying to replace weekday government nights.
The other sources fill in what flight search can't. Amadeus Demand360 gives on-the-books occupancy for the market across 35,000-plus hotels with a year of forward view, broken out by channel. PredictHQ tracks events in 19 categories, and the Hospitality.today review notes that 85 percent of recurring events change timing or location between occurrences, so a calendar copied from last year is often wrong. Google's free Travel Insights shows search volume by origin country for a destination.
| Replacement segment | Leading signal | Where it comes from | Typical lead |
|---|---|---|---|
| Domestic leisure, fly-in | Flight searches to your airport by origin city | Lighthouse Market Insight, Sojern, Google Travel Insights | 9-10 weeks for best fit |
| Domestic leisure, drive | Hotel searches and booking engine searches by date | Booking engine analytics, Google Hotel Insights | 4-5 weeks |
| Group and SMERF | Tournaments, reunions, conferences, school calendars | PredictHQ, local CVB, sports league schedules | Months, often a year |
| Corporate, local accounts | Market on-the-books by GDS channel, new local employers | Amadeus Demand360 and Agency360, local permits and hiring news | 2-8 weeks |
| Extended stay, project crews | Construction permits, plant openings, disaster response | Public permit data, state contract awards | 1-6 months |
The mechanism here is a weekly job, not a dashboard someone has to remember to open. Each Monday, a script pulls the forward signals for the next 120 days, lines them up against the hole calendar from Part 1, and produces a short list: these 14 nights are short by more than 15 rooms against last year, and here's what the signals say about who's coming instead. A model trained on the hotel's own history can weight each signal by how well it predicted pickup in the past.
That short list is what the sales team works from. It replaces the meeting where everyone agrees "we need to drive weekday business" and nobody knows which weekdays.
Part 3: Mining the CRM for Replacement Guests
The cheapest replacement guest is one who has stayed before. Most hotels have thousands of them sitting in the PMS and never write to them.
Revinate's 2026 benchmark found hyper-segmented email, a database split into four to seven groups, drives four times the revenue of batch-and-blast. The same report found complete guest profiles at just 11 percent in North America and email revenue down 19 percent globally. The data exists. It's mostly unused.
Mining starts from the cleaned stay history in Part 1. Group past guests by the segment they belong to and by the nights they stayed. A clustering model, or plain rules for a smaller hotel, finds the useful pockets: drive-market leisure guests within four hours who stayed on a weekend and never on a weekday; company profiles with two or more stays in 24 months that haven't booked in six; team managers and reunion organizers from past SMERF blocks; crews that stayed 14-plus nights on a project.
Then score each record for fit against the hole calendar. A drive-market couple who stayed twice in October is a good target for a Sunday-to-Tuesday shoulder offer. A contractor who booked a crew for a three-week job last spring is worth a call when a new permit shows up nearby. The CRM sends the first touch automatically; anything with a group or account behind it goes to a person.
One rule I'd hold to. Don't email Canadian guests a discount to come back. If they've stopped coming for political or currency reasons, a percentage off won't move them, and it trains every other segment that the rate is soft.
Part 4: Offers and Rates by Segment, Through Channels You Control
Once the hotel knows the nights and the buyers, the offers can be specific. That's what keeps them from turning into a public rate cut.
The booking engine carries closed-user-group rates that only show to a tagged audience: a member rate for past guests, a corporate code for a revived local account, a project rate for crews at 14 nights or more. Rauch suggests a member rate $5 to $10 below the public rate, which is still far cheaper than the commission it replaces. Length-of-stay rules do the rest: a Sunday-arrival, three-night package for the drive market fills exactly the nights the government business left.
The revenue management system needs to know about all of it. The Hotel Management guidance on federal travel shifts makes the same point, recommending forecasting by segment and testing group blocks against transient demand before cutting rate. In practice that means the RMS forecasts each segment separately, sees the closed-user-group rates as their own products, and holds a floor on the public rate so the hole-filling offers don't drag down nights that were going to sell.
OTAs stay in the mix. They become the last channel opened for a specific night, not the first, and only when the forward signals say the direct offers won't close the gap in time.
| Channel for a replacement room night | Cost basis | Cost per room night at $142 |
|---|---|---|
| OTA at 15% commission | Percentage of rate | $21.30 |
| OTA at 20% commission | Percentage of rate | $28.40 |
| OTA at 25% commission | Percentage of rate | $35.50 |
| Direct website booking | Card fees and loyalty perks | $3-7 |
| Member rate, $10 below public, booked direct | Rate given up plus direct cost | $13-17 |
| CRM email to past guests | Platform cost divided by booked room nights | Measure locally |
Hotels beginning a rebuild like this usually get the most from a structured look at forecasting and segment pricing first. That's the work behind our AI Revenue Optimization and Forecasting service, if you'd rather not build it alone.
Implementation: Ten Weeks to a Working Rebuild
None of this needs a new PMS. It needs clean exports, two or three subscriptions the hotel may already pay for, and someone who owns the Monday list.
| Weeks | Workstream | Output | Owner |
|---|---|---|---|
| 1-2 | Nightly PMS and CRS export, two years of history | Reservation-level table refreshed daily | Revenue manager with IT or vendor |
| 2-3 | Segment classifier, rules plus language model on free text | Corrected segment tag on every stay | Revenue manager |
| 3-4 | Hole calendar by night and segment | Nights short versus last year, by segment | Revenue manager, GM |
| 4-6 | Forward signal feed and Monday short list | Ranked list of short nights with likely replacement segments | Revenue manager, director of sales |
| 5-7 | CRM segmentation and scoring | Four to seven target groups with fit scores | Marketing, sales |
| 7-9 | Closed-user-group rates, packages, LOS rules, RMS floors | Segment offers live in booking engine | Revenue manager |
| 10 | First measured cycle | Replacement room nights by segment and channel cost | GM |
Measure one number above the rest: replacement room nights booked through direct and owned channels, by segment, against the hole calendar. If the hole is closing and the OTA share isn't climbing, the rebuild is working.
And keep the government business you still have. The audit will show which accounts are still traveling at per diem, and those guests are worth a phone call, not a mass email.
Back at Starwood, the red variance column never fixed itself, and nobody expected it to. What moved it was a sales team that walked out of the Monday meeting with names and dates instead of a mood. The software now does most of the work of producing that list. Someone at the hotel still has to pick up the phone.
Frequently Asked Questions
How much has government hotel demand fallen?
Kalibri Labs data covering more than 35,000 hotels showed U.S. government per diem room nights down 9 percent year to date in early 2025, with 76 percent of 334 markets declining and 30-day forward government bookings down 20 percent. Washington and government-reliant markets such as Santa Fe and Huntsville fell much further, and the fall 2025 shutdown pushed Washington occupancy down again.
Is international travel to the U.S. still declining in 2026?
U.S. Travel Association data shows inbound visits fell 5.5 percent in 2025 and Canadian visits fell from more than 20 million to about 16 million. U.S. Travel forecasts international visitor spending down 6.3 percent in 2026 to $163 billion, and does not expect arrivals to return to 2019 levels until 2030.
What data can a hotel use to find replacement demand?
Flight search data from providers such as Lighthouse and Sojern, booking engine search data, market on-the-books data from Amadeus Demand360, event data from PredictHQ or local sources, and the hotel's own reservation history. Lighthouse found flight searches correlate best with occupancy nine to ten weeks ahead and hotel searches four to five weeks ahead, with the strongest link in leisure destinations.
Why not just add OTA exposure to replace lost segments?
OTA commissions run 15 to 25 percent of the room rate, against $3 to $7 a night for a direct booking, and Kalibri Labs has found OTA business can require up to 6.5 times the acquisition investment. In a 140-room example losing about 1,770 room nights a year, refilling through OTAs at 20 percent costs about $50,300 against about $8,900 direct. OTAs work best as a backstop for specific nights.
How should a hotel segment its CRM to win back demand?
Start from cleaned stay history and split guests into four to seven groups tied to the nights the hotel needs to fill, such as drive-market weekend guests, lapsed local company accounts, past group organizers and long-stay project crews. Revinate's 2026 benchmark found this kind of hyper-segmentation drives four times the revenue of batch email. Score each record for fit and route accounts and groups to a person.
If you want help doing this
A segment rebuild is the kind of project AI for Hoteliers exists for: clean the PMS data, build the hole calendar, and get the Monday list running so the sales team works from names and dates. I spent years reading segment reports across 800-plus Starwood hotels and then owned ten of my own, so if your government or Canadian business has thinned, bring the numbers to a free 15-minute call and we'll find where the nights went.
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.