Everyone retreated from transborder. One carrier flew into it.

Change in scheduled transborder seats, January–April 2024 against the same months in 2026. Every Canadian carrier reduced capacity except one.

US carriers held capacity almost exactly flat.

That asymmetry is the most informative thing in the data. Canadian carriers principally serve Canadian-originating traffic; US carriers serve US-originating traffic. Capacity retreating on one side of the border while holding on the other is directional evidence about which side of the demand actually moved — inferred from what carriers did, not asserted.

Two different responses to the same contraction

Load factor separates capacity discipline from capacity retreat. Air Canada removed 16% of its transborder seats and load factor rose — supply was pulled ahead of demand, and yield was protected. WestJet removed 19% and load factor still fell, from 0.888 to 0.829: demand moved faster than the schedule could.

Porter added 55% and load factor did not move at all — 0.729 in 2024, 0.743 in 2025, 0.729 in 2026. Capacity and traffic grew together, from the thinnest base in the market.

Read naively, that looks like underperformance. It is not — or not mostly. Load factor is not comparable across carriers flying different equipment on different missions, and Section 02 shows what happens to this gap once fleet is held constant. The headline number is a mix effect.

Comparing carrier-level load factors assumes the carriers are flying comparable missions. They are not. Porter operates two fleets with entirely different economics, and the headline figure is an average of them.

On like-for-like equipment the gap nearly disappears.

Air Canada's 137-seat aircraft averages 1,206 miles per departure at a 0.792 load factor. Porter's 132-seat aircraft averages 1,319 miles at 0.767. Five seats of gauge apart, a hundred miles of stage length apart, same market, same four months — a 2.5 point gap, not the nine points the fleet-wide average implies.

The drag is the 78-seat turboprop network, running 0.607. That fleet is roughly a quarter of Porter's transborder capacity and pulls the whole average down. And it is not a Porter-specific weakness: WestJet flies the same 78-seat type transborder at 0.573, below Porter, while Jazz reaches 0.745. The type and the mission carry most of that variance.

This matters for what follows. The commercial question is not "why is Porter underperforming," because on a fleet-adjusted basis it largely is not.

What can be stated precisely is narrower. Porter flies the highest proportion of empty seats of any Canadian carrier on transborder routes — 27% of capacity in January–April 2026, against 18% for the Air Canada group and 17% for WestJet. In absolute terms Air Canada flies far more empty seats than Porter does; the difference is rate, not volume. And Porter is the only carrier among them whose published fare conditions include checked baggage and assigned seating at no additional charge.

Fare data for Canadian carriers is not published anywhere. DB1B, the US ticket survey, covers US carriers only. Establishing revenue per passenger would require assembling priced itineraries across routes, booking windows and days of week over a sustained period to estimate a central tendency — a data collection exercise, not an inference from capacity data. Porter may price its base fare above competitors precisely because more is included. Nothing in the public record settles it.

So the argument that follows rests on the two established facts and not on the third: there is a measurable pool of unsold seats concentrated on identifiable routes, and there is a published product difference. Whether that difference is currently priced correctly is the open question — and it is answerable by test, which is what Section 05 sets out.

What this analysis cannot see

Three limits worth stating plainly, because each one bounds the conclusion.

Costs are unknowable. US carriers file Form 41 financial schedules, including aircraft operating expenses by type. Foreign carriers file only traffic data under T-100(f). Porter is privately held, so there is no annual report either. Its CASM, crew costs, lease terms and gate fees are genuinely unavailable, and a carrier with lower unit costs is profitable at a load factor that would not sustain a competitor. Nothing here should be read as a claim about Porter's margins.

These are flown passengers, not sold seats. T-100 reports passengers transported. Sold load factor is higher than flown load factor for every carrier, and plausibly more so for a business-heavy network on flexible fares. That narrows the gap further rather than widening it — and it means some revenue is already being collected on seats that fly empty.

Block hours are estimated. Foreign carriers do not report aircraft hours, so any block-time or cost-per-hour figure here is derived from stage length by a stated formula, not measured.

The unsold seats are not spread evenly. They concentrate on the business network — Billy Bishop and the northeast US — while the leisure routes run close to full. Select a carrier and year to compare.

This inverts the obvious recommendation.

Ancillary pricing is not the answer to a half-empty business route. Those seats are a demand and schedule problem, and a bag fee is close to irrelevant to a Monday-morning business itinerary. Adding fees to a route that cannot fill half its seats is the wrong instrument entirely.

The opportunity is the opposite set. Porter's nine Florida and Nevada leisure routes carried 306,153 passengers in 375,408 seats over January–April 2026 — a 0.816 load factor, 8.7 points above Porter's transborder average. Two fifths of the capacity, carrying nearly half the passengers, and effectively subsidising the business network.

Those passengers shop hard on fare and demonstrably pay for extras on other carriers. They are also, critically, not the passengers whose loyalty rests on the all-inclusive promise — that is the business traveller on the Billy Bishop network, who is not in this test.

A carrier cutting capacity 16% might be responding to demand, to cost, or to price. Fuel is the largest variable cost in the operation and the first objection any revenue manager will raise, so it has to be eliminated before the reading in Section 01 stands.

Between 2023 and 2025 Gulf Coast jet fuel fell 24%, from $2.85 to $2.16 per gallon. Canadian carriers reduced transborder capacity across that period. Cheaper fuel makes marginal flying more attractive, not less — a cost-driven retreat in a falling-cost environment is not a coherent story.

In 2026 fuel rose 38% to $2.98 and the pattern held: Canadian capacity fell a further 13%, US capacity fell 1.2%. Across all four years US transborder seats stayed within a 1.2% band while facing identical fuel prices.

On attributing cause

The 2025 trade and border environment is the obvious candidate explanation for a Canadian-side demand shift. This analysis does not claim it.

T-100 measures what carriers flew. It cannot identify why. What the data supports is the directional finding — Canadian-side capacity retreated while US-side capacity held — and the elimination of fuel as a sufficient explanation. The direct measurement of the demand side exists: BTS publishes Canadian Travel to the U.S., monthly arrivals with province of origin, destination state, trip purpose and spending. That is the dataset that can speak to causation, and it is the correct next step rather than an inference from seat counts.

Stating the limit is not a hedge. A recommendation built on a causal claim the data cannot support is a recommendation that fails its first serious review.

Hypothesis

Seat selection is chosen deliberately over checked bags. A bag fee contradicts the all-inclusive promise directly and visibly. Paid advance selection is an upgrade to an existing free service — seats remain assigned at no cost at check-in — so the promise stays intact and the product is additive rather than subtractive. That distinction is the entire reason this test is defensible.

Design

Randomisation unit: booking session, not passenger. Passengers on one booking must see the same offer, or the experience is incoherent and the cost lands on the contact centre.

Assignment: 50/50, stratified by route and days-to-departure. Leisure booking curves are long and seasonal; unstratified assignment will imbalance on advance-purchase window and confound the result.

Primary metric: ancillary revenue per booking.

Guardrails, any one of which stops the test: booking conversion rate; contact-centre volume per thousand bookings; post-flight brand-perception score on treated routes; and check-in-time seat-request volume, which measures whether the free path is being pushed rather than chosen.

Sample size

Testing a 2 percentage point absolute difference in take rate around an assumed 25% baseline, at α = 0.05 and 80% power:

n per arm = 2 × (1.96 + 0.84)² × p(1−p) / δ²
          = 2 × 7.84 × 0.1875 / 0.0004
          ≈ 7,350 bookings per arm

The nine leisure routes carry 306,153 passengers over four months — about 76,500 per month. At roughly 1.6 passengers per booking that is approximately 47,800 bookings monthly, so both arms reach sufficient sample in nine days.

The volume is there. There is no reason to guess, and no reason to prefer a phased rollout to a real test.

Duration: four weeks regardless. Nine days satisfies power; four weeks covers two complete weekly booking cycles and avoids a result driven by a single promotional period.

Stop rule: fixed horizon, single analysis at four weeks. No peeking. Where interim monitoring is required for guardrails — and it should be, for conversion — use a pre-specified alpha-spending boundary rather than repeated naive tests. The most common failure in commercial A/B testing is stopping the moment a metric crosses significance, which inflates the false positive rate substantially.

What would make this test wrong

Stated in advance, because a design that cannot fail is not a test. If the leisure and business networks share a booking flow, treatment may leak to business passengers and contaminate the brand guardrail. Load factors above 0.85 mean seat inventory is genuinely scarce, so a paid option may reallocate rather than create revenue. And four weeks in Q1–Q2 does not test peak summer behaviour — the result should be treated as seasonal until replicated.

Three moves, ranked by expected effect on revenue per passenger against the brand risk each one accepts.

Summary

The through-line: with load factors this soft and share this newly won, the recoverable revenue is per-passenger and additive. It is not per-seat, and it is not subtractive.

What would complete this analysis

The gap in the public record is fare. Neither Porter's base fares nor its competitors' are published in any dataset, so this brief can establish a product difference and a capacity position but not a price position.

Closing that gap means systematic fare collection: priced itineraries for Porter and its comparators, sampled by route, booking window and day of week across a full booking curve, over enough weeks to estimate a stable central tendency rather than a snapshot. Booking-window stratification is the part that matters — a single-day price scrape measures whatever the revenue management system happened to be doing that morning, not the fare structure.

With that in hand, three things become answerable that are not answerable now. Whether Porter's base fare already carries a premium reflecting what is included, which would materially change the case for Recommendation 01. Whether the fare premium, if it exists, is smaller than the value of the included services, which is the quantified version of "unpriced value." And what the actual elasticity looks like on the leisure routes, which would replace the illustrative 25% take rate in the test design with an estimate.

The full dataset is loaded into this page. The SQL below runs locally, on your machine — nothing is sent anywhere, nothing needs to be logged into, and every figure in this brief can be reproduced from the queries here. Edit the query or use the shortcuts.