Case StudyRevenueMarketing

Case Study: Using AI to Increase Ancillary Revenue on Low-Cost Carriers

UUnknown

2026-02-19

10 min read

Hypothetical 2026 case study: how self-learning recommendation engines + AI video ads can lift LCC ancillary revenue by 12–28%.

Hook: Stop Leaving Easy Ancillary Revenue on the Table

Low-cost carriers (LCCs) live and die by ancillaries — seat selection, baggage, priority boarding — yet many airlines still rely on static offers and blunt segmentation. Travel teams and revenue managers face rapid fare swings, fragmented data, and the constant pressure to increase ancillary yield without degrading conversions. In 2026, the winning strategy is self-learning recommendation engines paired with AI-driven, targeted video ads that meet passengers where they are. This hypothetical case study shows how a mid-size LCC can boost ancillary revenue with a practical, measurable approach.

Executive Summary (Most Important First)

We model a 12–28% uplift in ancillary revenue per booking across three months by combining a self-learning recommendation engine with personalized, in-flow AI-created video ads. The engine optimizes offers in real time using contextual bandits and reinforcement learning. Video creatives are dynamically generated and served across the booking funnel and pre-departure channels. Key takeaways: start small, measure incrementally with holdouts, use first-party signals, and prioritize privacy-safe personalization.

Quick results snapshot (hypothetical)

Baseline ancillary attach rate: 23%
Post-implementation attach rate: 28–34%
Ancillary ARPU uplift: $3.80–$8.50 per passenger
Payback period: 6–10 weeks (for mid-size LCC)

Why This Matters in 2026

By late 2025 and early 2026, advertising platforms and airlines have fully adopted generative tools. IAB and industry reports show nearly 90% of advertisers using AI to build or version video ads, which means creative edge is now a differentiator rather than adoption itself. Meanwhile, privacy changes and cookie deprecation have accelerated reliance on first-party data and real-time recommendation models that learn from behavioral signals without excessive third-party tracking.

The Hypothetical Airline: "Coastal Air"

For this case study, imagine Coastal Air, a European LCC with 6 million annual passengers and three distribution channels: direct web/app bookings, OTA partners, and airport kiosks. Coastal Air wants to increase ancillary revenue from seat selection and checked baggage without lowering conversion on core ticket sales.

Business objectives

Increase ancillary attach rate by at least 15% within 90 days
Maintain or improve overall checkout conversion
Ensure all personalization is privacy-compliant (GDPR, CCPA-like rules)

Solution Overview

Coastal Air deploys a two-pronged system: a self-learning recommendation engine that personalizes ancillary offers in real time, and an AI-driven targeted video ad pipeline that serves micro-personalized creative during booking and as pre-departure prompts.

Core components

Data layer: First-party event stream (booking events, seat map interactions, bag searches), CRM signals, and flight context (length, aircraft, load factor).
Model layer: Contextual bandits for immediate offer selection and a reinforcement learning (RL) policy for longer-session strategies (bundles, sequencing).
Creative layer: Generative video templates that assemble short, 6–12s clips personalized by route, traveler type, and intent signals.
Serving layer: Low-latency API for real-time offers in the booking flow and programmatic delivery for in-app and pre-departure ads.
Measurement framework: Incrementality via holdout groups, conversion lift, and long-term retention metrics.

How the Self-Learning Recommendation Engine Works

The engine uses a hybrid approach: contextual bandits for immediate decisions and an RL policy for optimizing sequences across touchpoints. This balances fast learning with strategic longer-horizon rewards (e.g., a baggage sale now may enable a higher-margin bundle later).

Key signals and features

Trip characteristics: origin/destination, flight duration, fare class, day/time
Behavioral signals: pages visited, seat map hover patterns, search frequency
Ticket lifecycle: days to departure, previous ancillaries purchased
Contextual metadata: device, channel, load factor, weather at origin/destination

Why contextual bandits?

Contextual bandits are ideal for real-time personalization where exploration-exploitation tradeoffs matter. They allow the system to try different ancillary offers for similar contexts and rapidly converge on higher-revenue options without significant manual intervention.

Reinforcement learning for sequencing

Use RL when you need to optimize across multiple steps: the model learns policies like "offer seat selection at checkout; if declined, offer low-cost 24-hour upgrade via push message". RL maximizes lifetime ancillary yield rather than immediate per-touch revenue.

Targeted Video Ads: Creative + Signal Strategy

Video is powerful because visuals speed decision-making. In 2026, nearly all advertisers use generative tools to build video, shifting wins to teams that pair data-rich signals with strong creative inputs. Coastal Air uses short AI-generated video variants tailored by micro-segment.

Video ad use cases

In-flow booking: 6s explainer of seat benefits when a user hovers over the seat map
Cross-sell at payment: 8–12s dynamic bundle demo showing luggage and priority boarding for this route and price
Pre-departure nudge: 6s push video offering last-minute baggage with a small discount

Creative best practices (2026)

Use short, attention-grabbing hooks in the first 2–3 seconds
Personalize visuals to route context (e.g., surfboard for beach destinations)
Test multiple value props: convenience, price, comfort
Include clear, frictionless CTAs integrated with one-tap upsell flows

"Nearly 90% of advertisers now use AI to build or version video ads — creative inputs and data signals determine performance." — industry trend, 2026

Implementation Roadmap (90-Day Pilot)

Focus on a narrow set of routes and ancillaries to validate the approach quickly. Coastal Air pilots on three high-density leisure routes where ancillaries historically underperformed.

Weeks 1–2: Data & instrumentation

Stream booking events to a central data lake (Kafka or managed streaming)
Instrument seat map interactions and ancillary impressions
Define holdout and test cohorts (10–20% holdout recommended)

Weeks 3–4: Baseline and creative templates

Measure baseline attach rates and ARPU
Create video templates with generative tools and human-in-the-loop edits
Define KPIs and monitoring dashboards

Weeks 5–8: Model deployment and A/B tests

Deploy contextual bandit for in-flow offers (low risk, quick wins)
Run controlled A/B tests with video variants (holdout vs. video vs. static creative)
Iterate on reward definitions (immediate revenue vs. longer-term retention)

Weeks 9–12: Scale and RL sequencing

Introduce RL policies for multi-touch sequencing
Extend coverage to mobile app and pre-departure channels
Measure incrementality, update pricing, and roll out best-performing creative

Measurement & Attribution (Avoid False Positives)

In 2026, incrementality is the gold standard. With broad AI adoption, naive lift estimates overstate impact. Use holdout groups and randomized controlled trials to isolate the effect of the recommendation engine and video ads.

Essential metrics

Attach rate (ancillary purchases per booking)
Ancillary ARPU (average ancillary revenue per passenger)
Checkout conversion rate (to ensure offers don't harm ticket sales)
Incremental revenue per exposed user (vs. holdout)
Lifetime value impact (if bundling affects future purchases)

Best-practice measurement

Use statistically powered holdouts (10–25% depending on traffic).
Run lift tests for video creatives and sequencing policies separately.
Combine short-term RCT results with long-term observational studies to account for retention and cross-sell effects.

Results — Hypothetical Outcomes

Coastal Air's pilot produced clear, measurable wins within 90 days. Here are the hypothetical outcomes the team observed after rolling the system across the three pilot routes.

Performance summary

Seat selection attach rate rose from 18% to 24% (+33% relative)
Checked baggage attach rate rose from 14% to 19% (+36% relative)
Total ancillary revenue per booking increased by $5.60 (roughly 18% overall uplift)
Checkout conversion remained neutral (±0.5%), confirming offers did not harm ticket sales

Where revenue came from

Dynamic seat nudges at seat-map hover (20% of uplift)
Personalized video at payment showing bundled savings (45% of uplift)
Pre-departure targeted offers via push with small discounts (35% of uplift)

Practical Playbook: How Your Team Can Replicate This

Whether you’re a travel tech product manager, revenue manager, or developer, follow this condensed playbook to get started.

1. Start with the highest-impact touchpoint

Begin at the seat map or payment stage — these have the best intent signals. Instrument hover, selection attempts, and abandonment to capture intent.

2. Use first-party data and privacy-first IDs

Rely on logged-in user signals, hashed emails, booking reference, and device signals to personalize without third-party cookies. Update your consent flows to surface personalization benefits.

3. Choose the right algorithmic approach

Contextual bandits for rapid, low-risk personalization
RL for multi-step sequencing when you have enough traffic
Offline models (XGBoost/transformers) for feature engineering and propensity scoring

4. Make video creative an iterative experiment

Use generative tools to produce many micro-variants. But keep human review for compliance and brand voice. Measure which hooks (price, comfort, convenience) win for each segment.

5. Measure incrementally, then scale

Use holdouts and RCTs for both recommendation logic and video creatives. Only scale winning strategies and continue to monitor for distributional drift.

Risk Management & Governance

Self-learning systems can drift or exploit loopholes. Governance is essential.

Key controls

Human-in-the-loop for new offer types and creative templates
Guardrails to prevent price gouging or discriminatory offers
Regular audits for model drift and fairness
Privacy-preserving logging and data retention policies

Costs & Tech Stack Considerations

Rolling a self-learning recommender + video pipeline requires infrastructure but not a full ML org. Coastal Air reused managed services to accelerate delivery.

Recommended stack (cost-conscious)

Streaming: Managed Kafka or cloud streaming
Feature store: Managed feature store or simple Redis cache for low-latency features
Model infra: Contextual bandit library (Vowpal Wabbit or commercial alternatives) + RL frameworks if needed
Creative: GenAI video platform with APIs and brand guardrails
Serving: Low-latency REST/gRPC API + CDN for video assets

Future Predictions (2026 and Beyond)

Looking ahead, ancillary monetization will be shaped by three trends:

Self-optimizing offers will become standard: models that learn in production and adapt to micro-seasonality will outperform static price ladders.
Creative-as-code will enable on-the-fly video personalization integrated directly into the recommendation loop.
Privacy-first personalization will dominate: airlines that can deliver value with first-party signals will capture the most sustainable uplift.

Common Pitfalls & How to Avoid Them

Avoid overfitting to short-term clicks — optimize for revenue and retention.
Don’t deploy creative variants without holdouts — you may confuse the measurement signal.
Guard against aggressive discounting that trains users to wait for offers.

Checklist Before You Launch

Defined KPIs and power calculations for holdout tests
Instrumentation for all relevant events (hover, impression, click, purchase)
Brand-safe video templates and human review workflow
Privacy controls and consent capture aligned to regulations
Monitoring dashboards for lift, conversion, and model drift

Final Thoughts

Self-learning recommendation engines and targeted video ads are not theoretical — they are practical levers for LCCs in 2026. When implemented with careful measurement, privacy-first data, and creative rigor, these tools can unlock meaningful ancillary revenue uplift without hurting ticket sales. The formula is simple: data + AI + creative + rigorous testing = predictable ancillary growth.

Call to Action

Ready to pilot a self-learning ancillary system for your airline or travel product? Contact botflight’s advisory team to map a 90-day proof-of-value, or download our technical playbook for engineers and revenue managers to get started with contextual bandits, RL sequencing, and AI video creative templates.

Unknown

Contributor

Senior editor and content strategist. Writing about technology, design, and the future of digital media. Follow along for deep dives into the industry's moving parts.