TL;DR: In a Wyndham Hotels & Resorts owner survey, 98% of hotel owners and developers said they had started using AI, yet only 32% had embedded it across most operations. Separately, Skift Research has reported that 63% of hotel tech budgets still go to maintaining legacy systems. For mid-market travel platforms, the pricing algorithm is rarely the bottleneck. The data plumbing underneath it is.
The gap between "using AI" and "running on AI"
The most useful number in recent travel-technology research is not an adoption rate. It is the distance between two adoption rates. Wyndham's survey of hotel owners and developers (a franchisor-run survey, so read it as an indicator rather than an audit) found that 98% had begun incorporating AI, while only 32% had embedded it across most operations. Revenue optimization was a stated priority for 53% of respondents, behind operational efficiency (64%) and energy management (54%). Notably, 73% said they wanted to expand AI use but felt overwhelmed doing so.
Phocuswright's 2026 research on agentic AI tells a similar story across the wider travel industry: over 60% of surveyed travel businesses are experimenting with or scaling agentic AI, but only 6% describe themselves as actively scaling it, and 22% as in early scaling. A separate Canary Technologies survey of 400+ hospitality technology decision-makers (a vendor survey) found 85% planned to put at least 5% of IT budget toward AI in 2026.
Read together, the pattern is consistent: intent and pilots are everywhere, production-grade deployment is rare. Revenue management is one of the places where that gap is most expensive, because pricing decisions compound daily.
Where AI revenue management delivers today
Revenue management is a natural fit for machine learning because it is a repeated, measurable decision with a clear outcome (revenue per available room, yield, load factor). Where AI is earning its place:
- Demand forecasting at finer granularity. Models can combine booking pace, events, search demand, weather and competitor rates more flexibly than rule-based forecasts. The benefit shows up most for properties or routes with volatile demand.
- Rate recommendation with human approval. The Wyndham survey found 57% of respondents want human supervision of AI decisions. That matches what works operationally: the system proposes, the revenue manager approves or overrides, and the override data improves the model.
- Segment and channel mix. Deciding what to hold back from low-margin channels is a pattern-recognition problem that suits ML well.
- Anomaly detection. Flagging rate-parity breaks, rate-loading errors or inventory mismatches before a guest or partner finds them.
Where it is overhyped or premature
- Fully autonomous pricing for a mid-market operator. Autonomy needs clean inventory, rates and booking history across systems. Few mid-market platforms have that on day one.
- "Plug-in AI" on top of a fragmented stack. Skift Research's finding that 63% of hotel tech budgets goes to maintaining legacy systems (many lacking AI compatibility or clean integration) is the sober counterweight to vendor demos. A forecast is only as good as the property management, channel and booking data feeding it.
- Customer-facing generative AI making commitments. In February 2024, British Columbia's Civil Resolution Tribunal ordered Air Canada to pay a traveler about C$812 after its chatbot gave wrong bereavement-fare information, rejecting the argument that the chatbot was a separate entity. The tribunal's reasoning was that a chatbot is part of the company's website. If your assistant quotes fares or policies, you own what it says.
- Personalized pricing. The line between dynamic pricing and so-called surveillance pricing is attracting legislative attention in the US, including congressional letters to airlines. Pricing that varies by individual rather than by demand and inventory carries reputational and regulatory risk that a pure revenue lift calculation will miss. Treat the regulatory picture as unsettled and get legal review before individualized pricing in any market.
Legacy systems versus AI: the real decision
The framing "AI versus legacy" is slightly misleading. For most mid-market platforms the decision is among three paths:
- Buy an AI-native revenue management system and integrate it. Fastest to a first result, but you inherit its data model and pricing logic, and integration quality determines the outcome.
- Modernise the data layer first, then add models. Slower to start, and the most defensible when your inventory logic is unusual (packages, multi-supplier itineraries, group blocks, ancillaries).
- Keep the legacy engine and wrap it. Use an integration layer to expose rates, inventory and bookings through clean APIs, then run forecasting or recommendation services alongside the old engine. This lowers risk but can entrench technical debt.
Implementation risks and mitigations
| Risk | What it looks like | Mitigation |
|---|---|---|
| Data quality | Duplicate bookings, inconsistent room or fare codes, missing cancellations | Audit and reconcile source data before model work; define one booking record of truth |
| Integration cost | Point-to-point connections to PMS, CRS, channel manager, GDS | Build a thin integration layer with stable APIs rather than another custom link |
| Black-box trust | Revenue managers override recommendations they cannot explain | Show drivers behind each recommendation; log overrides and review them weekly |
| Hallucination in guest-facing tools | Chatbot invents policy, fare or availability | Ground answers in policy and rate sources, restrict to answerable questions, hand off to humans, log everything |
| Regulatory exposure | Personalised or opaque pricing practices | Limit to demand- and inventory-based signals; get legal review per market |
| Change management | Team treats the tool as a threat or ignores it | Start in recommendation mode, define who owns the final rate, measure against a control |
Is your business ready? A checklist
- Can you pull clean booking, cancellation, rate and inventory history for at least two years from one place?
- Do you know your current forecast accuracy and have a baseline to beat?
- Is there a named owner for pricing decisions, who will review model recommendations daily?
- Can you run the new approach against a control (certain properties, routes or dates) for 8 to 12 weeks?
- Are your PMS, channel and booking systems reachable through documented APIs, or does every change require a vendor ticket?
- Do you have rules for what an AI tool is never allowed to commit to a customer?
- Have you considered how your pricing signals would read to a regulator or journalist?
If you answer "no" to the first or fifth question, the highest-return investment is probably integration and data work, not a model.
Where Syslabs fits
Most of the work in this space is unglamorous: reconciling booking data, building API development and integration layers over legacy PMS and channel systems, and designing human-in-the-loop workflows so revenue teams can trust and audit recommendations. Syslabs does this through custom software development for mid-market operators who need their stack to support AI without a full rebuild. We covered a parallel pattern in Back-Office AI for Travel Management Companies, where the returns sit in the plumbing, not the chatbot.
Sources
- Wyndham Hotels & Resorts owner and developer AI survey, as reported by AltexSoft
- Phocuswright, "Budgets, Barriers and the Race to Agentic AI" (2026)
- Canary Technologies, "Navigating AI: Hospitality Shifts From Exploration to Execution" (2026), as reported by Hotel Management
- Skift Research, as cited in Skift, "Hotels Risk a Tech Trap" (September 2025)
- Moffatt v. Air Canada, Civil Resolution Tribunal of British Columbia (February 2024)
Next step
If you are weighing AI revenue management against your existing stack, we can spend 30 minutes on an AI-readiness conversation: where your data stands, which path fits, and what to test first.