Forget Me Not

TL:DR

The finding. Twilio’s 2026 Customer Insights Series says seven in ten APAC consumers have walked away from an AI interaction because it did not know who they were or what they had already said. 84% of brands believe their AI recognizes returning customers. The customers disagree.

Why hotels should care more than anyone. Recognition is not a feature of hospitality. It is the product. A bank that forgets you is irritating; a hotel that forgets you is failing at the one thing it sells.

The cause. AI amnesia is rarely a model problem. It is the guest-facing symptom of what this week’s h2c study exposes: 91% of hotel chains use AI, only 28% have an enterprise-wide strategy and only 13% report measurable ROI — while 69% of hotels still key guest preferences in by hand. The AI hears the guest. The hotel forgets.

The fix. Make every conversation write back, build one governed Guest Memory Layer with minimum-necessary access per agent, standardize the warm handoff, disclose up front — and put two numbers on the GM’s dashboard: the Repeat-Yourself Rate (what the guest re-tells us) and Manual Touches Per Stay (what staff re-type).

1.  The study that named the problem

On 17 September 2026, Twilio released the Asia-Pacific cut of its 2026 Customer Insights Series — a survey of 7,652 consumers and 660 business leaders across 18 markets, including Hong Kong, India, Indonesia, Japan, Malaysia, the Philippines and Singapore.^1,3 Its headline needed no translation: consumers are abandoning AI agents that cannot remember them. Twilio calls it “AI amnesia” — systems that fail to retain or retrieve enough context for a customer to continue a conversation without repeating themselves.³

The numbers that matter:

Sources: Twilio 2026 Customer Insights Series as reported by FutureIoT, TechNode Global, Context.ph, and NewsBytes.PH [1, 3, 5, 6]. Philippine sample: 353 consumers, 35 business leaders, April–May 2026.

Two further details deserve a hotelier’s attention. First, in the Philippines, 46% of consumers said a fast AI agent without personal context was more frustrating than a slower one that remembered them — only 16% felt the opposite.⁶ Speed, it turns out, is not the service. Second, the region’s consumers are not anti-AI: 68% say bots have improved over the past year, and they are most comfortable delegating exactly the tasks hotels automate first — appointments, restaurant reservations and returns.¹

A fair caveat. Twilio sells the cure — its Conversation Memory product launched in May to tackle what it calls “agent amnesia.”¹⁰ The Philippine business sample of 35 is thin. Read the figures as directional, not actuarial. But the direction is unmistakable, and every hotelier reading this has lived it from the guest side of the counter.

2.  What AI amnesia actually is

“The AI forgot me” is a symptom. Underneath it sit at least five distinct failures, and they need different fixes. In hotel terms:

Twilio’s own diagnosis is blunt and, for once, a vendor’s line is the right one: without a unified data foundation, AI does not eliminate silos — it operationalizes them at machine speed.⁸ A forgetful human night auditor annoys one guest. A forgetful AI agent annoys every guest around the clock, with a timestamped transcript of its failures.

3.  Why hospitality is more exposed than most

Recognition is the product

Hotels have spent a century monetizing memory. The doorman who says “Welcome back, Mr Tan.” The bartender who starts the Negroni before you sit down. The housekeeper who leaves the extra feather pillow without being asked. Loyalty programs are, at bottom, a memory system with points attached. When an AI agent becomes the first voice a guest meets — and Twilio says that is already true of about half of APAC service conversations — the agent inherits the brand’s promise of recognition. It usually cannot keep it.

Our data was fragmented long before AI arrived

None of this is new. AI makes it audible. Consider what hotel technologists already know:

• The PMS forgets by design. Many PMS platforms create a new profile for every booking, even for the same guest; deduplication is left to the CRM — if there is one.¹²

• A fifth of the guest is invisible. Ireckonu’s data from large hotel groups shows one in five guest profiles has no room reservation at all — restaurant, spa, and golf guests who never appear in the PMS — leaving up to 20% of recognition and re-engagement opportunities on the table.¹¹

• Migrations erase history. Moving to a cloud PMS frequently means only part of the historical stay data makes the journey.¹³ The hotel literally loses its memory during a technology upgrade.

• The OTA owns the email address. Masked or relay addresses mean that, to the hotel's systems, the guest who booked through an OTA is a stranger with a borrowed name.

• Vendor sprawl. Chatbot from one vendor, voice AI from another, WhatsApp concierge from a third, GRMS and in-room tablet from a fourth. Each remembers a sliver. None remembers the guest.

The handoff is where trust breaks

Hotels will, rightly, keep humans in the loop for anything emotional, expensive, or complicated: the delayed flight, the wrong room, the anniversary that went sideways. These are exactly the moments a handoff happens — and exactly the moments when making a guest repeat themselves is most corrosive. Twilio’s 65% repeat-after-handoff figure should be read in a hotel context as: two in three guests arrive at your front-desk agent already irritated, through no fault of the agent.

The cost lands in the wrong ledger

Twilio’s 5% “ended the relationship” figure looks small. It is not. In hospitality, churn is silent — the guest simply books elsewhere next time, often through an OTA that does remember them. The cost never appears against the chatbot line item. It appears months later as a softer repeat rate, a higher acquisition cost, and a commission bill. Meanwhile, the AI dashboard reports containment and handle-time savings, and everyone feels good about it.

4.  The absorption gap: why amnesia is a symptom, not the disease

Twilio told us what guests feel. A second study, published two weeks later, tells us why. h2c’s AI Opportunity Study 2026 — 113 hotel chains, more than 8,200 properties and roughly 1.3 million rooms, plus 230 documented AI implementations — finds AI in use at 91% of chains, up from 78% a year earlier. Company-wide AI strategies have quadrupled, from 7% to 28% — yet that still leaves 72% without one, and 49% running AI as experiments or pilots. Only 13% report measurable ROI, and chains rate AI’s contribution to business performance at 5.6 out of 10 (Asia-Pacific: also 5.6).¹⁸ Most deployments remain tactical, departmental or fragmented.^18,19

Look at where the benefits land. 67% of chains report better operational efficiency and 59% say AI frees staff for higher-value work — but only 32% cite an improved guest experience.¹⁸ Hotel AI, so far, has mostly been pointed inward. The guest-facing wave is next: 55% of chains plan guest-facing AI agents and 70% plan staff-facing ones. On today’s foundations, that is a plan to scale amnesia.

Talking more, remembering less

One chart in the h2c report should stop every hotel board in its tracks. Across ten AI use cases tracked year on year, almost everything grew: chatbots in use rose from 42% to 64% of chains, forecasting from 22% to 52%, guest engagement and the digital guest journey from 29% to 46%. One category went backward. AI-powered personalization fell from 22% to 16% in use — the only decline — even as 59% say they plan to use it.¹⁸ Read with Twilio, the pattern is stark: hotel AI is talking to guests far more and remembering them less. That is the arithmetic of amnesia.

Asia-Pacific adds a regional twist. Among chains using or planning AI agents, 75% in APAC plan staff-facing agents but only 33% plan guest-facing ones — the lowest guest-facing share of any region (the sub-sample is small, so treat it as indicative).¹⁸ Yet Twilio says AI already handles roughly half of APAC customer-service conversations. The implication: many APAC guests will meet AI agents that are not the hotel’s — OTA assistants, super-app agents, platform concierges — and those agents will remember the guest even if the hotel does not. APAC hoteliers have a short window to fix memory foundations before guest-facing agents arrive; otherwise, the guest’s memory ends up with someone else’s agent. Hyatt’s Mark Hoplamazian, for one, already expects AI assistants to become the primary way guests book.²⁹

Read that alongside Twilio and the mechanism is obvious. A guest-messaging bot bought by Front Office, a voice agent bought by Reservations, a review-response tool bought by Marketing, an upsell engine bought by Revenue: each department adopted AI, and each AI has its own little memory. Channel amnesia is not a malfunction of that estate. It is the estate working exactly as it was assembled.

Sources: h2c Hotel Chain Tech Report 2026, pp. 9–22 [18]; NYU SPS/RateGain/HEDNA State of Distribution 2026 [21]. h2c results are global and unweighted by chain size, with selected regional splits; sample sizes vary by question (N = 67–122). Twilio’s figures cover all APAC brands; h2c covers hotel chains specifically.

The amnesia begins at the write, not the read

The most important number for this paper is not Twilio’s. It is this: 69% of hotel chains still enter or enrich guest preference data manually, only 19% use AI to extract preferences from guest interactions — and another 19% do not actively enhance preference data at all.¹⁸ Even among large chains, 55% rely on manual entry. Take an ordinary pre-arrival message:

We’ll arrive late. My wife is allergic to shellfish, and we’d appreciate the same firm pillows we had last time.

Most hotels now have AI that will draft a charming reply. But does “late arrival” reach Front Office? Does “shellfish allergy” reach the F&B workflow and the profile? Does “firm pillow” reach Housekeeping? Or does someone copy and paste it into three systems — if they have time, on that shift?

If it is the latter, the hotel automated the sentence and not the work. The AI heard the guest perfectly. The hotel then forgot — not at the moment of retrieval, but at the moment it should have recorded. Every AI amnesia complaint Twilio measured on the read side has a twin on the write side. AI must move from writes to understands → structures → routes → acts → records. The last verb is the one that cures amnesia.

Where the hotel’s memory actually lives

h2c offers a second, quieter clue. Sixty per cent of daily AI use in hotel chains runs through external tools such as ChatGPT; 27% is vendor-embedded and only 12% internally built. Reliance on vendor-provided AI has risen to 31%, from 21% a year earlier. Internal AI knowledge is self-rated at 3.4 out of 10, and 56% cite a lack of AI skills as the leading barrier.¹⁸

Put plainly: a great deal of what hotel staff learns about guests is now being processed in browser tabs and vendor logs that the hotel neither owns nor reads back. A reservations agent pastes a guest’s email into a public chatbot to draft a reply; the reply is good; the insight evaporates. Call it shadow memory. It is not just a privacy exposure — it is the hotel’s recognition capability leaking out of the building one conversation at a time.

Hours saved are not guests kept

Hyatt’s SVP of Data and AI, Pat Nestor, framed the industry’s shift this week as moving from adoption to absorption — and noted that hotel companies know what AI costs far better than what it earns.²² AI amnesia sits precisely in that blind spot. A chatbot’s business case counts hours saved and conversations contained; the guest who quietly doesn't come back never makes it into the spreadsheet. A memory program therefore needs its own Value Conversion Path: fewer repeated questions → shorter handoffs → higher repeat-stay rate and direct-booking share → lower acquisition cost and commission. Measure each link, or the CFO will be right to doubt it.

The Pertlink lens: memory has a token cost

Under our Token Cost Per Guest (TCPG) framework, memory is not free. Every remembered preference injected into an AI conversation consumes context and compute. The answer is not to remember everything — it is to remember the right things, structured and retrievable, so the agent loads a concise guest brief rather than a 40-page stay history. Well-designed memory lowers TCPG; a dumped transcript raises it.

Agoda’s CTO, Idan Zalzberg, made the same point from the engineering side: “The real cost is not simply the model call.” The larger cost is the work that makes AI trustworthy in production — and clean, governed guest memory is a large part of that work.²⁸

5.  The paradox: remember more, but not creepily

Here is the uncomfortable part. The same guests who resent being forgotten also resent being watched. A hotel that recalls your pillow preference is gracious; a hotel that recalls the name of the person you shared a room with last March is alarming. Twilio’s data shows the trust conditions consumers attach: the option to switch to a human (58%), assurance that AI actions require human approval (57%), and visibility into what data the AI can see (51%).¹

Regulation is moving the same way. Article 50 of the EU AI Act has applied since 2 August 2026, requiring chatbots and AI agents to tell people they are interacting with AI, at the latest at the first interaction — and it reaches non-EU businesses whose chatbots serve EU users.^15,16 Any APAC resort with European guests is in scope. In the Philippines, the Data Privacy Act (RA 10173) already requires lawful basis, proportionality, and transparency for processing personal data.¹⁷ Against that backdrop, Twilio’s finding that only 22% of APAC businesses disclose AI up front is not just a trust gap; for some properties it is a compliance gap.

Hoteliers feel the same tension from the other side of the desk. Asked about guest-facing AI, chains in the h2c study cite unclear liability for AI errors (56%), data privacy and security (54%), guests’ reluctance to share personal data (53%), and accuracy and bias (52%).¹⁸ Guests want to be remembered; hotels fear remembering. Both are right, and neither is served by today’s fragmented middle. Note, too, that chains’ reliance on AI rose from 4.7 to 5.4 out of 10 in a year while trust stayed flat at 6.5: hotels are leaning on AI more without trusting it more.¹⁸

Remember centrally, reveal minimally. The way out of the paradox is architectural. The hotel can hold a rich, consented guest memory in one governed place, while each AI agent is given only the Minimum Viable Hotel Context its task requires: the minimum data (these six fields, not the whole PMS), the minimum authority (read, not modify) and the minimum duration (this reservation, this stay). It is least-privilege access, translated into hospitality. The airport-transfer bot does not need the guest’s spa history; the F&B agent must see the shellfish allergy. Memory is complete at the center and deliberately partial at the edge.

The principle: consented, purposeful memory. Remember what serves the guest, tell them you remember it, and let them see and edit it. Forget on request. Memory the guest can inspect is hospitality; memory they discover by accident is surveillance.

6.  The mitigation playbook

Six moves, in order of dependency. None requires waiting for a new AI model.

Move 1 — Resolve the guest before you remember the guest

Identity resolution comes first. Merge duplicate PMS profiles; link restaurant, spa, and event guests to the same person; match OTA-masked bookings at check-in with a consented capture of a real contact channel. Without a single guest identity, no amount of AI memory helps — the agent will faithfully remember the wrong person.

The most convincing hotel AI result of the season proves the order matters. Wyndham’s voice agent, built on Salesforce Agentforce, now answers calls at properties that previously went unanswered roughly a quarter of the time. But CEO Geoff Ballotti was explicit about the prerequisite: a half-billion-dollar technology-stack migration that began in 2016, consolidating dozens of PMS platforms down to two, long before any AI was layered on top.²⁶ Wyndham did not buy memory. It built the ground memory could stand on.

Move 2 — Make every conversation write back

Before buying memory, stop leaking it. Every guest-facing AI conversation should end with structured outputs — preferences, occasions, open issues, commitments made — routed to the owning department and recorded on the profile, with a human confirming anything safety-critical such as allergies. Specify it in the workflow, not in the hope that a busy agent will copy and paste. If 69% of preferences are hand-keyed today, the first win is not a smarter model; it is closing the loop between what the guest said and what the hotel recorded.

Move 3 — Build a Guest Memory Layer, not another silo

Create one governed context layer — a CDP or middleware tier sitting above PMS, CRM, POS, messaging, and GRMS — that every agent, human or AI, reads from and writes to. Define a Guest Memory Specification: which fields are remembered (preferences, open issues, stated occasions, accessibility needs the guest chose to share), which are ephemeral, and how long each is kept. Make it a requirement in every AI vendor contract that conversation outcomes are written back to this layer, not locked in the vendor’s own logs.

The foundations are thinner than the ambition. Only 42% of chains have a central guest profile database (CDP) in use and 43% a central data platform. A standardized AI data access layer — MCP or similar, the very thing that lets agents read and write shared context — is in use at 15%, with 52% having no plans. Agent-to-agent communication layers sit at 9% and orchestration at 8%.¹⁸ One small-chain respondent in the study described the goal well: an agentic mesh of “connected AI agents working across departments and sharing data” — which, by another name, is a cure for amnesia.

The report’s own conclusion puts the burden squarely on vendors: as agents take on more operational and guest-facing work, hotel systems must become more open, integrated and AI-ready, with accessible data, interoperability and orchestration.¹⁸ Memory is the clearest test of that claim.

Standards will make this cheaper. This week, Avalora’s VAIA became the first AI assistant validated against the HTNG Express PMS specification, which standardizes access to a limited core of guest, reservation, and room data.²³ That is unglamorous plumbing — and exactly what memory needs. No vendor can economically build a bespoke integration into a hundred PMS environments; a standard interface lets every agent read and write the same core context. Put HTNG Express (or equivalent) support and per-agent data scoping into your RFPs now.

Procurement is about to become policy in at least one market. From 1 November 2026, Singapore hotels can buy AI-enabled Digital Concierge solutions as a pre-approved category under Enterprise Singapore’s EDGE for Productivity scheme, with simpler applications and standardized pricing.²⁷ Pre-approved lists shape markets: vendors on the list win by default. Operators — and the agencies drawing up such lists, in Singapore and in any ASEAN ministry that copies the model — should make write-back to the hotel’s own guest record and a structured handoff packet qualifying criteria, not optional extras. A subsidized concierge that forgets the guest is amnesia at public expense.

Move 4 — Standardize the warm handoff

No escalation without a handoff packet: who the guest is, what they asked, what the AI already tried, what the guest is feeling, and what was promised. Displayed on the agent desktop before the human says hello. The test is simple — the human’s first sentence should prove they already know the story: “Ms Reyes, I can see the airport transfer didn’t show — I’ve got a car coming in ten minutes.”

Wyndham’s numbers turn this from a service argument into a revenue one. About 85% of callers stay with the AI agent; roughly 15% transfer to a person. Those transferred calls are not failures — Wyndham reports a 15% lift in booking conversion and a 16% lift in ADR on them.²⁶ The handoff is where the valuable conversations go, which makes it exactly the wrong place for the guest to start again from scratch. It also gives hoteliers a first public benchmark for the AI Override Rate: a transfer rate around 15%, reported as a success metric rather than an embarrassment.

The industry now has a live laboratory for this. HFTP has named Inntelo AI its official AI partner, with multilingual agents across its website, media and events including HITEC, escalating to HFTP staff when a human is needed.²⁴ Publishing its escalation rate, repeat-yourself rate on handoff, and failure categories over time would give hoteliers something rarer than another vendor demo: production evidence of how well AI-to-human handoff actually carries memory.

Move 5 — Disclose, consent, and show the memory

Every AI agent identifies itself at the first message. Every guest can see what is remembered about them — in the app, on WhatsApp, at the desk — and can correct or delete it. Recognition becomes a visible service rather than a hidden capability. This is where our Human Experience Orchestrator (HXO) principle applies: the AI carries the memory; the human delivers the moment.

Move 6 — Measure the forgetting — on both sides of the desk

If it is not on the dashboard, it is not managed. Add one metric alongside containment and handle time: the Repeat-Yourself Rate (RYR) — the share of conversations, and of AI-to-human handoffs, in which the guest had to re-supply information the hotel already held. Sample transcripts weekly.

Then measure its mirror image: Manual Touches Per Stay (MTPS) — every time an employee re-types guest information, moves it between systems, or copies an email into the PMS across one complete journey, from reservation to post-stay. RYR is the guest repeating themselves; MTPS is the staff repeating the guest. They fall together or not at all. Classify each manual touch as necessary human judgment, system limitation, integration failure, process legacy, or automation opportunity — and drive the last four down.

Give both numbers an owner. Not an “AI champion” who teaches prompts, but an AI Process Owner in Rooms and in Commercial, accountable for the question: which process in my department should change because AI now exists? Pair the metrics with the AI Override Rate and Decision Envelope boundaries from Pertlink’s governance work, so that what the AI may remember, and what it may do with that memory, are both explicit.

7.  A 90-day action plan

8.  Questions every owner should ask this quarter

1. If a guest messages us on WhatsApp tonight after booking on our website yesterday, does the agent know?

2. When the AI escalates to a human, what exactly does the human see — and before or after they say hello?

3. Which vendor owns the transcript of our guests’ conversations, and can we take it with us?

4. Do our AI agents disclose that they are AI in the first message — and can we prove it?

5. When a guest tells our AI about an allergy, how many people re-type it before it reaches the kitchen?

6. Which AI agent can read which guest fields, with what authority, and for how long?

7. What are our Repeat-Yourself Rate and Manual Touches Per Stay, and who owns them?

8. Could a guest see, correct, or delete what we remember about them today?

Closing

The first wave of hotel AI was judged on whether it could answer. The next will be judged on whether it remembers. Guests will forgive a slow response from someone who knows them; the Philippine data shows this in plain numbers. They will not forgive a fast answer from a system that treats them as a stranger on their fifth stay.

Remembered context also unlocks what comes next. Airbnb’s Brian Chesky argued this week that travel will not collapse into a single universal chatbot; interfaces will adapt to the traveler’s intent.²⁵ A hotel can only assemble the right screen — late-arrival, family, early-flight — if it remembers who is asking. Memory is the input to every adaptive experience that follows.

The good news: the fix is not a smarter model. It is the unglamorous work hotels have deferred for twenty years — clean identity, joined-up data, disciplined handoffs, honest disclosure. AI has removed the option of deferring it any longer. Hospitality no longer has an AI adoption problem; it has an absorption problem — and AI amnesia is what absorption failure sounds like to a guest.

The intelligence may be artificial. But the experience is human.

Made with the help of various SI [AI] tools, but always with a HITL

References

1. FutureIoT Editors, "Seven in 10 APAC consumers abandon AI support over memory failures," FutureIoT, 28 September 2026. https://futureiot.tech/

2. "AI amnesia is driving APAC consumers away from brands, study finds," FutureCIO, September 2026. https://futurecio.tech/ai-amnesia-is-driving-apac-consumers-away-from-brands-study-finds

3. TNGlobal, "Twilio says 70 percent of APAC consumers abandon AI service interactions that lack context," TechNode Global, 24 September 2026. https://technode.global/2026/09/24/twilio-apac-ai-amnesia-consumer-trust/

4. "AI memory gaps are damaging customer trust in Asia-Pacific: Twilio," Techgoondu, 25 September 2026. https://www.techgoondu.com/2026/09/25/ai-memory-gaps-are-damaging-customer-trust-in-asia-pacific-twilio/

5. "AI amnesia erodes brand trust as AI takes half of Philippine customer service – study," Context.ph, 22 September 2026. https://context.ph/2026/09/22/ai-amnesia-erodes-brand-trust-as-ai-takes-half-of-philippine-customer-service-study/

6. "‘AI amnesia’ frustrates Filipino customers, survey finds," NewsBytes.PH, 28 September 2026. https://newsbytes.ph/2026/09/28/ai-amnesia-frustrates-filipino-customers-survey-finds/

7. "AI Takes Over Customer Service, but Indian Consumers Still Struggle to Be Recognized – Twilio," Digital Terminal, 21 September 2026. https://digitalterminal.in/trending/ai-takes-over-customer-service-but-indian-consumers-still-struggle-to-be-recognised-twilio

8. Twilio, "The 2026 CDP Shift: Why AI Agents Are Only as Smart as Your Data Infrastructure," Twilio blog, September 2026. https://www.twilio.com/en-us/blog/insights/why-AI-agents-are-only-as-smart-as-your-data-infrastructure

9. Twilio, "The Customer Data Platform Report 2026." https://www.twilio.com/en-us/report/the-cdp-report-2026

10. "Twilio Takes On ‘Agent Amnesia’ So Customers Never Have to Repeat Themselves Again," The Letter Two, 6 May 2026. https://thelettertwo.com/2026/05/06/twilio-agent-amnesia-customer-engagement/

11. Ireckonu, "Hotel Groups Miss Out 20% of Guests Due to Data Silos, New Ireckonu Data Shows," 18 February 2026. https://ireckonu.com/article/hotel-groups-miss-out-20-of-guests-due-to-data-silos-new-ireckonu-data-shows/

12. Leadheed, "Hotel PMS vs Hotel CRM: Key Differences & Integration Benefits," 2026. https://leadheed.com/blog/hotel-pms-vs-hotel-crm/

13. DailyPoint, "Now is the Moment for a Revival in Hotel Technology!" (cloud PMS migration and loss of stay history). https://www.dailypoint.com/newsroom/blog/restart

14. Stay Insiders, "The hotel data integration imperative: why collapsing PMS, POS and labor silos is the real AI prerequisite," July 2026. https://www.stayinsiders.com/the-hotel-data-integration-imperative-why-collapsing-pms-pos-and-labor-silos-is-the-real-ai-prerequisite

15. Cloud Security Alliance, "EU AI Act Article 50: Transparency Obligations Take Effect," July 2026. https://labs.cloudsecurityalliance.org/research/csa-research-note-eu-ai-act-article-50-transparency-20260729/

16. Senyo Labs, "The EU AI Act Kicked In in August 2026: A Transparency Guide for Chatbots and AI Content," September 2026. https://www.senyolabs.com/en/blog/ab-yapay-zeka-yasasi-madde-50-seffaflik-2026

17. Republic Act No. 10173, Data Privacy Act of 2012 (Philippines); National Privacy Commission. https://privacy.gov.ph/data-privacy-act/

18. h2c GmbH, "The AI Opportunity," h2c Hotel Chain Tech Report 2026 (public extract). 122 responses from 113 hotel chains (101 surveys, 21 executive interviews); fieldwork May–July 2026. Page references in text. https://h2c.de/work/research/2026-AI-opportunity-study

19. "AI Use Reaches Most Hotel Groups as Enterprise Readiness Lags," Hotel News Resource, October 2026. https://www.hotelnewsresource.com/article143157.html

20. "Hotels still stuck on manual work despite increased AI adoption," PhocusWire, 1 October 2026. https://www.phocuswire.com/news/technology/hotels-still-stuck-on-manual-work-despite-increased-ai-adoption

21. NYU SPS, "More Than 50% of Hotels Use AI, but Under 10% See Real Impact, Finds State of Distribution 2026 Report from RateGain, NYU SPS and HEDNA," 2026. https://www.sps.nyu.edu/about/news-and-ideas/articles/press-releases/2026/more-than-50-of-hotels-use-ai-but-under-10-see-real-impact-rategain-nyu-sps-hedna.html

22. "The P&L Doesn’t Lie: Hotels See More AI Costs Than Bottom-Line Gains," Skift, 1 October 2026. https://skift.com/2026/10/01/the-pl-doesnt-lie-hotels-see-more-ai-costs-than-bottom-line-gains/

23. "Avalora’s VAIA is first AI assistant to complete HTNG Express PMS integration," Hospitality Net, 1 October 2026. https://www.hospitalitynet.org/news/4134669/avaloras-vaia-is-first-ai-assistant-to-complete-htng-express-pms-integration

24. "HFTP names Inntelo AI as AI partner, deploying AI agents for members, media and events," Hospitality Net, 1 October 2026. https://www.hospitalitynet.org/news/4134683/hftp-names-inntelo-ai-as-ai-partner-deploying-ai-agents-for-members-media-and-events

25. "Brian Chesky interview: AI agents need their own operating system," TechCrunch, 1 October 2026. https://techcrunch.com/2026/10/01/brian-chesky-interview-ai-agents-need-their-own-operating-system/

26. "Hotels, AI and building for long-term economic value," Skift, 25 September 2026 (Wyndham CEO Geoff Ballotti, Skift Global Forum). https://skift.com/2026/09/25/hotels-ai-and-building-for-long-term-economic-value/

27. "Singapore government expands AI support for hotel digitalization," OpenGov Asia, September 2026 (Enterprise Singapore EDGE for Productivity, Digital Concierge category). https://opengovasia.com/singapore-government-expands-ai-support-for-hotel-digitalisation/

28. Agoda, "AI Developer Report 2026," PR Newswire, 24 September 2026. https://www.prnewswire.com/apac/news-releases/agoda-releases-ai-developer-report-2026-agentic-ai-adoption-outpaces-enterprise-readiness-across-southeast-asia-and-india-302888472.html

29. "What hotel CEOs are actually doing with AI, from voice agents to a digital Hyatt CEO," Skift, 28 September 2026. https://skift.com/2026/09/28/what-hotel-ceos-are-actually-doing-with-ai-from-voice-agents-to-a-digital-hyatt-ceo/