1 hour ago

The Baker's Dozen Checks In

What McKinsey's Technology Trends Outlook 2025 Means for Hospitality's Next Eighteen Months

A McKinsey 2025 Technology Trends Outlook analysis filtered for hospitality, arguing that agentic AI is reshaping the guest booking layer and that governance, not model capability, is now the key constraint on adoption.

By Terence Ronson, Founder & Managing Director, Pertlink Limited

Photo by Pertlink Limited

FRAMING

Thirteen Trends, One Front Desk

McKinsey calls it a baker's dozen — thirteen frontier technology trends, scored on innovation, interest, investment, and adoption — in its fifth annual Technology Trends Outlook. Hospitality doesn't get its own chapter. It never does. But read the report the way an owner-operator has to read everything — for what it does to tomorrow's P&L, not tomorrow's headline — and eleven of the thirteen trends land somewhere on your property before your next capex cycle closes.

This Viewpoint sits McKinsey's 2025 outlook next to what Pertlink has been arguing all year: that the guest decision layer is migrating into AI-controlled channels, that the unit economics of AI service now run through a Token Cost Per Guest, and that governance — not model capability — is the binding constraint on how fast any of this can responsibly scale. The McKinsey data doesn't just support that argument. In places, it dates it.

Three numbers set the scene. Equity investment in agentic AI grew from a standing start to $1.1 billion in 2024. Job postings referencing agentic AI roles rose by 985 percent over the same two-year period. And 78 percent of organizations McKinsey surveyed are already using AI in at least one business function — while only 1 percent describe their deployment as fully mature. That gap, between adoption and maturity, is where the next eighteen months of hospitality technology strategy will actually be fought.

TREND 01 — AGENTIC AI

The Decision Layer Keeps Moving

McKinsey's framing of agentic AI is unambiguous: these are systems built to act, not merely to converse — booking, filling in forms, executing multi-step workflows, coordinating with other agents through emerging standards like Anthropic's Model Context Protocol and Google's Agent2Agent protocol. The report's own case studies read like a distribution strategy memo — a general-purpose agent that books flights and restaurants unprompted, a research agent that shortlists and synthesizes before a human ever opens a tab.

Regular readers of Pertlink Viewpoint will recognize this as exactly the terrain covered in “The AI's Have It,” “The Layer That Owns the Guest,” and “The Terminal Is the New Homepage”: the hospitality decision layer — search, shortlist, compare, book — is migrating out of the OTA interface and into the agent's reasoning trace, whether that agent lives in a chat window or, as Shanghai's tourist terminals show, on a screen in the physical city. McKinsey's contribution is to quantify the momentum behind that migration and to name the plumbing (MCP, A2A) that will determine which properties are surfaced to the agent and which are quietly skipped.

McKinsey also flags something owner-operators should sit with: agentic AI's adoption score currently sits at 2 out of 5 — small-scale experimentation, not scaled deployment. Most of what agents are doing today is closer to a prototype than a production channel. The strategic window is therefore now, while the standards are still being written and before the large aggregators finish encoding their own preferences into the agent layer. AEO — agent engine optimization — and its cousin, GEO, are no longer forward-looking concepts. They are the SEO of 2027, and the properties that wait for the concept to mature before acting will be optimizing for a channel that has already settled without them.

TREND 01 — AGENTIC AI, CONTINUED

The Agent Books It. Who Keeps the Guest?

Since this Viewpoint's agentic AI section was first drafted, the picture has moved from theoretical to operational — and it happened inside a single news cycle. Airlines are now running the exact architecture McKinsey describes in production. Amadeus is deploying agent teams across network planning, marketing, aircraft turnaround, and disruption management. Southwest is using agentic tooling to rebuild its crew-scheduling environment. Most tellingly, Alaska Airlines and Volantio have agents that identify likely-oversold flights, evaluate which passengers fit a set of business rules, construct a compensation offer, and propose the change directly — a sequence that runs from observation to commercial action with no human step in between.

That sequence — observe a condition, evaluate the customer, construct an offer, act — maps onto hotel oversell and upsell management almost without translation. A forecast identifies an oversold night; an agent identifies flexible reservations, evaluates loyalty status and arrival profile, builds an alternative offer, contacts the guest, and updates the PMS. Before a vendor arrives asking a property to hand over that authority, this firm's recommendation is to define it first, one workflow at a time, as a set of Agentic Decision Rights:

OBSERVE — which conditions may the agent monitor?
DECIDE — which decisions may it make unassisted?
OFFER — what commercial proposition may it construct?
COMMIT — what may it transact without approval?
ESCALATE — at what point must a human take over?

Consumer appetite for delegation is outpacing most operators' expectations. Amadeus's APAC research finds that 40 percent of travelers are already open to letting AI book on their behalf, with similar shares welcoming AI-generated destination ideas. A separate Amadeus Hospitality study of 500 hoteliers and 6,000 travelers found AI most useful during planning and discovery, and — the number worth sitting with — 69 percent of travelers now consider an AI summary sufficient to decide without further investigation, rising to 87 percent in India and 86 percent in China.

That shortlisting-before-search dynamic revives, in sharper form, the guest-ownership problem that hospitality has known for two decades of OTA distribution. Guest contact data supplied by OTAs has always been partial and time-limited by design. AI agents risk reproducing that architecture at a more consequential layer, because the agent increasingly holds the traveler's intent before the property's own search or booking engine ever sees the guest. Three things worth keeping distinct on any KPI sheet: guest identity — knowing who is staying; guest contactability — being able to reach them; and guest relationship — the guest having knowingly consented to an ongoing connection with the property. An intermediary, human or agentic, can hand over the first while withholding the second and third.

There is a parallel infrastructure shift worth flagging for the same reason. Eviivo's new distribution partnership with Dida exposes independent-hotel inventory to roughly 40,000 APAC distributors — and Dida has paired that with an MCP-native AI booking gateway designed to sit inside third-party agent applications. Distribution is becoming infrastructure a property cannot see, not a channel it can name. The practical response is to track distribution provenance for every reservation — guest interface, demand source, distributor, inventory source, and commission — so management doesn't discover after the fact that three apparently separate channels were all fed by the same intermediary.

The commercial response to all of this is a First-Party Conversion Rate: of guests arriving through any intermediary, agentic or otherwise, how many leave with a permissioned direct relationship — loyalty enrolment, consented contact, a direct-booking account, opted-in post-stay communication. Track the full funnel as discovered, selected, booked, known, and retained, and as the strategic point sharpens: hotels cannot own every discovery interface and will not stop guests from delegating decisions to agents. What they do own, and what no intermediary can substitute for, is the stay itself. Converting that stay into a known, retained relationship — lawfully and transparently — is the whole game from here.

TREND 02 — ARTIFICIAL INTELLIGENCE

Ninety-Two Percent Are Investing. One Percent Are Ready.

The headline AI numbers in McKinsey's report will be familiar to anyone who has sat through a hotel technology steering committee this year: 92 percent of executives plan to increase AI investment over the next three years, yet the same research finds only 1 percent of organizations describe their AI deployment as fully mature. McKinsey's own explanation is structural, not technological — the gap is explained by how long it takes organizations to adapt processes, build complementary capabilities, and reskill a workforce, not by any shortfall in what the models themselves can do.

That is precisely the argument behind TCPG, the Token Cost Per Guest framework this firm has been developing through 2026. McKinsey's report supplies the macro tailwind: inference costs are falling sharply, and a new generation of smaller, domain-specific models is putting AI within reach of organizations that could never have justified a frontier-model bill. For a mid-market hotel group, that is the difference between AI remaining a corporate pilot and AI becoming a line item every property can budget for. The 99 percent gap between investment intent and deployment maturity is not a technology problem. It is an operating-model problem — precisely the ground covered in “Flat Is the New Front Desk,” “Every Hotel Could Now Be a Software Company,” and “The Camry Guest Experience”: unglamorous, reliable, correctly costed AI beats an impressive pilot that never scales past the flagship property.

There is a caution built into that same argument, one this firm made directly ahead of HITEC 2026 in “Going to #HITEC? Ask about Tokenmaxxing!” Falling inference costs are not, by themselves, good news. They are how the bill quietly grows, as the technology sector's own tokenmaxxing hangover — leaderboards, consumption targets, bonuses tied to token burn, then the wave of retreats that followed — demonstrated within a single half-year. TCPG exists precisely to stop hospitality from importing that mistake: it prices AI against guests served, not tokens consumed, so cheaper inference translates into better unit economics rather than into busier, costlier automation for its own sake.

McKinsey's other observation worth flagging to any GM weighing a build-versus-buy decision: multimodal AI — models that read text, images, video, and audio together — has moved from a research curiosity to a production tool within three years of gen AI's commercial availability. For hospitality, that is the technical foundation for everything from AI-assisted property inspections to voice- and camera-enabled concierge tools. The capability is no longer the constraint. The organization around it still is.

That 78-percent-to-1-percent gap also has a name at property level, and it isn't technical. A large hotel-brand rollout of the same AI-enabled guest-engagement platform recently produced highly engaged properties generating over $60,000 in incremental annual revenue apiece — and passive properties running the identical software generating close to nothing. Same vendor, same contract, same code. The difference was adoption, not capability, which is worth holding as three separate milestones rather than one: go-live, the system works; adoption, staff actually use it correctly; value, the property produces a measurable result. Most technology projects celebrate the first and assume the other two will follow. A 90-day value review after go-live — feature utilization, workflow compliance, manual workarounds, and the revenue or cost outcome — is the cheapest insurance against discovering, a year later, exactly how much purchased capability went unused.

TREND 07 — DIGITAL TRUST AND CYBERSECURITY

Trust Is the New Uptime

McKinsey's cross-cutting theme for 2025 is blunt: as technologies become more powerful and more personal, trust becomes the gatekeeper to adoption, not a compliance afterthought. The report tracks digital trust and cybersecurity investment at $77.8 billion in 2024 — modest growth of 7 percent year on year, but from an already-large base, and rising in lockstep with agentic AI specifically because autonomous systems that can take real-world actions raise the stakes on every governance question a hotel group has been able to defer until now.

This is the ground this firm has been walking with the “Who Controls the Controls?” series (Parts One and Two) and “The Invisible Ink” watermarking paper — and McKinsey's data gives that argument a harder edge. An agent that can book, cancel, or reprice on a guest's behalf can also be socially engineered, prompt-injected, or simply wrong at a moment when there is no human in the loop to catch it. McKinsey's own big question for the trend — how to balance AI autonomy against human oversight — is not a philosophical one for a hotel group; it is a legal and reputational exposure question, and it belongs on the same governance agenda as PCI compliance and data residency, not bolted onto IT's existing security review as an afterthought.

TRENDS 06 & 09 — IMMERSIVE REALITY AND ROBOTICS

The Robots Are Already on Shift

Two further trends deserve a shorter mention because their hospitality relevance is more operational than strategic. McKinsey documents robotics expanding well beyond manufacturing into services — cobots now working food-preparation lines, robots delivering within hospital corridors — and pegs the addressable robotics opportunity at close to $900 billion by 2040. Humanoid platforms remain early, but purpose-built service robots for housekeeping runners, room-service delivery, and back-of-house logistics are past the novelty stage that the report describes for most other trends.

Immersive reality's hospitality case is narrower but sharper: AR-guided maintenance and training, the kind McKinsey illustrates with technicians repairing wind turbines through augmented-reality goggles, translates directly into engineering and housekeeping onboarding for properties that are perennially short of experienced trainers. Neither trend changes the guest-facing experience as fundamentally as agentic AI or the broader AI trend does. Both belong in the Human Experience Orchestrator [HXO] conversation set out in “The Lobby Boy v2031” — where technology's job is to give staff back the time and context to be better hosts, not to stand in for them at the point of guest contact.

CLOSING

What This Means for You

Read across the baker's dozen, and McKinsey's own cross-cutting themes for 2025 — the rise of autonomous systems, new human-machine collaboration models, scaling challenges, and responsible-innovation imperatives — map almost exactly onto the fronts hospitality owner-operators need to be working simultaneously: where you show up in agent-mediated search and booking, who keeps the guest relationship once an agent has made the introduction, how you cost and govern AI at the level of the individual guest interaction, how you build trust and oversight into systems that can now act rather than merely suggest, and where automation genuinely frees your people to do the parts of the job only a person can do.

None of the thirteen trends require a hotel group to become a technology company. They do require treating 2026 and 2027 as the years when the agent layer, the cost model, and the governance framework all get set — by someone. The properties that show up now, with a defensible AEO presence, a costed AI operating model, and a governance framework that can survive an incident, will be the ones the agents recommend by default in 2028. The rest will be negotiating for visibility inside a channel someone else already built.