Every Hotel Could Now Be a Software Company

When building software stops being the hard part, the edge moves to judgement — and hospitality is next in line. What the next 12–24 months ask of owners, GMs and commercial teams

Terence Ronson, Pertlink Limited

A McKinsey conversation is doing the rounds this month. Snowflake’s chief executive, Sridhar Ramaswamy, sits with McKinsey’s North America chair and makes an argument that, at first, sounds like a problem for the technology sector. AI, he says, has collapsed the cost of building software. Almost anyone can now describe what they want in plain English and watch the application appear. So the scarce thing — the thing that decides who wins — is no longer code. It is judgment: how quickly an organization can learn, adapt, and move its people to where the value now lies.

Read that again with a hotel in mind. This is not really a software company story. It is an every-company story. And hospitality — an industry that has spent three decades buying its technology rather than building it — is directly in its path.

For thirty years, we bought software. We did not build it. That distinction is now dissolving — and with it, a comfortable set of assumptions about who holds the advantage.

The constraint was never the code.

Ask a general manager what has held back their property’s technology, and you will rarely hear “the PMS.” You will hear about data trapped in silos, reports nobody reads, revenue decisions made on gut because the number arrived too late, and guest complaints that three systems knew about but none surfaced. The constraint was never the code. It was the operating judgment to turn what the building already knows into something a guest can feel.

Ramaswamy calls the underlying shift the industrialization of intelligence — packing reasoning, not just computation, into tools anyone can summon. For us, it lands squarely in the gap we have always struggled to close: the distance between a hotel’s data and its decision. AI does not remove the need for judgment. It removes most of the excuses for not exercising.

When anyone can build

The most quietly radical line in the interview is about disruption. Software, Ramaswamy notes, has long behaved like a cottage industry: once a population adopts a system, switching feels unthinkable, so incumbents coast. That description fits hospitality technology almost perfectly. Switching costs, integration dread, and the fear of a failed migration have protected vendors for years, even when the product still didn't earn its keep.

Now, a commercial director with no engineering background can stand up a working dashboard by describing it. A revenue team can prototype the report, but has always wanted to, and never has. None of this makes your PMS obsolete tomorrow. But it changes the question in the room. Not just “which system do we buy?” but “what do we now build for ourselves — thinly, cheaply, around the edges of what we buy?”

This is the build-versus-buy calculus inverting in slow motion. Snowflake wrote its own internal support system in six weeks and emptied its queues. A hotel group will not rebuild a property-management platform in six weeks, and should not try. But the guest-messaging layer, the internal knowledge agent, the bespoke arrivals dashboard, the one report that finally ties channel cost to guest value — those are now within reach of a capable team. The Hybrid Hospitality Framework was always about choosing the right blend of human and machine. It now extends to a second choice: what you license, and what you assemble yourself.

The vendor moat was never the software. It was the switching cost. AI is draining the moat while everyone is still admiring the castle.

Not every token has the same value.

Buried in the exchange is a remark I have been making for the better part of a year. As the two discuss consumption-based pricing — where you pay for what you use rather than a flat fee — one of them observes, almost in passing, that “not every token has the same value.” Correct. And that single sentence is the whole argument behind TCPG — Token Cost Per Guest.

The large model providers already price by consumption. That model is now arriving in hospitality contracts dressed as innovation, and it carries a tail risk the industry has not yet metabolized: costs that scale with usage rather than rooms. A token spent resolving a billing query at 2 am has a different value from a token spent auto-generating marketing copy nobody asked for. If you cannot see the difference, you cannot manage it — and you will discover the bill at the worst possible moment.

This is why TCPG belongs in the operator’s vocabulary, and — I will keep saying it — ultimately in USALI. Total AI token cost divided by the number of guests served. A simple ratio that turns an invisible, elastic expense into a line a controller can defend, and a GM can act on. Expect the providers who win in our sector to copy what the sharper coding tools already do: pair a per-seat fee with a consumption fee, then add per-person and per-account ceilings so one enthusiastic user cannot run up a runaway bill. Ask for those guardrails by name.

The conversational lobby

The interface is changing shape. The bespoke, stylized dashboard — designed once for the masses — is giving way to something more fluid: you ask, it answers; you refine. Ramaswamy’s analogy is traffic. You ask the map what to do, then decide whether to follow it. The tool proposes; the human decides.

In a hotel, this cuts two ways. Guest-facing, the conversational layer finally makes “ask for anything” plausible without a thirty-person contact center behind it. Staff-facing — and this is the underrated half — a duty manager can interrogate the building in plain language instead of hunting through four systems. “Which arrivals tonight have an unresolved complaint from a previous stay?” should be a sentence, not a project. That is AI Literacy in its most practical form: not knowing how the model works, but knowing how to ask.

Rethinking roles, not rotas

Here is the part that owners should read twice, because it is where the money and the morale both live. When Snowflake found it no longer needed a dedicated demo team — every executive could now make their own — it did not announce redundancies. It moved those people into roles the business could not previously fill. The framing was growth, not subtraction: produce more, reach further, and redeploy the freed-up talent.

That is the optimistic path, and it is available to us. The honest footnote is the one Ramaswamy does not dodge: some people struggle to adapt, and that becomes a performance conversation. The leadership task is to make the transition genuinely possible — not to dress up cuts as a transformation. In hospitality, the destination role is one I have called the HXO, the Human Experience Orchestrator: the person who no longer keys data or chases reports, but directs machine output toward the moments that make a stay. The coordinator becomes a conductor.

Literacy spreads; it is not mandated.

The most useful change-management lesson in the conversation is almost an accident. Snowflake’s coding tool was not forced on anyone. It was simply useful, and it spread on its own. No training program, no mandate — just utility, plus a culture in which the data was already open to everyone who needed it.

For a GM facing an AI rollout, that is the playbook in miniature. You do not create literacy by decree. You create it by removing friction, opening the data, choosing one or two uses with obvious payoff, and letting the floor discover the rest. The properties that win over the next two years will not be the ones with the biggest license. They will be the ones where asking the building a question feels as natural as picking up the phone used to.

The next 12–24 months

Predictions are cheap, so here are several framed the way an owner needs them: what to watch and what to do. The split is deliberate. The near term is about discipline and experiments; the midterm is when structural advantage opens up between operators who treat AI as a capability and those who still treat it as a procurement line.

What this means for you

Strip away the technology talk, and the McKinsey argument leaves operators with a short, awkward, useful list.

• The moat moved. Your suppliers’ advantage was switching cost, not brilliance. Re-test every “settled” contract against a question you could not ask a year ago: could we now build a thin version of this ourselves?

• Measure the tokens. Consumption pricing is coming whether you choose it or not. TCPG turns an invisible, elastic cost into a number you can manage — and contract ceilings turn enthusiasm into a budget, not a hostage situation.

• Promote the orchestrators. The roles that endure are the ones that direct machine output toward guest moments. Fund that transition honestly; do not relabel cuts as change.

• Open the data, then get out of the way. Literacy spreads through utility, not mandates. Remove the friction, surface the data, pick the obvious wins, and let the floor teach itself.

• Brief the owners early. The next 12–24 months reward operators who treated this as capability, not procurement. That conversation is far easier to have now, by choice, than later, under pressure.

None of this means hotels become technology companies in costume. It means the cost of building useful software has fallen far enough that the advantage has shifted to something older and more human: judgment, adaptability, and the willingness to change what we do faster than the tools around us change. That has always been the job. AI removed the last good excuse for not doing it well.

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

Terence Ronson · Founder & Managing Director, Pertlink Limited

Source framing: “AI is turning every company into a software company,” McKinsey & Company, June 2026 (a conversation between Snowflake CEO Sridhar Ramaswamy and McKinsey’s Eric Kutcher). This viewpoint is an independent Pertlink interpretation for a hospitality audience; it paraphrases rather than reproducing the original.

Made with the help of various AI tools, but with a HITL.