01 · Talent & ownership
Build a team — or adopt a platform
AI talent is scarce, expensive, and rarely fluent in hospitality operations. The question is whether your edge comes from building models and infrastructure, or from configuring a hospitality-native platform to your brand, your PMS, and your service standards.
Path A — Build in-house
Hire ML engineers and stand up your own stack
Full control, but 12–24 months to production, ongoing platform maintenance, and a roadmap that competes with every other IT priority. You inherit the cost of keeping pace with a model landscape that resets every quarter.
12–24 months to value
Path B — Adopt a hospitality platform
Configure a native platform to your estate
Production in weeks, not years. Your team owns prompts, business rules and integrations — not GPU clusters and model evals. Vendor R&D keeps you current automatically as foundation models improve.
Weeks to value
Our take
Build where you are differentiated — your guest experience, your data, your brand voice — and buy the orchestration layer. The winning teams spend their scarce engineering hours on integrations and workflows, not on rebuilding infrastructure that a platform commoditizes.
Career signal: "We built our own LLM stack" ages badly. "We were live in production across three brands in one quarter" compounds.
02 · Architecture
Bolt AI onto your PMS — or run an AI-native layer
Your PMS vendor will sell you an AI add-on. It will be convenient — and it will be locked to one system, one data model, and one vendor's release cadence. AI that orchestrates across your whole stack is a different animal from a feature inside a single tool.
Path A — PMS bolt-on
Enable the AI feature inside your PMS
Fast to switch on, but trapped behind one vendor's API surface. It cannot reach your CRM, housekeeping, F&B, or messaging tools — so it answers questions but rarely completes the work that spans systems.
Single-system ceiling
Path B — AI-native orchestration
A layer that sits above every system
One adaptive brain connected to PMS, CRM, housekeeping and channels via open APIs. It reasons across systems, executes end-to-end, and is not held hostage to any single vendor's roadmap.
Estate-wide execution
Our take
Treat AI as an orchestration layer, not a PMS feature. Your systems will change over the next five years; the intelligence layer should be the constant that outlives any one of them — and that means it has to be vendor-neutral by design.
Career signal: Bolt-ons make you a tenant in someone else's roadmap. An orchestration layer makes you the landlord of your own guest experience.
03 · Velocity
Pilot forever — or deploy to production
The most expensive AI program is the one that never leaves the lab. Endless proofs-of-concept feel safe, but they burn budget and credibility while competitors learn from real guests in production.
Path A — Perpetual pilot
Keep testing until it is "perfect"
Demos impress the board, but nothing touches a real guest. You accumulate slideware and sunk cost, never the operational learning that only comes from live traffic — and the window closes around you.
No compounding learning
Path B — Scoped production launch
Go live on a narrow, high-volume use case
Start where volume is high and risk is contained — pre-arrival messaging, FAQs, after-hours coverage. Measure resolution and CSAT, then expand. Real data beats a perfect plan.
Learning in weeks
Our take
Set a production date before you start, and pick a use case you can ship to real guests within a quarter. Constrain scope, not ambition. A live, narrow deployment teaches you more in a month than a pilot does in a year.
Career signal: Boards forgive a contained, measured launch. They do not forgive a two-year program with nothing in production to show for it.
04 · Capability
A chatbot that answers — or an agent that acts
A generic chatbot can describe your cancellation policy. It cannot actually move the reservation, trigger housekeeping, or apply the loyalty credit. The gap between answering and executing is the gap between a novelty and an operational asset.
Path A — Answer-only bot
Conversational FAQ over your content
Helpful for deflection, but every real request — a change, an upsell, a recovery — still lands on a human. Guests feel the hand-off, and your team still does the work.
Deflects, doesn't resolve
Path B — Agents that execute
Skills that take action across systems
The agent understands intent and completes the task: modifies the booking, dispatches the team, sends the upsell, logs the CRM note — with the guardrails you set. Requests become resolved actions, not tickets.
Resolves end-to-end
Our take
Evaluate vendors on what the agent can do, not what it can say. Ask for a live demo where the AI executes a multi-step task against a real system. If it can only talk, it is a cost center wearing an AI badge.
Career signal: "Our AI deflects 40% of messages" is fine. "Our AI resolves them end-to-end" is the number that changes the P&L.
05 · Reach
One channel — or unified across channels & languages
Guests do not think in channels. They start on WhatsApp, switch to email, then call. If each channel has its own bot with its own memory, the guest repeats themselves and your brand feels fragmented — in every market and every language.
Path A — Per-channel point tools
A separate bot for each touchpoint
Web chat here, a phone IVR there, an email auto-responder somewhere else. No shared context, inconsistent answers, and a separate integration bill for every channel and language you add.
Fragmented & siloed
Path B — One brain, every channel
Unified across messaging, voice & email
A single agent with shared memory across WhatsApp, web, email and voice, fluent in every guest language. Context follows the guest; staff see one unified inbox instead of five disconnected tools.
One context, all languages
Our take
Buy the channel-agnostic brain, not a channel-specific tool. Multilingual, cross-channel continuity is now table stakes for international guests — and it is far cheaper to start unified than to stitch silos together later.
Career signal: Every channel-specific bot you buy today is an integration you will rip out tomorrow. Unify once.
06 · Data & knowledge
Siloed data — or an AI-ready, open knowledge base
AI is only as good as the knowledge it can reach. If your policies, rates, SOPs and property facts live in PDFs, inboxes and someone's head, no model can serve them reliably — and you will rebuild that knowledge for every tool you buy.
Path A — Knowledge stays siloed
Each tool ingests its own copy
Every vendor re-imports your content into its own black box. Updates drift out of sync, answers contradict each other, and you can never fully audit what the AI was told — or take it with you.
Lock-in & drift
Path B — One AI-ready knowledge layer
Structured, governed, open via MCP
A single source of truth, structured for the LLM era and exposed through open standards so any model or agent can use it. Update once, serve everywhere, and keep ownership of your own knowledge.
Portable & auditable
Our take
Invest in an AI-ready knowledge base as core infrastructure, and insist on open connectivity (e.g. MCP). Your knowledge is a strategic asset — it should never be trapped inside a single vendor's proprietary index.
Career signal: The teams that win in 2027 are the ones that made their knowledge portable in 2026, while everyone else was re-uploading PDFs.
07 · Trust & governance
Govern later — or govern from day one
The fastest way to kill an AI program is one bad guest-facing answer that makes the news. Guardrails, escalation paths and audit trails are not a phase-two nicety — they are what lets you move fast without betting the brand.
Path A — Ship now, govern later
Add controls after something breaks
Speed today, exposure tomorrow. Without limits, escalation and logging, a single hallucination or data-handling mistake can trigger a brand, legal or regulatory incident you cannot undo.
Unbounded risk
Path B — Guardrails from day one
Controls, escalation & audit built in
Define what the agent may do autonomously, where it must hand off to a human, and log every action. Governance becomes the accelerator that lets you expand scope with confidence, not a brake.
Confident scaling
Our take
Choose a platform where business rules, human-in-the-loop and full auditability are native, not bolted on. The ability to say exactly what the AI did, and why, is what keeps the program alive long enough to compound.
Career signal: Governance is not the thing that slows you down. It is the thing that lets you keep your job when something inevitably goes sideways.
