Every hotel tech roadmap now has an AI line on it. A guest messaging vendor wants AI replies, a PMS wants an assistant for the front desk, an RMS wants natural-language explanations of its pricing. The question founders bring to us is rarely whether to add AI. It is whether to build the feature themselves or integrate someone else's.

This post gives you a decision tree for that choice. Work through the questions in order; each one sends you to a branch. It is written for founders and product and revenue leaders at hotel tech companies.

The three real options

"Build or buy" hides a middle option that most vendors end up taking. Be precise about which one you mean.

Why this decision matters more than it looks

Two pieces of evidence should shape your default. First, MIT NANDA research reported by Fortune found that about 5% of generative AI pilots achieve rapid revenue acceleration, and that purchasing AI tools from specialised vendors and building partnerships succeeded about 67% of the time, while internal builds succeeded only one-third as often. That study looks at organisations adopting AI, not at software vendors, but the lesson carries: building is the harder path.

Second, your buyers are not asking for AI first. HEDNA's 2025 State of Distribution report found hotels are prioritising tech consolidation over AI investments and that AI tools rank lowest in investment priority across all hotel segments. In the Hotel Yearbook's 2025 tech survey, ROI was the top priority when evaluating new technology, chosen by 53.48% of respondents to that question, with AI and automation described as secondary concerns. An AI feature has to earn its place by fitting the workflow and proving a return, however it is built.

The decision tree

Question 1: Is the AI output the reason customers buy you?

• If yes, and the feature depends on data only you hold (for example years of pricing decisions across thousands of properties), go to Question 2.

• If no, and the AI is a convenience on top of your core job, go to Question 3.

Question 2: Do you have the team to own a model for years?

• If yes, meaning engineers who can evaluate, monitor and retrain models and a budget for it, build. Start on a commercial model API and only move to your own model when you can show it beats the API on your data.

• If no, build on a model API and hire for the capability before you promise anything proprietary in sales.

Question 3: Does the AI need to act inside your core workflow?

• If yes, for example drafting a reply inside your messaging inbox or explaining a rate inside your RMS screen, go to Question 4.

• If no, because the job sits next to your product (a voice agent for reservations calls when you sell a PMS), go to Question 5.

Question 4: Do the unit economics work at your price?

Model costs are now published and low per request. On Anthropic's pricing page, Sonnet 5.5 is listed at $2 per million input tokens and $10 per million output tokens, and Haiku 5.5 at $0.10 and $0.50 for prompts up to 100,000 tokens. At the Sonnet rate, a request with 2,000 input tokens and 300 output tokens costs $0.007. Multiply by your expected requests per property per month and compare with what you can charge.

• If the cost per property is small against the price, build on a model API. The real costs are engineering, evaluation and support, so budget for those.

• If usage is heavy and unpredictable, build on a model API but price the feature by usage or as a paid tier, not inside the base fee. See how to price a second product.

Question 5: Is there a specialist that already does this well and connects to you?

• If yes, integrate. Agree the commercial model (referral fee, resale or co-sell), make the integration two-way and certified, and put it in your marketplace.

• If no, ask whether customers will pay for it at all. If they will, build on a model API as a narrow feature. If you cannot find that evidence, do not build it yet.

The tree on one page

A public example of the integrate path

Apaleo shows what the integrate path looks like at platform level. In September 2025 it launched an MCP server, described as the first in property management, to give AI agents a standardised way to read hotel data and take actions without custom integrations for every vendor and use case. It built on Apaleo's Agent Hub, which the company calls the first AI agent marketplace for hospitality. Apaleo did not try to build every agent itself; it built the connection layer and let others build the agents. For a PMS with an open platform strategy, that is a clear integrate decision.

What changes in your go-to-market

• If you build: you must be able to show results on the customer's own data during the sales cycle, and your positioning should say what the AI does, not that it exists. Our post on AI-native versus AI-enhanced positioning covers the language.

• If you integrate: your sales team sells the outcome of the combined product, and partner enablement becomes part of your revenue motion, much like the choice covered in build or partner on integrations.

• Either way: lead with integration and ROI, because that is what hotel buyers rank first. Cloudbeds cites h2c research that 70% of hoteliers name integration with existing systems as a top priority in AI investments. See also what the h2c study means for selling AI to chains and how hotels get their AI.

AI decisions sit within the product and positioning stages of the hotel tech GTM strategy guide, and they interact with the pressure hotels feel to cut vendors: an integrated AI partner is one more supplier from the hotel's point of view unless you package it as part of your product.

The short version

Build only when the AI output is why customers buy you and you can own it for years. Build on a model API when the feature acts inside your core workflow and the unit economics work. Integrate when the job sits next to your product and a good specialist already connects to you. If none of those holds, wait.

If you are unsure whether your AI roadmap is helping deals close or just adding demo slides, the free leak audit will show where buyers drop out and whether AI is part of the reason.

Frequently Asked Questions