The Year Operto Became an AI Company

By the fourth quarter of 2024, every hospitality software vendor had an AI strategy. Very few of them had an AI engine.

The gap between strategy and engine is the most expensive sentence in enterprise software. It is the difference between a roadmap deck and a production system, between a feature list and a confidence-gated multi-agent architecture answering guest messages at three in the morning in the guest's own language. Most companies underestimate the gap by a factor of three. A small number cross it; most stall in pilot, where the demo works and the deployment doesn't.

Operto, a 25-year-old hospitality technology company best known for IoT door management, guest experience tooling, and operations software across thousands of properties, sat exactly inside that gap. It had the customer base. It had the operational depth. It did not, in late 2024, have a single line of generative AI in production.

The capability gap

Operto's Connect platform, where property staff handle guest interactions, ran on human triage. Every booking question, every late-checkout request, every broken-thermostat complaint landed in an operator inbox and was answered by a person. Knowledge lived in scattered documents and the heads of long-tenured staff. Translation was a tab open to Google or a polite apology. There was no AI assist, no autopilot, no contextual retrieval, no concept of an agent.

Ancillary revenue from upsells, parking, late checkout, pet fees, early check-in, is worth ten to twenty percent of stay revenue. Operto had no native mechanism for capturing it. The workflow was a form, the form had drop-off, the drop-off was money.

The competitive pressure was the more uncomfortable part. It was coming from AI-native entrants with thinner products but better stories. An operations company without an AI engine was structurally behind a chatbot vendor with a slick demo. That asymmetry triggered the decision.

Three paths

There were three options at the end of 2024, and the obvious two were both wrong. Building internally meant six to nine months of hiring before anything shipped, in a category where six months is two cycles of competitive movement. Buying a chatbot vendor meant inheriting someone else's architecture, roadmap, and view of what AI in hospitality should look like.

The third option was a partner who already had production AI in the segment, who could embed inside the next-generation product as a co-builder rather than a vendor, and who had the depth to design and ship the engine in months rather than years. INTO, a Montreal-based AI company already operating a generative-AI guest-interaction platform across the short-term rental side of the market, fit that description.

The conversation did not start as a partnership. It started, in late 2024, as an acquisition discussion. By mid-December both sides had concluded that a partnership was the better answer. It preserved speed and IP boundaries on both sides, and let work begin in months rather than after a deal. The first work started March 1, 2025.

Engineering taste

The decisions that determine whether an AI deployment ships are almost never the headline ones. The headline decision is which model to use, and it is the least important. The decisions that matter are the ones underneath.

The first was a confidence gate. Every AI response is scored, and the score determines what happens next. Above a threshold, the response sends autonomously. In the middle band, the operator sees a draft they can edit. Below it, the AI stays out of the way. Most AI deployments fail not because the model is bad but because they over-promise on autonomy and under-deliver on judgment about when to stop.

The second was a clean data boundary. Operto owns its operational data. INTO owns the knowledge layer the AI retrieves from. The two systems sync via webhooks; the responsibilities are unambiguous. Ambiguous data ownership is where partnerships die.

The third was the refusal of single-vendor dependency. The engine is model-agnostic; it can route across OpenAI, Google, Groq, Azure, Fireworks, and Huggingface, with fallback logic when one is degraded. None of this is visible to a guest sending a message. All of it is why the product stays up.

What got built

The system is a layered architecture: an engine at the centre, six specialised agents around it, three operator-facing modes. The engine handles intent classification, retrieval, response generation, multi-language inference, and confidence scoring. The operator surface is embedded directly inside Operto Connect. In the first mode, the AI suggests a reply. In the second, the operator drafts and the AI completes against the knowledge base. In the third, the AI sends autonomously when its confidence crosses the threshold.

The agents do the work a single chatbot cannot. The Guest Service Agent handles on-property queries; the Concierge Agent handles off-property recommendations; the Revenue and Upsell Agent posts charges directly to the property folio, turning chat into the actual sales surface rather than a form; the Operations Manager Agent detects tasks in guest conversations and routes work in Connect; the Voice Agent picks up phone calls in real time over an 8 kHz telephony stack with warm transfer to a human; the OpsAssistant Agent translates natural-language requests into deterministic SQL reports. Property management connectors are live in production on Cloudbeds and Mews.

The asymmetry, narrowed

The framing inside Operto has shifted. The company no longer describes itself as a hospitality technology company adding AI features. It describes itself in agentic terms: an Agentic Operating System for accommodations, with intelligent agents underneath the marketing. That positioning would not have been credible at the start of 2025. It is credible now because there is production code behind it.

The asymmetry between an operations veteran with no AI engine and an AI-native entrant with no operations depth was, until recently, one-directional. With the engine in production, it narrows. The IoT footprint, the operations depth, and the customer base are durable advantages. The AI engine closes the missing dimension.

Most strategic transformations are visible the moment they conclude. The interesting ones are the ones already running before anyone notices.