When a city’s data needs a “lifeguard”: What Kelowna’s AI mistake reveals about the real threshold for digital public services

The city of Kelowna in Canada once asked its AI chatbot which beaches had lifeguards. The answer sounded very confident, and very wrong. The bot listed several beaches that did not have lifeguards on duty, and the reason was not complicated: deep in the city’s website sat an outdated PDF, and that document recorded an arrangement that had existed in the past. The file had not been updated in time, but the model treated it as usable fact.

This kind of mistake is worth taking seriously not because it is “funny,” but because it is profoundly urban. It is not an isolated AI hallucination, but a typical symptom of public-sector digital infrastructure: fragmented information, document buildup, aging systems, blurred lines of responsibility, all finally exposed through a seemingly advanced interface. AI did not create the problem here; it merely made the problem impossible to ignore.

In Kelowna CTO James McGregor’s description, the case was “interesting, but also critically wrong.” That sentence could almost serve as a footnote to many city AI projects today. For government agencies, the real challenge is not connecting a chatbot to a website, but ensuring that what it can access is genuinely governed, still valid, and traceable to its source.

The first reality of public-sector AI: not intelligence, but plumbing

Cities begin adopting AI not because they suddenly develop romantic notions about technology itself, but because the time structure of public services has changed. Municipal offices keep fixed hours, but residents do not only ask questions, file complaints, submit applications, and make inquiries during office hours. Information services must respond around the clock, especially in high-frequency scenarios such as permits, transportation, community events, public safety, and facility use.

Kelowna’s chatbot was initially built to answer common service questions and cover after-hours demand. According to public information, the system handles about 180,000 queries a year, around 40% of which require no human intervention. That means it is no longer a mere experiment, but a digital front end that has been preliminarily embedded into the municipal service chain.

But what matters more than that number is what it reveals about the basic logic of city AI applications: AI does not begin with “decision-making,” but with “triage.” Its first job is to reduce service friction, extend response time, and cut down repetitive inquiries. For cities, the value of such applications is not in replacing people, but in allowing scarce human capacity to be concentrated on the tasks that require judgment more urgently.

This also explains why Kelowna’s management team emphasized “small ball” — like a conservative baseball strategy, starting with singles and doubles rather than going straight for home runs.This also explains why Kelowna’s management team emphasizes “small ball” — like a conservative strategy in baseball, getting singles and doubles first rather than going straight for the home run. If a city AI deployment initially tries to automate across departments, at scale, and end to end, the risk does not come from the model lacking capability, but from the city system itself being uneven: some processes are standardized, while others rely on human experience; some data are clean, while others are still buried in old websites, legacy systems, and old files.

Old PDFs are not a joke; they are a visualization of a city’s technical debt

The most valuable part of Kelowna’s story is that it turns “technical debt” from an abstract concept into a concrete scenario. The ColdFusion system left behind 30 years ago, long-unorganized historical documents, the ERP upgrade still underway, and a data governance framework not yet fully established — these are not backend details, but prerequisites for whether AI can safely enter public services.

When many cities talk about AI, they instinctively focus on models, interfaces, and new features; but the difficulty in the public sector is the opposite. The key lies in “what the model sees.” If the data catalog is unclear, the document lifecycle is not managed, and the responsible departments are not defined, then AI will only accelerate historical errors into real-time errors.

That is also why more and more cities are shifting their digitalization path from “front-end innovation” to “foundation rebuilding.” In Kelowna, AI deployment is not competing with infrastructure modernization for resources; instead, it is being advanced in parallel with ERP implementation and data governance. In other words, the city does not treat AI as a shortcut to skip system upgrades, but as a stress test that forces governance upgrades.

This approach is closer to reality, and also closer to how future city systems will evolve. A truly sustainable smart city is not one that layers AI onto old systems to produce demo effects, but one that first aligns data flows, business flows, and permission flows, and then lets automation enter the appropriate links.

Why city AI will not first change “the smartest parts,” but will first change “the most repetitive parts”

From the perspective of global urban governance trends, the first stage of AI entering municipal systems is usually not deep autonomy, but process relief. It will preferentially enter scenarios such as consultations, triage, knowledge retrieval, case classification, appointment guidance, frequently asked questions, and multilingual services, because these scenarios are highly repetitive, have relatively clear boundaries of responsibility, and have controllable error costs.

This does not conflict with cities’ long-standing pursuit of “efficiency.” On the contrary, it is a more realistic view of efficiency: not letting machines make more decisions for the government, but letting machines free people from low-value repetitive work. For a municipal department with limited staff, limited budget, and increasing compliance pressure, the significance of this kind of automation lies in expanding service coverage, not in creating a technological myth.The Kelowna case also highlights a trend that is often overlooked: in the future, differences in AI capability will depend not only on the model itself, but also on the quality and governance of an organization’s proprietary data. Models trained on the public internet can answer many general questions; but a city’s real competitiveness comes from how well it understands its local processes, geographic information, service records, permit rules, and facility conditions. Whoever has cleaner, more traceable, and more usable data is more likely to turn AI into a public capability rather than a public risk.

The next stage of urban digitalization: from “launching a tool” to “operating an information ecosystem”

Over the past decade or so, many smart city projects have failed not because the technology was not advanced enough, but because they treated the city as a product launch event rather than a complex system that must be continuously operated. On the surface, they seemed to have many functions; in reality, each system operated independently: websites, hotlines, case management systems, GIS, permit platforms, ERP, and archives were all fragmented, and data could not flow across different service scenarios.

The arrival of AI, paradoxically, makes this fragmentation even harder to hide. Because a chatbot must know where information comes from, when it expires, who is responsible for updating it, and which content requires human confirmation. In other words, AI turns the gray areas long tolerated in urban governance into visible problems.

This is precisely the deeper significance of the Kelowna case: it reminds the public sector that digitalization is not as simple as “adding a new entry point,” but requires building a sustainable urban information ecosystem. Including:

• lifecycle management of documents and data
• knowledge base governance for citizen services
• data standards linked to ERP and back-end processes
• automatic removal and review mechanisms for outdated information
• clear rules for human review and responsibility transfer

These tasks are not glamorous, and they are unlikely to appear on a launch stage, but they determine whether a city can truly move toward operationally viable intelligence.

The future competition among cities increasingly resembles a contest of governance capability

If we place the Kelowna case in the broader urban technology landscape, it corresponds to a wider shift: competition among cities is moving from “whether there is an AI project” to “whether AI can be embedded into governance systems.”

This shift will affect multiple levels. First is public service. In the future, city hotlines, websites, apps, and neighborhood access points will increasingly be supported through human-AI collaboration, and the citizen experience will be closer to “getting information instantly” rather than “submitting a form and waiting for a callback.”

Second is governance structure. AI forces cities to redefine which processes can be automated, which must retain human judgment, and which information belongs to public knowledge versus controlled data. This process will drive stricter data governance, and also a more mature awareness of digital sovereignty: cities cannot rely solely on external tools; they must also control their own data boundaries and content sovereignty.Once again, the issue is infrastructure. What appears on the front end as a chatbot actually requires a much higher level of systems integration behind the scenes, including identity authentication, access control, knowledge base updates, log auditing, cybersecurity, and cross-departmental coordination. The so-called “city operating system” is not a single platform, but a combination of these capabilities.

In this sense, AI is changing cities, but what it is changing is not the superficial layer of “what a city looks like,” but the underlying layer of “how a city operates.”

Conclusion: What cities truly need is not a robot that answers questions better, but a system that no longer creates mistakes

The reason Kelowna’s beach lifeguard incident matters is that it clearly exposes a common misunderstanding: the value of AI in cities is not to replace governance, but to test it. It can reveal all at once whether the data is accurate, whether the processes are clear, whether responsibilities are defined, and whether the system is outdated.

So the path to maturity for city AI may not be romantic. It will not begin with grand visions of autonomous driving, nor will it start with a citywide digital twin. It is more likely to begin with an old PDF, an ERP system that is being rebuilt, a citizen service portal that people ask questions through every night, and a group of technology managers willing to admit that “the system still isn’t clean enough.”

And the dividing line for future cities is forming right here: not which city is first to adopt AI, but which city is first to learn how to keep AI from being led astray by outdated data.

SEO Description

Kelowna city’s AI chatbot incorrectly cited an outdated PDF, exposing the core challenges of public-sector AI deployment: data governance, legacy systems, ERP upgrades, and urban digital infrastructure. From a smart city perspective, this article analyzes why the prerequisite for AI in municipal services is not model capability, but the reconstruction of the city’s information ecosystem.

Source URL

https://www.digitaljournal.com/article/when-your-data-has-a-lifeguard-problem/