Blog > Don’t Make The New York City Mistake
Don’t Make The New York City Mistake
AI is good for some problems, terrible for others. Learn how to spot which is which
Uncategorized
New York City began 2026 with dumping its flagship AI experiment. Its chatbot was meant to help the public navigate city rules, but ended up being described as “unusable”.
For marketers in financial services firm, this story has a familiar ring. Before, employees may have been covert with their AI use. Now that companies have set up safe access to protect data, they are encouraging employees to use the tools to improve the business. Yet adoption has been patchy.
Whenever employees have set ways of getting work out the door, it’s not always obvious where the AI fits in. A first step towards capturing value is spotting the type of problems where models can actually help.
100% problems vs 80% problems
By and large, problems come in two flavours.
A 100 percent problem is one where errors carry a severe penalty, so the fault tolerance is very low. A good example is blood tests, where getting a marker wrong can have dire consequences.
In regulated financial services, some processes naturally fall into this bucket. Regulated statements must have the correct figures, and product specs must be correct.
An 80 percent problem is one that aims to clear a “good enough” bar. Think of choosing a bouquet of flowers; it needs to fit your taste, the budget, the occasion. Many different bouquets will do.
Large language models are badly suited to the first category. They are also unusually good at the second.
An important comment on the Pareto principle. The Pareto principle is all about the trade-off between effort and output. It states that for a little effort you get most of the benefit.
Here the issue is the task type. No amount of effort will get the model to behave reliably on a 100 percent task.
So the very first step is to steer clear of 100 percent problems, and push AI models toward 80 percent work.
The NYC AI Case Study
In October 2023, New York City announced a plan to harness AI. Eric Adams’s administration released a chatbot for checking city rules and regulations, pitched at people starting and running businesses.
But by March 2024, The Markup had recorded the bot giving guidance that was wrong (and in some cases illegal). The AI told employers they could take workers’ tips, and told and landlords they could discriminate based on a tenant’s source of income.
On top of model hallucinations, the information on rights and rules was inaccurate and often incomplete. For example, the bot didn’t know that in 2020 the city mandated businesses to accept cash, thus protecting unbanked customers.
At the time, the city admitted the bot (which had cost about $600,000 to build) was still in pilot stage, but it was committed to “ironing out the kinks”. Alas, the bot was failing at a 100% problem, by giving definitive guidance on rights and rules.
In January 2026, at a press conference on a $12 billion budget gap, incoming mayor Zohran Mamdani singled out the chatbot as an extravagant cost, calling it “unusable”.
A Lesson from the Big Apple
Incoming mayors blaming outgoing mayors is as predictable as the tides. Equally, the city would not have cut a trusted and genuinely useful bot – it would have touted it.
So the bot suffered from two issues, closely linked:
- It tried to use AI to solve a 100 percent problem. Notably, the NYC system could have been designed differently: in response to a question, it could have pointed users to the relevant rules, while adding a cautious note on edge cases and exceptions. While this would have made its advice less definitive, its scope would be reduced to and 80 percent problem.
- Users weren’t using it. After the early gaffes, the chatbot quickly faded from view.
Irrelevancy is a poor outcome. Companies are now making substantial upfront investments in AI, and are left hoping employees’ use delivers a return.
FinText trains financial services marketing teams to get real value from AI. That includes choosing the right problems, mapping workflows, and designing guardrails that make ‘good enough’ reliable.
We have a wealth of experience with financial services workflows, and understand companies’ unique challenges in putting AI to use. Talk to us to learn more!
Discover more
The sheer variety of games available at magic reels casino would be impressive on its own, but what elevates the experience is how thoughtfully everything is presented and organised for the player. Slots are grouped not just by provider but by theme, volatility, and feature type, making it simple to find exactly the kind of session you are in the mood for today. Table games and live dealer rooms receive equal attention, ensuring that no category feels like an afterthought.
Players who appreciate transparency will feel right at home at red lion casino, where terms and conditions are written in plain language and important details are never buried in fine print. The game lobby is organised with practical filters that actually work, letting you sort by provider, category, or popularity in seconds. Deposits land in your account almost immediately, and the withdrawal process follows a predictable timeline that the platform consistently meets without excuses.
The team behind wild dice casino clearly understands that a great online casino is more than just a collection of games thrown together on a website. The entire user journey, from sign-up to cashout, has been refined to remove friction and maximise enjoyment at every touchpoint. Game loading times are consistently fast regardless of device, the promotion schedule offers genuine value without impossible wagering requirements, and the support desk operates with visible competence and care.
One of the quietly impressive things about allwins casino is how the platform manages to stay current without chasing every passing trend at the expense of what already works well. The core game library is rock solid with hundreds of proven titles, while new additions are clearly vetted for quality before they earn a spot in the catalogue. This curated approach means players can pick almost any game at random and still end up with a polished and engaging experience.
There is a reason why so many players keep returning to mr james week after week without ever feeling the need to look elsewhere for entertainment. The game selection covers every major genre you could want, from high-volatility slots with massive win potential to relaxed table games perfect for a quiet evening session. Customer support agents respond quickly and knowledgeably, treating each enquiry with the professionalism it deserves. The overall experience simply feels well thought out.
The banking experience at vipzino casino is one of the smoothest in the industry, with a range of deposit methods that all process instantly and a withdrawal system that meets its stated timelines with impressive consistency. But fast payments would mean little without great games to spend them on, and the catalogue delivers handsomely across slots, table games, and live dealer options. Everything is tied together by an interface that never gets in the way of the fun.
The banking experience at madslots casino is one of the smoothest in the industry, with a range of deposit methods that all process instantly and a withdrawal system that meets its stated timelines with impressive consistency. But fast payments would mean little without great games to spend them on, and the catalogue delivers handsomely across slots, table games, and live dealer options. Everything is tied together by an interface that never gets in the way of the fun.
What makes the game library at magius casino special is not just its size but the evident care that has gone into selecting every title it contains for the player catalogue. Each slot, each table game, and each live dealer room feels like it belongs and has been chosen to serve a specific player preference or mood. The result is a collection where browsing itself becomes enjoyable, because you trust that whatever catches your eye will deliver a quality gaming experience.