AI is probably not what you think it is. You're probably not wrong about the bits you hate.
The complaints about AI are everywhere now, and most of them are fair.
The writing that sounds plausible and means nothing. The chatbot that answers a slightly different question than the one you asked, confidently, and then apologises when you point it out. The tool someone paid a lot of money for that turned out to be a wrapper around something free. The prediction that AI would transform your industry by now, which has not happened, or has happened in ways nobody predicted and not necessarily for the better.
If you have spent any time near this conversation and come away feeling that AI is mostly noise, a lot of evidence supports that view.
This is not an article that will try to talk you out of it entirely. It is going to try to do something more specific: separate the parts of the sceptical case that are well-aimed from the parts that are firing at the wrong target.
The things the sceptics are right about
Most AI content is bad. Generated text that sounds authoritative and says nothing. Blog posts that could have been written about any company, product or industry with the word swapped. Customer service bots that deflect rather than answer. If you have been on the receiving end of this, your irritation is proportionate to the offence.
AI has been sold at the wrong level of abstraction. Business owners have been told to "use AI" and handed access to a general-purpose tool with no specific task, no clear output and no realistic expectation. When that produces bad results, which it usually does because vague inputs produce vague outputs, the tool gets the blame. The tool is not entirely innocent, but the briefing is where things actually went wrong.
The hype has been enormous and the results have been uneven. If you have watched AI get announced as a transformation of your sector for three years and your sector looks more or less the same, your scepticism is empirically grounded. Some things have changed. Many things have not.
A lot of what gets called AI is not AI. This one matters and the site has a whole page on it, but the short version is: much of what gets sold under the AI label is automation: rules-based, predictable, doing exactly what it was told to do. Automation is not unintelligent, it is differently intelligent, and it is often the better solution: cheaper, more reliable, easier to maintain. The conflation is not accidental. "AI-powered" is easier to sell than "we built a sensible system."
Where the complaints sometimes miss
The sceptical case tends to be strongest when applied to AI at its most general (the chatbot trained on everything, the text generator pointed at no particular task) and weakest when applied to AI doing something specific.
There is a difference between asking AI to write your marketing strategy and asking it to draft replies to the twenty standard customer enquiries your business gets every week, in your tone, based on your actual answers. The first task is too broad for any tool to do well. The second is the kind of thing where AI genuinely saves time and does it reasonably.
The complaints are also sometimes aimed at the category rather than the implementation. A bad chatbot is evidence that a particular chatbot was poorly built or wrongly deployed. It is not evidence that all systems that understand natural language and generate useful responses are failures. The distinction feels pedantic until you realise it has practical consequences for decisions about what to try.
The specific things that actually work
Not as transformation. Not as a replacement for expertise or judgement. As tools that handle defined tasks in less time than it took before.
Summarising long documents and pulling out the relevant points. Drafting a first version of text where the content is mostly standard but the wording matters: a quote, an email, a product description. Turning messy input into structured information: a voicemail, a photo of a receipt, a recording of a meeting. Answering customer questions outside business hours on topics the business actually knows the answer to.
None of these will change your company. They will each save a modest amount of time on a task that currently takes longer than it needs to. Compounded, that adds up. On its own, each one is useful in a specific and unspectacular way.
That is a much smaller claim than the one AI has mostly been sold on. It is also a more honest one.
What this site is actually for
The scan on this site is not an attempt to sell you AI. It is an attempt to look at your specific business and tell you plainly which of the things above, if any, apply to you, and whether the answer is AI, automation, or neither.
The sceptic's instinct, that most of what gets sold as AI is not worth the money or the disruption, is a reasonable starting position. The only place it costs something is if it prevents a straightforward look at the smaller, quieter set of things that do work, for some businesses, in specific circumstances.
Whether your business is one of them is what the scan is for.
For a more personal version of this argument, including why our own accountant's scepticism turned out to be useful, that's here.
If you arrived here after watching one of our short videos and the format was part of the issue, that's explained here.