Dorothy has a byline in Wholesale Manager. Read the full article here:
https://wholesalemanager.co.uk/why-ai-wont-fix-broken-processes
Or read the summary below.
Generative AI tools like ChatGPT, Gemini and Claude are remarkably convincing to talk to, and that can lead you to believe they can do absolutely anything for your business. It's worth pausing on how they actually work before deciding how much weight to put on them.
In simple terms, these systems break language down into small chunks called tokens and learn how likely different tokens are to follow one another. From those patterns, they predict the next most probable token, over and over, and that's how they generate text, code or images. Predicting the next likely word isn't the same thing as understanding your business, however impressive the result feels.
Rubbish In, Rubbish Out
That distinction matters because of a much older principle: rubbish in, rubbish out. If your systems hold messy or inconsistent data, building AI on top of that mess won't tidy it up, it will amplify it. Building on poor data is rather like adding an extension to a house with dodgy foundations.
Before training any AI on your own data, it pays to go looking for the little inconsistencies that creep into most systems over time: a full name typed into a first name field, a product code sitting in the wrong ID field, records linked only by matching free text rather than a proper reference, or fields left open to free text when a fixed set of choices would have prevented errors in the first place. Most businesses have several of these lurking somewhere, whether they realise it or not.
Where to start?
A good starting point, before spending anything, is to look honestly for what is broken, or as we tend to put it, shonky. Watch for processes that grind to a halt whenever one particular person is away, repetitive manual tasks, information being retyped from one system into another, and processes that regularly go wrong. Talking to your own staff is often the quickest way to find these, since they're usually the ones quietly patching over the rough edges. Bringing in an outside pair of eyes helps too, precisely because familiarity can make problems invisible.
It's worth being clear that AI is a tool, not a technical resource in its own right. It works well when you genuinely understand your business and have someone who can check that what it produces is actually right. It's particularly strong where there's a large volume of data to draw predictions from, or for brainstorming, summarising long documents, and helping you find things buried somewhere on your server. Where accuracy really matters, a traditional software approach, or a considered mix of the two, is often the safer choice, since AI can still get things confidently wrong.
Cost
Cost and data privacy are worth thinking through as well. Pricing among the big AI providers is competitive for now, but that won't necessarily last as demand grows. And before you commit, it's worth reading the terms carefully to understand who owns the data once an AI has been trained on it, and whether that information could end up benefiting your competitors.
Ultimately, the innovation still has to come from you. AI can copy and combine what already exists, but it can't originate the ideas that will move your business forward.