Somewhere between the headlines promising AI will transform everything and the frustrated forum posts from people who've been let down by it, there's a more useful conversation to be had. Not "is AI good or bad?" but "what do you actually need to know before you start — or before you go further?"
I've been using AI tools seriously for a while now, and the UKBF community has been having this conversation in real time across hundreds of posts. What follows is drawn from that thread and from my own experience. It isn't a beginner's guide — we've published those. It's the stuff that tends to only become clear once you've got some mileage on the clock.
Start with the problem, not the tool
This sounds obvious. It isn't, in practice.
@fisicx who started the thread this series has been drawing on, made a point that I think is the single most important thing anyone considering AI for their business should hear: do the business analysis first. Before you ask whether AI can help, ask what you actually need. He gave an example from his own experience — a company spending days each month producing a set of analytical reports, only to discover when they actually asked the recipients what they needed that almost all of it could be binned, replaced with a short email containing the few data points people actually used.
No AI required. Just asking the right question.
@Data Swami made a similar point from the opposite direction: most businesses that struggle to see value in AI haven't actually embedded it meaningfully. They've simply taken work they were already doing and passed it through a chat interface. The AI becomes a new layer on top of an existing process rather than a genuine improvement to the process itself. The result is underwhelming, and the AI gets the blame.
The lesson: before you reach for any tool, AI or otherwise, be clear about what problem you're actually trying to solve.
The free versions are not neutral
I've never used the free versions of AI tools, and this is deliberate. My reason, as I mentioned in the thread, is straightforward: I don't want my data used for training, and I'd rather have something ringfenced for my own use that I can control.
Data Swami put it more bluntly: with a free product, you are the product. If you're feeding business information, client details, or anything commercially sensitive into a free AI tool, you should understand clearly how that data is being used. Even with paid versions, it's worth checking the terms.
fisicx raised a counterpoint worth hearing: most small businesses using AI are using free versions, and they aren't going to pay. For a plumber using AI to help draft a quote or knock together a cash flow, the risk calculus is different to a professional services firm feeding client data into the same tool. Context matters. But knowing the distinction exists is the starting point.
The practical takeaway is simple: if you're using AI for anything business-critical or commercially sensitive, use a paid plan and read the data policy. It's not expensive, and the cost of not doing it can be.
It confidently makes things up — and that's a feature, not a bug
This is the one that catches the most people out, and it's worth being direct about it.
@Newchodge, who advises businesses on employment law, described seeing AI-generated legal advice posted on a UKBF thread that was clearly fabricated and at least seven years out of date — presented with complete confidence, reshared without anyone checking. She also noted that false case citations from AI have appeared in actual UK legal proceedings, with consequences for the firms involved.
@fisicx put it plainly: AI doesn't look things up the way a search engine does. It generates what a plausible answer looks like based on patterns in its training data. If the correct answer isn't well-represented in that data — because it's too recent, too niche, too specialised — the AI will produce something that sounds right but isn't. It doesn't flag uncertainty the way a careful human would.
@Mark T Jones shared a real-world example that makes this tangible in a different way: his wife uses an AI transcription tool for NHS patient notes, which saves significant time — but requires close management, because left unchecked, it will occasionally fuse unrelated parts of a conversation into the notes in ways that could be, as he put it, potentially dangerous. His favourite absurd example: it once attributed a patient's shoulder pain to tomato growth in warm weather, having picked up a piece of casual chat and incorporated it into the medical record.
The tool isn't broken. It's doing exactly what it was designed to do, generating plausible text based on what it heard. The problem is that plausible and accurate are not the same thing, and in a high-stakes context, that gap matters enormously.
The rule I apply: always verify anything an AI tells you before you act. Not because the AI is incompetent, but because confident fluency and factual accuracy are genuinely different things, and AI has plenty of the former without guaranteeing the latter.
The prompt is doing more work than you think
One of the most consistent observations across the thread is how much the quality of the output depends on the quality of the input.
@Frank the Insurance guy said it well: ask it for slop and you will get slop. The more specific the prompt, the better the result. @YasmeenLondon went further. In their view, most complaints about AI's capabilities come from people who haven't invested in learning to prompt effectively.
Fisicx described writing a 300-word prompt for a plugin project — carefully specifying requirements, conditions, and constraints — and getting a strong first result because the instruction was precise enough to be useful. He also noted that his background as a technical writer probably helped: the skill of communicating precisely what you need is directly transferable.
This isn't about learning a set of tricks. It's about treating AI the way you'd brief a capable but junior colleague who has no context about your business, your clients, or your industry unless you tell them. The more clearly you explain the task, the constraints, the audience, and the format you need, the more useful the output.
And when you're in a back-and-forth that isn't working — the AI going round in circles, giving you the same wrong answer with slight variations — starting a fresh conversation with a better-constructed prompt will almost always outperform trying to correct the existing one.
Know where it genuinely helps — and where it doesn't
One of the more useful things to emerge from the UKBF thread is a fairly clear picture of where AI delivers real value for small businesses and where it struggles.
Where it tends to help: low-level, repetitive tasks that don't require specialist knowledge. Drafting emails and standard correspondence. Summarising meeting notes. Reformatting content. First drafts of things that will be edited before use. Code for standard tasks if you know enough to review and fix what it produces. Debugging existing code. Brainstorming starting points for problems, you then work through them yourself.
Where it tends to struggle: anything highly specialised or niche. Anything that requires up-to-date knowledge it wasn't trained on. Legal, regulatory, and compliance tasks where accuracy is non-negotiable. Anything that requires a definitive answer rather than a plausible one. And as fisicx found, anything built around a custom API or edge cases that sit outside the mainstream, the AI simply has no data to draw on.
My own experience maps closely to this. AI is most valuable to me for the tasks I'd otherwise describe as "high effort, low value", the work that needs doing but doesn't require my direct judgment. Freeing up time and thinking space for the work that does.
Watch the pricing trajectory
This one isn't about how AI works — it's about how AI businesses work.
@UKSBD raised a concern that I think is underappreciated: once you're embedded in a tool and your workflows depend on it, the pricing leverage shifts to the provider. Several members noted that free tiers are becoming more restrictive over time, and that the pattern of tech tools — low cost to acquire users, higher cost once they're dependent — is well established.
Data Swami's advice was pragmatic: take advantage of the current pricing while it works in your favour, but build your processes so you're not irreversibly locked in to any single provider. The market is competitive enough right now that switching is possible. That may not always be the case.
The practical implication: don't build mission-critical processes around a single AI tool without thinking about what you'd do if the pricing changed significantly or the tool changed in ways that didn't suit you.
AI is genuinely useful for a meaningful range of business tasks. It is not a shortcut around thinking. It is not infallible. It is not going to replace the judgment, experience, and relationships that make your business what it is.
What it can do, used thoughtfully, is take some of the weight off the parts of the day that drain time without adding value, so that more of your energy goes where it actually matters.
The businesses that will get the most from AI are the ones that were clear about what they needed before they started, realistic about what the tools can and can't do, and disciplined about keeping human judgment in the loop where it counts.
That's not a complicated formula. It just requires actually thinking about it before you start, which, as it turns out, is good advice for implementing almost anything.
This article draws on the "AI does have its uses" discussion thread on UK Business Forums, with contributions from members @fisicx, @Data Swami, @Newchodge, @Mark T Jones, @Frank the Insurance guy, @YasmeenLondon, and @UKSBD.
Related reading: The AI Divide: What UKBF Members Really Think in 2026 | Is AI Making the Internet Worse? | AI and Your Business: Useful Tool, Not a Magic Wand
I've been using AI tools seriously for a while now, and the UKBF community has been having this conversation in real time across hundreds of posts. What follows is drawn from that thread and from my own experience. It isn't a beginner's guide — we've published those. It's the stuff that tends to only become clear once you've got some mileage on the clock.
Start with the problem, not the tool
This sounds obvious. It isn't, in practice.
@fisicx who started the thread this series has been drawing on, made a point that I think is the single most important thing anyone considering AI for their business should hear: do the business analysis first. Before you ask whether AI can help, ask what you actually need. He gave an example from his own experience — a company spending days each month producing a set of analytical reports, only to discover when they actually asked the recipients what they needed that almost all of it could be binned, replaced with a short email containing the few data points people actually used.
No AI required. Just asking the right question.
@Data Swami made a similar point from the opposite direction: most businesses that struggle to see value in AI haven't actually embedded it meaningfully. They've simply taken work they were already doing and passed it through a chat interface. The AI becomes a new layer on top of an existing process rather than a genuine improvement to the process itself. The result is underwhelming, and the AI gets the blame.
The lesson: before you reach for any tool, AI or otherwise, be clear about what problem you're actually trying to solve.
The free versions are not neutral
I've never used the free versions of AI tools, and this is deliberate. My reason, as I mentioned in the thread, is straightforward: I don't want my data used for training, and I'd rather have something ringfenced for my own use that I can control.
Data Swami put it more bluntly: with a free product, you are the product. If you're feeding business information, client details, or anything commercially sensitive into a free AI tool, you should understand clearly how that data is being used. Even with paid versions, it's worth checking the terms.
fisicx raised a counterpoint worth hearing: most small businesses using AI are using free versions, and they aren't going to pay. For a plumber using AI to help draft a quote or knock together a cash flow, the risk calculus is different to a professional services firm feeding client data into the same tool. Context matters. But knowing the distinction exists is the starting point.
The practical takeaway is simple: if you're using AI for anything business-critical or commercially sensitive, use a paid plan and read the data policy. It's not expensive, and the cost of not doing it can be.
It confidently makes things up — and that's a feature, not a bug
This is the one that catches the most people out, and it's worth being direct about it.
@Newchodge, who advises businesses on employment law, described seeing AI-generated legal advice posted on a UKBF thread that was clearly fabricated and at least seven years out of date — presented with complete confidence, reshared without anyone checking. She also noted that false case citations from AI have appeared in actual UK legal proceedings, with consequences for the firms involved.
@fisicx put it plainly: AI doesn't look things up the way a search engine does. It generates what a plausible answer looks like based on patterns in its training data. If the correct answer isn't well-represented in that data — because it's too recent, too niche, too specialised — the AI will produce something that sounds right but isn't. It doesn't flag uncertainty the way a careful human would.
@Mark T Jones shared a real-world example that makes this tangible in a different way: his wife uses an AI transcription tool for NHS patient notes, which saves significant time — but requires close management, because left unchecked, it will occasionally fuse unrelated parts of a conversation into the notes in ways that could be, as he put it, potentially dangerous. His favourite absurd example: it once attributed a patient's shoulder pain to tomato growth in warm weather, having picked up a piece of casual chat and incorporated it into the medical record.
The tool isn't broken. It's doing exactly what it was designed to do, generating plausible text based on what it heard. The problem is that plausible and accurate are not the same thing, and in a high-stakes context, that gap matters enormously.
The rule I apply: always verify anything an AI tells you before you act. Not because the AI is incompetent, but because confident fluency and factual accuracy are genuinely different things, and AI has plenty of the former without guaranteeing the latter.
The prompt is doing more work than you think
One of the most consistent observations across the thread is how much the quality of the output depends on the quality of the input.
@Frank the Insurance guy said it well: ask it for slop and you will get slop. The more specific the prompt, the better the result. @YasmeenLondon went further. In their view, most complaints about AI's capabilities come from people who haven't invested in learning to prompt effectively.
Fisicx described writing a 300-word prompt for a plugin project — carefully specifying requirements, conditions, and constraints — and getting a strong first result because the instruction was precise enough to be useful. He also noted that his background as a technical writer probably helped: the skill of communicating precisely what you need is directly transferable.
This isn't about learning a set of tricks. It's about treating AI the way you'd brief a capable but junior colleague who has no context about your business, your clients, or your industry unless you tell them. The more clearly you explain the task, the constraints, the audience, and the format you need, the more useful the output.
And when you're in a back-and-forth that isn't working — the AI going round in circles, giving you the same wrong answer with slight variations — starting a fresh conversation with a better-constructed prompt will almost always outperform trying to correct the existing one.
Know where it genuinely helps — and where it doesn't
One of the more useful things to emerge from the UKBF thread is a fairly clear picture of where AI delivers real value for small businesses and where it struggles.
Where it tends to help: low-level, repetitive tasks that don't require specialist knowledge. Drafting emails and standard correspondence. Summarising meeting notes. Reformatting content. First drafts of things that will be edited before use. Code for standard tasks if you know enough to review and fix what it produces. Debugging existing code. Brainstorming starting points for problems, you then work through them yourself.
Where it tends to struggle: anything highly specialised or niche. Anything that requires up-to-date knowledge it wasn't trained on. Legal, regulatory, and compliance tasks where accuracy is non-negotiable. Anything that requires a definitive answer rather than a plausible one. And as fisicx found, anything built around a custom API or edge cases that sit outside the mainstream, the AI simply has no data to draw on.
My own experience maps closely to this. AI is most valuable to me for the tasks I'd otherwise describe as "high effort, low value", the work that needs doing but doesn't require my direct judgment. Freeing up time and thinking space for the work that does.
Watch the pricing trajectory
This one isn't about how AI works — it's about how AI businesses work.
@UKSBD raised a concern that I think is underappreciated: once you're embedded in a tool and your workflows depend on it, the pricing leverage shifts to the provider. Several members noted that free tiers are becoming more restrictive over time, and that the pattern of tech tools — low cost to acquire users, higher cost once they're dependent — is well established.
Data Swami's advice was pragmatic: take advantage of the current pricing while it works in your favour, but build your processes so you're not irreversibly locked in to any single provider. The market is competitive enough right now that switching is possible. That may not always be the case.
The practical implication: don't build mission-critical processes around a single AI tool without thinking about what you'd do if the pricing changed significantly or the tool changed in ways that didn't suit you.
AI is genuinely useful for a meaningful range of business tasks. It is not a shortcut around thinking. It is not infallible. It is not going to replace the judgment, experience, and relationships that make your business what it is.
What it can do, used thoughtfully, is take some of the weight off the parts of the day that drain time without adding value, so that more of your energy goes where it actually matters.
The businesses that will get the most from AI are the ones that were clear about what they needed before they started, realistic about what the tools can and can't do, and disciplined about keeping human judgment in the loop where it counts.
That's not a complicated formula. It just requires actually thinking about it before you start, which, as it turns out, is good advice for implementing almost anything.
This article draws on the "AI does have its uses" discussion thread on UK Business Forums, with contributions from members @fisicx, @Data Swami, @Newchodge, @Mark T Jones, @Frank the Insurance guy, @YasmeenLondon, and @UKSBD.
Related reading: The AI Divide: What UKBF Members Really Think in 2026 | Is AI Making the Internet Worse? | AI and Your Business: Useful Tool, Not a Magic Wand
