What can AI actually do for a small business, and what it can't?

AI is very good at language outputs, such as drafting, summarising, rewriting, answering questions, and spotting patterns in text. However, it isunreliable at judgment, factual accuracy, and anything that needs context it was not given. Knowing which side of that line a task sits on is the difference between AI helping you and AI causing problems.

Key Takeaways

  • AI is strong at language tasks: drafting, summarising, rewriting, answering.

  • It is weak at judgment, factual accuracy, and generally anything needing context it was not given

  • It can be confidently wrong, because it predicts plausible language rather than checking truth and makes assumptions where there is ambiguity.

  • The skill is matching each task to what AI is actually good at and providing it enough context and direction to reduce ambiguity.

What is AI good at?

The honest list is a bit narrower than some of the hype would have you believe, and more useful. AI reliably helps with drafting a first version of almost any text, summarising something long into something short, rewriting for tone or clarity, brainstorming options, explaining an unfamiliar concept, and giving structure to a messy set of notes. These can be real, everyday wins for a small business owner

We have also experienced AI to be good at identifying patterns, analysing data, creating hypotheses, assisting with research, and tracking digital marketing metrics.

What it can’t do reliably, yet?

  • Guarantee facts. It can state something false with total confidence.

  • Make a judgment call. Pricing, hiring, a delicate customer decision; these all need a human to make a judgement.

  • Know your business. It only knows what you tell it in the conversation and what you have provided in context.

  • Stay current. Unless it has live access, its knowledge has a cut-off.

  • Take responsibility. The accountability for any output is always yours.

We’ve also found that it tries to cut corners, much like a human tries to, when faced with a big task. It can lose context during long conversations so will ‘forget’ what was instructed earlier in the chat. Most of the LLMs powering the tools most of us use currently have been trained to be agreeable, so many tools will agree with you and follow the direction you give it, even if that isn’t the best decision.

Why does it sound so confident when it is wrong?

Because of how it works. AI predicts the most plausible next words, not the most true ones. Most of the time plausible and true line up, which is why it is useful. But when they don’t, it does not hesitate or flag the doubt. It delivers the wrong answer in the same confident tone as the right one. That’s not a bug you can switch off, unfortunately, it is the nature of the tool, and it is exactly why a human check matters so much.

How do I use it safely given all that?

Match the task to its strengths, give it the context it needs, and verify anything that matters. Use it freely for drafting and summarising, where a wrong turn is easy to catch. Be careful with facts, figures, and judgment, where a confident error is costly. The rule of thumb - make it a sandwich: give it direction and context, let AI do the first draft, and you keep the final say.

A capable assistant, not an oracle

The most useful way to think about AI is as a fast, capable assistant who is occasionally and confidently wrong. You would not let an assistant like that send work unchecked, and you would not refuse their help either. You would use them for what they are good at and apply your own judgment to the rest. That judgment is the capability worth building, and it outlasts any tool.

Start here

Our free video series walks through exactly these misconceptions and how to work around them. Short videos, with an AI ethics checklist and use policy template included.

Go to the video series

Subscribe to the newsletter for weekly capability building insights

Sign up for the weekly newsletter

Who is 25eight?

  • 25eight is a certified B Corp deeply committed to our impact on people and planet.

  • We have been developing and integrating artificial intelligence systems since 2020, ethically and with a human-centred approach.

  • Our capability building methodology has been delivered to 11,300+ businesses, in sectors from advanced manufacturing to defence and health.

  • We have measured impact of AI capability building in the businesses we have supported since 2025.

Frequently Asked Questions

  • Language tasks: drafting, summarising, rewriting, answering questions, brainstorming, and giving structure to messy notes. These are where it reliably saves time.

  • It cannot guarantee facts, make judgment calls, know your business without being told, stay current without live access, or take responsibility for its output. Those remain human.

  • Because it predicts plausible language rather than checking truth. When plausible and true diverge, it still answers in a confident tone, which is why human review matters.

  • Trust it for drafting and summarising, where errors are easy to catch. Verify anything involving facts, figures, or judgment before you rely on it.

  • Match tasks to its strengths, give it the context it needs, and keep a person reviewing anything important. Let AI draft, and keep the final say