There is a lot of AI that demos brilliantly and helps no one. It answers a clever question on stage, then quietly does nothing useful once it is live. The interesting question is not whether AI is impressive. It is whether a given tool does a real job, day after day, without needing a person to babysit it.
The test we use
Before we build anything with AI, we ask three things:
- Does it do a specific job? "An assistant" is not a job. "Answers the ten questions we get asked most and books the good enquiries in" is a job.
- Is it grounded in something true? A tool that makes things up is worse than no tool. Useful AI answers from your actual content, with limits, and hands over to a person when it is out of its depth.
- Can you measure it? If you cannot see what it handled and what it passed on, you cannot trust it or improve it.
If a tool fails those three, it is a demo, not a system.
Where it genuinely helps
The quiet, repetitive work is where AI earns its place:
- Qualifying enquiries. A chatbot trained on your content can answer the common questions around the clock and write the qualified leads straight into your CRM, so your team only sees the ones worth their time.
- Checking work at scale. On the KC Landscapes platform, completed-work photos are audited automatically, so every visit is held to a consistent standard without someone reviewing each one by hand.
- Removing manual admin. Sorting, summarising and routing information that used to eat an afternoon.
None of these are flashy. All of them save real time.
Honourable mentions: off-the-shelf tools that pass the test
We build AI where it needs to be bespoke, but plenty of ready-made tools clear our three questions for everyday jobs. A list like this dates quickly, so treat it as a snapshot of what we would actually recommend to a small business today, organised by the job, because the job is the point:
- Drafting and thinking. Claude or ChatGPT. Pick one and learn it properly rather than dabbling in both. The job: first drafts, summaries, rewording, thinking a problem out loud. The caveat: they will confidently invent facts, so anything factual gets checked before it leaves the building.
- Meeting notes. Fathom has a genuinely useful free tier, and Otter is the familiar name. The job: notes and actions written for you, so you can actually be in the meeting instead of scribing it.
- Research with sources. Perplexity answers questions with citations you can click and check. The job: the first twenty minutes of any research task, done in two.
- Everyday graphics. Canva has folded AI into a tool many businesses already pay for. The job: social posts and simple graphics to a consistent standard without waiting on a designer.
- Video and podcast editing. Descript lets you edit a recording by editing its transcript, deleting a sentence of text deletes it from the video. The job: turning one recording into usable clips without learning an editor.
- Receipts and bookkeeping. Dext reads receipts and invoices into your accounts, and Xero now suggests its own reconciliation matches. The job: the envelope of receipts, gone.
Notice what is not on the list: anything whose pitch is "AI for everything". Every tool above does one job you already have. That is the pattern.
Buy it or build it?
The honest split: buy when your job is the same as everyone else's, meeting notes are meeting notes, and the tools above are cheap, polished and maintained by someone else. Build when the value lives in your data and your process: answering questions from your own content, checking your work to your standard, feeding your CRM your way. That is when off-the-shelf becomes a compromise and something built to fit starts earning its keep. Most businesses need one or two built things and a handful of bought ones, not a platform.
The part people skip
If a tool touches customer data, it needs to be built responsibly: sensible limits, a clear handover to a human, and the data protection and governance to match. That is not a nice-to-have. It is the difference between a tool you can put your name to and one that embarrasses you later.
The short version
Ignore the demo. Ask whether the tool does a specific job, stays grounded in the truth, and can be measured. The AI worth building is usually the quiet kind that just gets on with it.
Wondering where AI actually fits in your business? Book a discovery call and we will give you an honest view, including "nowhere yet" if that is the truth.
Common questions
What should a small business use AI for first?
Start with the job that eats the most repeated time: usually answering the same enquiry questions, meeting notes, or drafting routine documents. Pick one tool for that one job, run it for a month, and judge it on time saved. Resist starting with the most impressive demo; start with the most boring job.
Are free AI tools good enough for business use?
Often, yes, to begin with. Fathom's free tier, the free versions of Claude and ChatGPT, and Canva's included AI features cover a surprising amount. Pay when you hit a real limit, not before. The main thing to check on any free tier is what happens to your data, especially whether your content is used for training, and most tools let you switch that off.
Can I just use ChatGPT for everything?
For drafting and thinking, close to it. But a general chatbot does not know your prices, your availability or your policies, and it will guess when asked. The moment AI speaks to your customers or checks your work, it needs to be grounded in your actual content with a proper handover to a human, which is the point where off-the-shelf stops and built-to-fit begins.
How much does it cost to add AI to a business?
The bought tools above run from free to roughly £10 to £30 per person per month, and the running cost of a well-built bespoke assistant is usually pennies per conversation. The real cost question is the build, which depends entirely on the job, which is why we scope it around a specific job rather than "adding AI". If the job does not justify the cost, we will say so.
