AI, workflow automation or plain software: which one fixes your problem? 

Plain software follows fixed rules and gives you the same answer every time, and workflow automation moves your data and triggers actions between tools on those rules. AI reads, writes and judges where rules can't, like a messy document or a question in plain language. Use the simplest one that works and keep a person checking the AI.

What are the three ways to remove manual work from your business?

You can take manual work off your plate in three ways - plain software, workflow automation or AI. Each one suits a different kind of task and each one breaks in its own way.

• Plain software is a tool with fixed rules like the accounting software that adds up your invoice.
• Workflow automation is a connection between your tools like a reminder that goes out when a job is booked.
• AI is a model that reads or writes, like one that pulls the supplier and amount out of a photographed bill.

People use the word AI for all three, and I think that's where a lot of bad purchases start. A quote for an "AI project" can turn out to be a simple connection between two of your tools, which would have cost you less and run more reliably.

What is the difference between them, side by side?

I'd read the table from left to right as a ladder. Start with plain software, and if that doesn't fit then try automation, and use AI only for the one step that nothing simpler can handle.

Which one fits the everyday tasks in a small business?

Here are four everyday tasks, and in each case the right answer may be less exciting than you expect, which is usually a good sign.

Sending a reminder

A reminder before an appointment follows a rule - send this message a set time before the booking. That's workflow automation or a built-in feature of your booking software. AI adds nothing here and using it would only add cost and the chance of a strange message.

Reading a supplier invoice

Invoices arrive in many layouts as PDFs and photos, and writing a rule for every layout is slow work that nobody on your team will thank you for. This is a good place for AI to read the document and pull out the fields. Automation then carries the result to your accounting software, and a person reviews it before it posts.

Answering a guest enquiry

Your guests ask the same questions in many different words. AI can understand the question and answer from your own written information, and if the answer isn't in there then it should hand over to a person. Software with fixed rules then handles the booking.

Matching a payout file

Matching a payment file to your sales records is a rule-based job, because the amounts and references either match or they don't. Plain software or automation handles it, and the same input always gives you the same result. AI is the wrong tool here because you want certainty, not a best guess.

Why is "AI" so often the wrong answer for a small business?

AI is often the wrong answer because most business tasks already follow rules. Where a rule works it is cheaper, faster and easier to test, and it gives you the same answer every time, which is what you want for your money and your records.

• AI can give you a confident answer that is wrong while a rule fails in a way you can predict.
• AI needs review and monitoring, and a rule needs very little.
• AI results can change when the model behind them changes, but a rule stays exactly as written.
• AI can be slower and it costs more each time you use it.

None of this makes AI a bad choice. It makes AI a choice for a specific job, and reading, writing and judging text are jobs that rules can't do well, so that's where I think AI earns its keep.

What does each one need to keep running after launch?

Each option needs care after launch, but in very different amounts.

Plain software needs updates and the occasional change when your rules change. Workflow automation needs someone to notice when a connected tool changes its fields or its login, and since a broken connection often fails quietly you'll want to ask for alerts. AI needs the most attention, because someone has to read a sample of its output, correct its mistakes and update the instructions it follows.

I'd ask any provider who looks after the solution once it's live. If the answer is nobody then the first change in your tools will break it. In my opinion a named person who is accountable for the result is worth more than any feature.

What about keeping a human in the loop?

Keep a person in charge of anything AI produces that affects your money, clients or records. That person doesn't redo the work - they check it, approve it and fix what's wrong, which is quicker than doing the task by hand and protects you from confident mistakes.

The check can be light for low-risk tasks like drafting a reply to a routine question, and it should be firm for high-risk ones like posting an entry to your accounts. My rule of thumb is to decide the level of check before you build, and to write it down.

Software and automation are much easier to test in advance, because you can run a hundred sample cases and know what will happen, whereas with AI you keep watching.

How do you choose between the three of them?

Ask yourself three questions in this order, and stop at the first yes.

1. Can a tool you already own do it? Check for a setting, a template or a built-in feature, and if there is one then use it.
2. Can you write the rule in a few clear lines, and do your tools connect? If yes then workflow automation is your answer.
3. Does the task need reading, writing or judging messy content? If yes then use AI for that step only, and add a person to check the result.

Many real projects use all three, with software holding the records, automation moving the work and AI handling the one hard step. Keeping the AI step small makes it easier for you to test and cheaper to run.

Be wary of anyone who proposes AI before they've even looked at your process. A sound adviser will tell you when a spreadsheet fix or a simple connection beats an AI build, and I'd trust that person far more than the one selling the shiny option.

What if you are still not sure which one you need?

Run a small test before you buy anything. Take twenty real examples of the task, such as past invoices or enquiries, and try the simplest option on them first. Note how many it handles correctly and what goes wrong with the rest, and if a rule handles most of them then you may need nothing more.

If the rule fails on the messy cases then try AI on those cases alone. This tells you with your own data whether AI is worth adding, and it costs very little.

Write down what you learn and the cases that failed, because that record helps whoever builds the final version and shows you where a person is still needed.

Frequently asked

My short answer to most of the questions below is to use the simplest tool that works and add AI only where rules can't do the job. Hope this helps, and if you're still stuck you know where to find me!