Do you actually need AI for this?

The short answer

Ask five questions before you buy or build anything. If any answer is "no", stop. Most small businesses that regret an AI purchase skipped at least one of them.

This guide exists because the most useful thing a problem-first resource can do is tell you when you do not need another tool.

The five questions

1. Does this task repeat?

If you do it once a quarter, do it by hand. Automation has setup cost, review cost and maintenance cost. A task that happens four times a year never repays them.

Pass condition: the task happens at least weekly, or it happens less often but takes hours each time and follows the same steps.

2. Can you describe the output well enough to judge it?

You need to be able to look at the result and say "correct" or "wrong" quickly. If you cannot define what good looks like, you cannot review the output, and an unreviewed automation is a liability.

Pass condition: you can write a one-paragraph description of a correct result, with an example.

3. Is the input mostly text, or structured data you already have?

AI is strongest on unstructured text: emails, notes, documents, reviews. It is weak when the input is missing, inconsistent, or locked inside a system you cannot export from.

Pass condition: the inputs are available, in a usable form, without a data-cleanup project first.

4. Would a mistake be cheap to catch?

A bad subject line is cheap. A wrong price quoted to a customer, a missing compliance disclosure, or a deleted record is not.

Pass condition: errors surface before they cost money, or a person reviews the output every time.

5. Have you checked whether your existing software already does this?

Many tools already include summarisation, drafting, routing or reporting that you are paying for and not using. Adding a second product to do the same job creates two sources of truth.

Pass condition: you have looked at what you already own, and it genuinely does not cover the need.

When the answer is "use what you have"

• Your email client already sorts, labels and schedules.
• Your CRM already has templates, reminders and pipelines.
• Your accounting software already extracts receipt data.
• Your help desk already routes by keyword.

Enabling a feature you already pay for beats integrating a new vendor almost every time.

When manual is the right answer

• The task is infrequent and judgement-heavy.
• The volume is low enough that reading everything is faster than configuring anything.
• The work builds customer trust that automation would weaken, such as a personal follow-up after a complaint.

Manual is not a failure state. It is often the correct design.

When AI is genuinely a good fit

You answered yes to all five, and the task is drafting, summarising, classifying or extracting. Those are the four jobs where current tools are reliable enough to save time without much drama.

Red flags in a vendor pitch

• A claimed percentage saving with no stated baseline.
• "Set it and forget it" for a workflow that touches customers.
• No clear answer about where your data goes or how long it is retained.
• A price that only makes sense at a volume you do not have.
• A demo that is really a generic chatbot wearing your logo.

How to decide in 20 minutes

Write the task, the frequency, the current time cost, the review cost, and the failure cost. If the time saved per month is smaller than the setup plus review time, do not proceed. That arithmetic is unglamorous and it is usually right.

Notes on sourcing

This guide is a decision framework, not a product review. It is labelled researched. No savings figures are claimed because none were measured on your business.