Start with a task you already understand.

Perhaps you turn customer enquiries into job briefs, prepare a weekly update or pull the same details from several documents. You know the information, the intended result and what a mistake would look like. That gives you a useful starting point for AI in your business.

The first goal is a finished piece of work you can check and use. Once you have that, you can decide whether the process is worth repeating.

What does AI for business mean?

In this guide, it means using a generative AI assistant to help interpret information and prepare an output: a summary, comparison, document or draft reply. You describe the work, provide relevant material and review what it produces.

That is different from a spreadsheet formula or a fixed rule. A formula follows the calculation you set. An AI assistant can help organise an untidy enquiry or suggest clearer wording, but you still need to check whether its interpretation fits the source.

It also helps to distinguish three stages:

Stage What you do What you have
First attempt Give an instruction in a chat and review the answer One useful output
Reusable workflow Save the instructions, input format and checks A process you can run again
Configured automation Set up the supported tools, access and trigger A process with agreed automatic steps

Writing “do this every Monday” in an ordinary prompt does not establish a working weekly automation. Scheduling, access to files and actions in other tools need a supported setup and testing.

1. Choose one task with a clear finish

Look for work that repeats, uses information you can supply and ends with something easy to inspect.

Recurring task Useful first output What you check
Reading new enquiries Internal brief and missing questions Customer request, dates and scope
Preparing a client update Draft from your approved notes Progress, commitments and next steps
Reviewing supplier proposals Comparison against your criteria Prices, exclusions and source references
Planning marketing Campaign brief and draft content Offer, audience and factual claims

Choose one. Write down its starting material and its finish: “From this enquiry, prepare a job brief and a draft reply for me to approve.”

For a first attempt, keep the task away from commitments you cannot easily reverse. A draft quotation can be checked before it goes out. An automatically sent quotation with the wrong price creates a different problem.

2. Prepare a small, usable source pack

Gather only what the task needs. An enquiry workflow might use the enquiry text, your service description and a list of questions you normally ask. A weekly update might use approved meeting notes and the previous update’s format.

Keep one clearly labelled version of each source. Tell the assistant which document controls facts such as current services or agreed dates. If two sources disagree, ask it to flag the conflict.

Use fictional or anonymised material while learning. Before providing business information, check your organisation’s rules and the selected tool’s data settings. A desktop app can still use cloud services; OpenAI, for example, documents cloud and local work options in its desktop app. Check the actual processing arrangements before sharing sensitive material.

Choose a tool you are permitted to use, then try this actual task. Check that you can provide the required input and retrieve a usable result. If you are choosing between assistants, compare Claude and ChatGPT for marketing using the same brief.

3. Give the assistant a useful instruction

A prompt is your instruction to the AI. Include the result you need, the source material, the output format and any important limits.

OpenAI’s official prompting guidance recommends giving ChatGPT the relevant goal, context, output and boundaries. It also recommends reviewing the response and refining it with follow-up instructions. You can apply that approach without learning technical terminology.

“Help with enquiries” leaves most decisions open. “Turn this enquiry into a factual job brief, list missing information and draft a reply without promising a price or date” gives the work a clear finish.

4. Work through an example

Northbank Decorating is a fictional business. This fictional enquiry arrives:

Can you quote for repainting our meeting room? Walls only, roughly four metres by five metres. We would prefer the work after 6 pm. Is next Tuesday possible?

The owner wants a brief before replying. A useful result would record:

  • Requested work: Repaint the meeting-room walls.
  • Information supplied: Approximate room dimensions of four by five metres; evening work preferred.
  • Timing: Next Tuesday requested; availability unconfirmed.
  • Missing information: Location, wall height and condition, access, and whether furniture needs moving.
  • Next action: Ask for the missing details before confirming the quotation process.

The room dimensions do not establish the paintable wall area. “Next Tuesday” also needs a calendar date before anyone commits to it. These are exactly the details the owner should catch during review.

Here is a reusable prompt. Replace the bracketed sections with your own material:

Turn the enquiry below into an internal job brief and a reply draft.

Business context: [services we provide and our quotation process]
Enquiry: [paste an approved or fictional enquiry]

Return:
1. The requested work and facts supplied by the customer.
2. Missing information, in order of importance.
3. Any unclear dates, conflicting details or assumptions to check.
4. A short, friendly reply asking the necessary questions.

Use only the supplied facts. Distinguish customer requests from
confirmed arrangements. Do not invent a price, availability or promise.
Prepare a draft for my approval; do not send anything.

The result is a prepared brief and reply, ready for the owner to check. It has not booked the job or verified availability.

5. Review against the source

Read the output beside the input. Check names, quantities, dates, prices and commitments first. Then check whether it omitted something that changes the work.

For Northbank, the owner should confirm that the reply says “walls only,” preserves the evening preference and asks about next Tuesday without accepting it.

If the answer needs improvement, give precise feedback: “Keep the questions, shorten the introduction and make clear that we have not confirmed the date.” Save the instruction that produced the better result.

Try the process on a straightforward example, an incomplete one and an unusual one. Record where you had to intervene. If the assistant repeatedly misses the same detail, improve the source material or review step before expanding the workflow.

6. Decide whether to repeat or automate it

Compare the complete process with your normal method, including preparation, checking and corrections. Record time spent, missed details and whether the finished output was useful. A quick first draft is only part of that comparison.

If it works well, save the prompt, input checklist, output example and review instructions together. Name the person responsible for the final decision.

Only then consider automation. Define what starts the process, what it can access, where the result goes and what happens when information is missing. Confirm each connection in the actual tool and account. Keep approval before sending customer messages or making commitments.

You can explore AI email automation or move into a complete AI marketing workflow when your first task is dependable.

If you want help setting up one recurring desktop AI workflow, request a free assessment. We review your task and agree the scope, required tools and setup before payment, then build and repeat the agreed workflow with you.

Frequently asked questions

Do I need to know how to code?

You can begin with an approved assistant, written instructions and material you supply yourself. Whether a later automation needs coding depends on the task and the tools it must use.

What should I do if the AI gives a confident answer I cannot verify?

Treat that part as unresolved. Ask which source supports it, open that source and check the original wording. If you cannot establish the answer, remove the claim or ask the person who owns the information.

When is a workflow ready for someone else to use?

When they can follow the saved instructions, recognise an incomplete result and know who approves the output. Have them run a fresh example while you observe; use their questions to improve the instructions.