How to start using a dot for one useful business task

Start with a task you can judge.

That is the simplest advice I can give a business owner who wants to use a dot. Pick something specific enough that you will recognize a good result: a meeting follow-up draft, an organized set of next actions or a checked website page.

In my own work, the most useful requests have a clear output and a visible review step. The frustrating ones are often missing one of those pieces. There is plenty of activity, but it is hard to tell whether the actual problem has been solved.

Here is a practical way to start.

1. Choose an outcome that matters this week

Look for a recurring task you already spend attention on.

Perhaps you finish client calls with useful notes but no clear follow-up. Perhaps routine messages sit in your inbox because every reply starts from a blank page. Perhaps a website change needs checking before you approve it.

Choose one of those situations. Describe what should be different when the work is finished.

For example: “After this meeting, I want a short draft follow-up with the decisions, the next actions and anything that still needs an answer.”

Avoid beginning with a request to automate the whole business. A smaller task is easier to inspect and easier to correct.

2. Confirm what your account can actually do

OpenAI describes dots as ongoing agents that can use connected apps and work between conversations. Access is rolling out, and availability depends on the plan, region and account. Check the official setup guide before choosing a workflow that depends on a particular feature.

Begin with the information needed for your chosen task. That might be a document you provide or an app connection you deliberately authorize.

If the task requires a calendar or inbox, check that the right account is connected. Ask the assistant to confirm what it can read before relying on it.

Be careful with customer information. Use the minimum information needed, check your organization's rules and avoid copying sensitive material into a new tool simply because it makes the demonstration easier.

3. Explain the business context

An assistant needs to know what matters in this situation.

For a follow-up draft, that includes who will read it, what was agreed and what remains unresolved. For inbox sorting, it includes the senders and situations that should get your attention.

In a recent client setup, we discussed having the assistant ask the owner questions about how he manages his inbox. That matters because his system should reflect his work. Reusing my labels does not automatically give him my priorities.

Give the assistant a good example when you have one. A previous follow-up you actually sent is more useful than asking for a vaguely “professional” tone.

Correct the facts before polishing the language.

4. State the review boundary

Be explicit about the difference between preparing work and taking an external action.

Here is a starter request you can adapt:

“Use the meeting notes I provide to prepare a follow-up email for my review. Include confirmed decisions and next actions. Mark missing dates or owners as questions. Do not invent commitments or send the email. Return the draft with a short list of anything I need to check.”

That request specifies the input, the output and the stopping point.

If you later want the assistant to send a message, authorize that separately with the intended recipient and content clear. Some actions will still need additional confirmation or your direct involvement.

For early work, a reviewable draft is often a comfortable starting point. One client I work with is already using that approach for email: have the assistant prepare the message and keep it in drafts for his own check.

5. Test the result where it will be used

Read the follow-up against the original notes. Check names, dates, ownership and promises. See whether an unresolved question accidentally became a confident statement.

For a calendar task, inspect the actual event, including the date, time zone and participants.

For a website, open the page and follow the relevant customer path. For an app, use the interaction yourself.

I have found app work where the result did not match what I wanted and an input flow where the experience was too cumbersome. The useful test was the screen itself. A polished explanation could not settle those questions.

Ask for corrections that point to the problem: the field that is blocked, the step that asks too much or the slide that does not explain the decision.

6. Judge whether the task is worth repeating

Before expanding the setup, look at what happened.

Did the draft save you from starting over? Did the assistant capture the important decision? How much checking and correction did it need? Was the output easy to find?

If you want to measure time, record a comparable task done manually and include the review and correction time in the assisted version. Keep task completion, output quality and business results separate.

A draft produced quickly may still need too much rewriting. A website published successfully still needs working inquiry and booking paths. Tool usage alone does not answer either question.

7. Expand after the first workflow is dependable

Repeat the task enough to learn where it fails. Include an ambiguous input or an exception, rather than testing only the easiest example.

When it works reliably enough for your needs, consider an adjacent task. Meeting notes might lead to draft follow-up, then to reviewed task updates. Each added step should have a clear owner and an appropriate permission boundary.

You do not need an elaborate setup to learn whether this is useful. You need one real task, the right context and a result you can check.

Want help choosing the first task? Book a 30-minute consultation with me.