Start Small

Run your first AI experiment.

You don't need an AI strategy, an implementation team, or a perfectly defined use case. Pick one real task, give yourself 15 to 30 minutes, and see what happens. Your first goal isn't to automate anything. It's to learn whether AI can help.

Step 1: Pick the right first problem

A good first experiment is:

Low Risk

Choose a task where a mistake won't create a major business, customer, safety, financial, legal, or compliance problem.

Easy to Verify

Choose something you understand well enough to recognize whether the AI's answer is good.

Actually Useful

Solve a real problem. Don't invent an exercise just so you can say you tried AI.

Small Enough to Finish

Think 15–30 minutes, not a six-month implementation project.

A Little Annoying

Repetitive, time-consuming, tedious, or something you've been putting off is often a great place to start.

Good First Experiments

Try using AI to:

Draft an email from rough notes
Summarize meeting notes and identify action items
Turn a procedure into a training checklist
Explain an unfamiliar Excel formula
Brainstorm interview questions
Compare two non-confidential documents
Help troubleshoot an error message
Analyze an appropriate, anonymized spreadsheet
Rewrite technical instructions for a different audience
Create the first draft of a report or presentation
Organize scattered notes into a usable document

Start with something useful — but safe enough to experiment with.

⚠️ Before you experiment: Protect your information.

Your first experiment should also be a safe one. Before entering anything into an AI tool, check:

Your organization's AI and information-security policies
Whether the tool is approved for business use, and what it does with what you submit
Whether the content includes customer, supplier, employee, or personal information
Whether it includes proprietary or confidential information

While you're learning, use sample, anonymized, or non-confidential information. And never assume that because a tool lets you upload something, you should.

Step 2 - Define "Better"

Before you open your AI tool, take a minute to write down your starting point:

The task: _______________________________

How I do it today: _______________________

Approximate time: ________________________

What I want AI to help with: ______________

How I'll know it helped: ___________________

You're not trying to find out whether AI is impressive.

You're trying to find out whether it makes the work better.

Step 3: Give AI the job

You don't need to be a prompt-engineering expert. Just give the AI four basics:

WHO

Who are you? What is your role, industry, or situation? Is there a particular role or perspective you want the AI to take?

WHAT

What are you trying to accomplish?

CONTEXT

What does the AI need to know to understand the situation?

OUTPUT

What would a useful answer look like?

Want to go deeper? Prompt Better walks through five steps to more useful answers.

Copy-and-paste prompt: Get Started

“I work in [ROLE / INDUSTRY] and I need help with [TASK].

My goal is [DESIRED OUTCOME].

Here is the relevant context: [CONTEXT]

I would like you to [SPECIFIC REQUEST].

Please give me the result as [FORMAT — email, table, checklist, steps, summary, outline, etc.].

If you don't have enough information to do this well, ask me questions before answering.”

Step 4: Don't accept the first answer

AI works best as a conversation. If the first response isn't quite right, don't start over. Tell it what isn't working. Try:

“That's too technical. Rewrite it for someone without technical experience.”

“You're making an assumption that isn't correct. Here's how our process actually works…”

“I like this direction, but make it shorter and more actionable.”

“What information am I missing that would help you give me a better answer?”

“Give me three different approaches.”

“Challenge your recommendation. What could go wrong?”

💡 Pro tip: You don't always need a better prompt

Sometimes you just need another sentence. Correct it, add context, tell it what you like and don't like, or ask a follow-up question.

Have a conversation.

Step 5: Verify before you trust

AI can be confidently wrong. It can produce an answer that sounds polished, professional, and completely believable, and still be wrong. Before using the result, check:

Check the Facts

Are names, dates, numbers, calculations, quotations, and factual claims correct?

Check the Source

If the answer depends on a policy, regulation, customer requirement, specification, contract, or current information, verify it against an authoritative source.

Check the Logic

Does the recommendation actually make sense?

Check the Output

Is anything important missing? Did the AI misunderstand part of the assignment?

Would you be comfortable putting your name on it?

AI can help do the work.

It doesn't transfer responsibility for the work.

Step 6: Measure what happened

Don't just ask, "Was that cool?" Ask, "Was that better?" Compare the result with how you did the task before:

TIME
Did it take less time?

STEPS
Did it simplify the process?

QUALITY
Was the finished product better?

FRUSTRATION
Was the task easier to complete?

ERRORS & REWORK
Did it reduce mistakes — or create additional work?

CAPABILITY
Did it help you accomplish something you couldn't easily do before?

THINKING
Did it help you see the problem differently?

And Be Willing to Say…“Nope. This isn't better.”

Not every problem needs AI. Sometimes it creates more work than it saves, another tool is a better fit, the current process is already pretty good, or the task simply needs human judgment.

That's still a successful experiment. You learned something.

Step 7: Decide what happens next

After your experiment, choose one of three paths.

Keep it

It worked. Keep using AI for this task. Before you close the conversation, ask:

"Based on this conversation and the final result, create a reusable prompt I can save and use the next time I do this task."

Save it somewhere your team can find it.

Improve it

AI helped, but the process could still be better.

Ask:

“What could I change about my instructions, the information I provide, or my workflow to make this more effective next time?”

Then experiment again.

Connect it

You've done this AI-assisted task several times and it works. Now you may be ready to go further. Ask:

"I do this task regularly. What options exist to automate or partially automate it using systems we already have? Ask me about our current technology before recommending new software."

This is where AI may start connecting with tools you already use: Microsoft 365, SharePoint, Power Automate, your ERP, BI dashboards and alerts, APIs, or AI agents.

Automate what you've learned — not what you haven't understood yet.

Your first AI experiment doesn't have to be impressive. It just has to teach you something.

Pick one problem.
Try one experiment.
Verify the result.
Measure whether it helped.
Decide what to do next.

Then do it again.

Start small. Learn fast. Build from what works.

And most importantly…

Stay curious.

Next: Prompt Better

Page updated

Google Sites

Report abuse