How to Pick a Business Idea Worth Testing for 2030

On this page

1. Start with a list you can actually compare
2. Run each idea through a free roast first
3. Describe every idea the same way
4. Use a simulated test to rank the shortlist
5. Read the results for patterns, not winners
6. Turn the top two or three into real-customer checks
7. Your next step

You’ve got a list of business ideas, and one of them keeps pulling at you. Go with it? That’s the usual move, and it’s a shaky way to choose. The idea you love is often just the one you’ve carried around the longest. That says nothing about whether a customer is waiting for it.

A simulated test can rank a short list quickly and cheaply. Treat the result as a hint about where to look first. The participants are AI personas, not real people, so the numbers tell you which ideas deserve your next hour. They don’t prove anyone will pay.

Here’s what this guide covers: building a list you can actually compare, describing every idea the same way, reading the results for patterns, and taking the best two or three to real customers. No income is promised. Real customers settle that.

Start with a list you can actually compare

Write down 5 to 8 ideas. Rough is fine. One line each is enough for now, and you’ll probably feel a bit silly about two of them. Keep them anyway.

Now cut. If you couldn’t start an idea with the skills, time and money you have today, take it off the list. Not forever, just for this round. An idea that needs a year of savings and a team you don’t have isn’t something you can check this month.

Then look at where each idea came from. The better ones usually trace back to something you’ve already seen in your own work or your own town. Maybe a shop owner you know gets asked the same question all week. Maybe a neighbor keeps losing an hour to the same chore. You can check that by asking a few people. A hunch about what 2030 will look like is another matter. Nobody can check it, so you end up arguing with yourself and never settling anything.

Last, narrow each idea to one customer and one problem. “Something with AI for everyone” can’t be tested, because you don’t know who to ask. “A scheduling tool for solo piano teachers who lose lessons to no-shows” can.

Imagine a former bookkeeper with six ideas: a bookkeeping service for dog groomers, a receipt-scanning app, a meal-prep subscription, a local tutoring marketplace, a template shop and a repair-cafe booking tool. She doesn’t cut any yet. She just sharpens the fuzzy ones, so “a template shop” turns into “spreadsheet templates for freelancers who track their own expenses.” Now she has six things she can set side by side.

Run each idea through a free roast first

Take your list and write two sentences for each idea: who it’s for, and what problem it solves. Then paste each one into the free Idea Roast. It takes a minute per idea and costs nothing.

You get back a few potential weaknesses, what looks promising, open questions, and the riskiest assumption. Start with that last one.

Read the riskiest assumption for each idea and ask one thing: can I check this myself, with real people, in a few weeks? If not, drop the idea for now. “A big partner has to say yes” is the classic case. You can email them, but you can’t make them answer, and meanwhile the rest of your list sits there untouched. Ideas that hang on a handful of people you can’t reach make poor first-round candidates.

Say the former bookkeeper’s roast flags two of her six. The tutoring marketplace needs enough tutors and enough parents at the same moment. The repair-cafe booking tool needs community centers to adopt it before anyone else can use it. Neither assumption is impossible. Both are hard to test cheaply. She sets them aside and keeps four.

Keep in mind what you’re reading, though. It’s one AI’s take on a short description, nothing more. It finds holes but doesn’t give a score, and a gentle roast doesn’t make an idea good. So don’t rank your list by how harsh or kind the feedback sounded.

If a roast keeps coming back with the same open question, look at your description first. It’s usually vague, so rewrite it before you move on.

Describe every idea the same way

Now you’ve got four survivors. Before you test any of them, write them up so they can fairly compete.

Use the same length, the same format and the same price logic for every idea. If one description runs six sentences and another runs two, you’re testing your writing, not the ideas. The idea you happen to be excited about will get the better paragraph, and it’ll win for the wrong reason.

A simple template keeps you honest:

• What it is, in one sentence.
• Who it’s for, in one sentence.
• What it costs, and how it’s charged (once, monthly, per job).

That’s it. Skip hype words like “revolutionary” or “all-in-one.” They add nothing a participant can react to.

Pricing is where people slip. If the bookkeeping service is $150 a month and the template shop is $12 once, that’s fine, as long as each price is what you’d realistically charge. A price you’ve quietly lowered to make an idea look good tells you nothing.

Then name the audience specifically: age range, job, budget level, where they shop. “Small business owners” is too loose. “Owners of one-person pet grooming shops, 28 to 55, tight budgets, who find suppliers through social media” gives the test something to work with. For a fuller walkthrough, see how to describe your audience for an AI test.

Say the former bookkeeper writes this for her receipt-scanning app:

A phone app that scans receipts and sorts them into tax categories. For freelancers aged 25 to 50 who track their own expenses. $8 a month.

It’s short and easy to compare with the others. She writes the other three the same way.

Use a simulated test to rank the shortlist

Now the shortlist goes into a simulated test. Pick the idea type that fits each one. The bookkeeper’s receipt-scanning app is an app or software test, so participants answer Would try, Maybe or Wouldn’t try. Her expense templates for freelancers fit better as a startup idea, where the answers are Interested, Maybe or Not interested. Her meal-prep subscription would run as a physical product, and a bookkeeping service for groomers as a service.

Her answer scales will differ a bit by type, and that’s fine. What matters is that each idea gets the right framing and the same audience settings.

Your first 1,000-participant test is free if you’re eligible (one free test per person). After that, 1,000 participants costs $19, so running four ideas is a small spend. A test takes about 10 to 12 minutes. You can start at the free test, or look at a sample report first.

Keep in mind who’s answering. The participants are AI personas, not real people. You’ll see the share of each answer overall and group by group, plus short reasons for each verdict.

When you compare ideas, use the same audience for all of them. Otherwise you’re comparing audiences, not ideas.

Then be strict about what counts as a difference. Reruns give somewhat different numbers, so if two ideas land within a few points of each other, call it a tie. Don’t crown a winner over a gap that a second run could flip.

A tie is useful too. It tells you the simulation can’t separate them, so pick based on what you could test with real people soonest.

Read the results for patterns, not winners

Once the four reports are in, resist the urge to scan for the biggest “Interested” number and stop there. The overall share hides the more useful part.

Open the group results. Which groups lean in, and which don’t? An idea that one group loves can be a better bet than one everybody mildly likes. Say the bookkeeper’s bookkeeping service for groomers gets a modest overall score, but one group of tight-budget solo shop owners is clearly keen. That’s a customer you can actually go find. A flat, lukewarm result across every group gives you nobody to call.

Then read the reasons behind “Not interested” and “Wouldn’t try”. Each one is short, so it’s quick work. Look for the same objection showing up again and again: too pricey for what it does, I already use a spreadsheet, I don’t trust an app with my receipts. Write those down. They’re your questions for real people, almost word for word.

Be careful with the nice reasons too. Repeated praise can be the AI repeating a cliché, so check whether the reasons name something specific.

Keep the limits in front of you:

• Personas don’t spend real money.
• They don’t see your competitors or your website.
• They can be more generous than real buyers.

So a strong result means “look here first.” It doesn’t mean “build this.”

If your top two sit within a few points of each other, they’re tied. Let the group patterns and the objections break the tie, not the headline number.

Turn the top two or three into real-customer checks

Say the bookkeeper’s top three are the groomer bookkeeping service, the receipt-scanning app and the template shop. Each one now gets a small check with real people. Not a build. A check.

Pick the lightest test that fits the idea:

• Conversations. Talk to 5 to 10 people who match the audience. For the groomer service, that means actual groomers. Ask how they handle their books now and what it costs them, and use the objections from the simulated reports as your questions. If you need help finding people, see how to find people for customer interviews.
• A landing page. Good for the app and the templates. One page, one promise, one button. How to write a landing page that tests an idea covers the details.
• A pre-sale offer. The strongest signal of the three, because someone has to hand over money. See how to run a pre-sale.

Now the part people skip. Write your drop rule before you start. Something like:

If none of ten groomers will book a call or put down a deposit within two weeks, I drop this one.

Without that line, you’ll talk yourself into one more month, and then another. I’ve done it. A lukewarm result always looks like “almost there” once you’re attached to the idea.

Watch out for kind words, too. “Great idea, I’d totally use that” costs the person nothing to say, so it tells you nothing. A booked call, a deposit or a signed-up customer is different. Someone spent time or money.

Be honest about what the simulation gave you. It picked where to look first. It didn’t predict a single sale, and nothing in this process predicts income. Real customers decide whether any of these ideas deserves your time.

Your next step

Start with the cheap part, and do it today. Write a list of 5 to 8 ideas. Cross off the ones you couldn’t begin with what you have right now. Then paste a two-sentence description of each into the Idea Roast. If you can’t think of a way to test an idea’s riskiest assumption, drop it.

The survivors get described the same way and run against the same audience. Your first 1,000-participant test is free if you’re eligible, so you can rank a shortlist before you’ve spent much. Want to see what the output looks like first? There’s a sample report. When you’re ready, start the free test.

Keep the result in proportion. Those participants are AI personas, and the ranking only tells you where to look first.

The part that counts comes next. Choose one idea and get a real conversation on the calendar, this week if you can. A single call with someone who fits your audience will teach you more than another round of list-making.

Written by the SellTo.us Team. Examples in this article are hypothetical. SellTo.us results come from AI-simulated participants, not real people, and don’t predict sales or guarantee success.