What Business Should I Start?

The short answer: start from a problem you already have access to, then check two things before you write any code. First, is the problem documented by people other than you. Second, does anyone already pay to have it solved. If you are asking what business should I start, that sequence will decide the outcome more than the idea you happen to like this week. Access gets you a candidate. Documentation and money tell you whether the candidate survives.

Most people run this backwards. They pick a category that sounds promising, then hunt for a problem inside it, then look for evidence that the problem is real. Strip the wording and what business do I start becomes a sorting problem rather than a creativity problem: you already have more candidates than you think, and what you lack is a filter strict enough to kill most of them quickly. This page is the filter, not a list of ideas. If you want candidates, the sibling articles linked further down carry them.

Key takeaways

• Start from a problem you can already see, then verify it outside your own experience.
• Our publication bar for a recurring pain is evidence from at least 3 distinct authors, not one loud review.
• A real complaint is not always buildable. Platform limits can block an outsider even when the pain is well documented.
• Money already moving through freelancers is the cheapest proof of demand available: 3,622 buyer reviews sit behind one cluster of Excel dashboard gigs.
• Of 3,261 published pain cards, 804 carry an explicit monetization signal.
• Competitor coverage changes your decision more than complaint volume does.

Start from a problem you can already see

You have three kinds of unfair advantage, and each points at a different set of problems.

The first is the tool you use every day. If you have spent two years inside Asana, Shopify admin, QuickBooks or Grafana, you know where the workflow breaks, which workaround everyone uses, and which support answer is a polite no. One Asana user described the failure mode plainly: “Every campaign being its own project with Planning, Ops and Marketing sections works fine until three different project managers each have their own interpretation of what goes where.” (source) If you have lived that, you can tell in ten seconds whether a proposed fix is real.

The second is the industry you worked in. Domain access is not about software. It is about knowing who signs the check, what the budget line is called, and which month of the year they have time to talk to you.

The third is a platform where you already sell or already have an audience. Distribution takes longest to build from zero, so starting where you already have it removes the hardest unknown. This is also why marketplace pain is more actionable than generic pain: an app store is both a source of complaints and a channel. Our corpus covers 75 software marketplaces, and Shopify leads with 298 published pain cards, followed by Autodesk at 254 and Grafana at 163.

Pick the advantage you actually have. There is no answer to what business is good to start in the abstract, because the same opportunity is excellent for someone with a mailing list of 4,000 ecommerce operators and worthless for someone with none.

Four questions that eliminate most ideas

Once you have a candidate problem, run it through four questions in this order. Most candidates die at question one or three, which is the point.

Is the problem repeated? One angry review proves a person had a bad day. We treat a pain as recurring only once it has evidence from at least 3 distinct authors, and similar complaints get clustered into one card instead of counted as separate opportunities. Fifty posts about the same broken workflow should strengthen one finding, not create fifty ideas. The clustering and evidence checks are described on our how it works page, and pain card, opportunity type and monetization signal are defined in the glossary.

Is it recent? A complaint from 2021 may already be fixed, and a fixed problem is a wasted month. The Shopify chargeback and pre-order stock cluster showed interest up 231% in the last week; a QuickBooks Online payroll cluster showed 305%. Momentum tells you whether you are looking at a stable annoyance or a wound that just opened.

Does anyone pay for a workaround today? If people currently solve the problem with a spreadsheet, a virtual assistant or a freelancer, that spend is your price signal. This question has the highest kill rate and the cheapest answer, so do not skip it.

Can an outsider legally and technically build the fix? This is where good ideas quietly fail. Asana card 296 collects a well-documented complaint from eleven distinct authors: “there does not appear to be an easy way to apply that new field retroactively to projects that already exist … there is no easy way to see which projects it has not been added to” (source). It is still a poor build target for a third party, because applying a custom field retroactively across every existing project is a platform capability, not an app capability. A real complaint is not automatically a buildable one, and the fastest way to find out is to read the API docs before anything else.

Check willingness to pay before you build

Willingness to pay is not a survey question. People say yes to surveys. The stronger signal is money that has already moved, and freelance marketplaces make that visible: every buyer review is a completed, paid transaction for a specific job.

Here is what that looks like in our paid demand catalogue, collected 2026-08-27. Buyer reviews count reviews across the gigs in each cluster, and prices are starting prices in USD.

Gig cluster	Gigs	Buyer reviews	Starting price
Build Excel dashboards and reports	7	3,622	$47
Develop custom browser extensions	45	2,095	$114
Build Gantt charts and project plans	8	1,532	$32
Set up and customize a Shopify store	8	1,325	$209
Build n8n automations and workflows	8	1,147	$71
Airtable expert and consultant	4	1,107	$89

Read those rows as evidence of a job people hire out, not as a business plan. Three thousand six hundred buyer reviews on Excel dashboard gigs mean reporting is still unsolved for a large group of buyers with budget and no patience. The Shopify setup row matters more than its review count suggests, because $209 is a starting price, which puts it in a different bracket than the $32 Gantt work.

Paid demand also answers what small business should I start when you do not want to build software at all. Every row above is a service someone runs today, and a service is the fastest way to learn a problem well enough to productize it later.

A worked example

Two Shopify pain cards, similar surface, opposite conclusions.

Card 3705 covers chargebacks and pre-order stock. It has 118 posts from 107 distinct authors, a score of 0.997, explicit monetization signals and interest up 231% in the last week. One merchant wrote: “I didn’t fulfill any high-risk orders after asking on this subreddit; the fear of chargebacks just wasn’t worth it for me personally, and I ended up refunding them all. I’ve lost a lot of money from it” (source). It also lists six paid or freemium competitors already: OrderPoint, Stockcast, Skucast, Purchase Order Hero and Stockie among them.

Card 4136 covers profit reconciliation, hidden COGS and shipping. It has 43 posts from 41 distinct authors and a score of 0.986. A former operator described the damage: “I used to manage inventory for a small brand and the COGS thing alone gave us completely wrong numbers for months before someone caught it” (source). No direct competitor is listed on that card.

Same marketplace, same evidence bar, different answer. If you are a solo builder with no distribution, card 3705 asks you to win attention against five or six incumbents who already rank in the app store, and complaint volume will not carry you there. Card 4136 has roughly a third of the authors and no listed competitor, which means the market is either unserved or unservable, and finding out which one costs a week of interviews rather than a quarter of engineering. For one person choosing between the two, the competitor column decides which business is good to start, not the author count. Give yourself an existing audience of Shopify merchants and 3705 becomes reasonable again, because your distribution cancels their head start. The underlying evidence sits in the Shopify marketplace view and the pain card index.

Where to look if you have no idea at all

If none of the three access routes apply yet, you are not stuck, you are just at the research stage. Three paths work.

Read the recurring complaint categories and pick one you can stand working in for two years. Performance and reliability leads our corpus with 642 cards, followed by billing and pricing at 601 and integrations and API at 496. Our analysis of the complaint corpus covers what those categories can and cannot tell you.

Read candidate lists, then apply the four questions above to each one. Our companion article on 12 unique business ideas gives evidence-backed candidates, and new business ideas covers what has changed recently in the complaint data, which is where the least crowded openings usually sit. If you want to know what businesses are good to start with software margins and one founder, start with the micro SaaS guide.

Start with the service version first. Sell the fix by hand, learn the workflow, then automate the part you repeat.

Common wrong answers

Chasing a category because it is large. Accessibility is the biggest Chrome extension gap category in our data at 1,643 scored feature gaps, and that number tells you nothing about whether you can reach those users or charge them. Category size measures noise, not opportunity.

Copying a competitor with no distribution plan. Cloning a product that already ranks means competing on the axis where you are weakest. If your only differentiator is that your version is newer, you are asking buyers to take a risk for no reason.

Building for a market you cannot reach. The question of what company should I start is really a question about who will answer your first email. If you cannot name twenty specific people or forums where your buyers already gather, the idea is not ready yet.

So, what business should I start

Take the problem you already have access to. Confirm at least three independent people describe it, check that the complaints are recent, find out whether anyone pays for a workaround today, and read the platform’s API limits before you fall in love with the fix. If any of those four checks fails, drop the candidate and move to the next one. That is the whole method, and it runs faster than it sounds once you stop treating idea generation as the bottleneck.

You can run this against our evidence instead of your memory. The idea validation workflow walks a candidate through the same checks against 3,261 published pain cards, and pricing covers what access to the full corpus costs. Once you know which problem you are taking on, what kind of business should I start covers the next decision: whether it becomes a service, a productized package, a marketplace app or an extension.