What Kind Of Business Should I Start?

What kind of business should I start is an easier question than which idea should I pick, and it comes first. The type follows from three things you can check today: what you can deliver on your own, how the buyer already pays for this job, and whether a platform hands you distribution. Answer those three honestly and you are down to one or two types before you have looked at a single idea.

Picking the idea itself is a separate filter, and it already has a home. The four-question method, meaning is the problem repeated, is it recent, does anyone pay for a workaround today, and can an outsider technically build the fix, is worked through in what business should I start. Run it after the type decision, because the type decides which evidence you need. A service needs one buyer with a deadline this month. A marketplace app needs a pain documented by at least 3 distinct authors and a host API that lets a third party perform the operation.

Key takeaways

• The type comes out of delivery capacity, existing buyer behaviour and available distribution, not out of how interesting the idea sounds.
• Services get paid first. Web scraping gigs carry 37,685 buyer reviews at a $55 starting price, which proves the work is bought, not that it scales.
• A productized service is the same job sold at fixed scope and price, and it is the only type that turns your own repetition into margin without code.
• Marketplace software rents distribution from a host. Shopify alone holds 298 published pain cards in our corpus.
• Extensions and automations sit between the two, with 8,558 scored Chrome feature gaps and 2,095 buyer reviews behind custom extension gigs.
• The evidence bar does not change with the type: at least 3 distinct authors describing the same problem.

Four types worth considering

Four types cover almost everything one person can start this year. They differ in who buys, how delivery scales, and how long you wait for the first payment. All four numbers below come from the paid demand catalogue, collected 2026-08-27, where every buyer review is a completed purchase.

A service you deliver yourself

The buyer is someone with a broken thing and a date. Delivery is your hours, so scaling means charging more per hour or working more of them, and there is no third option. The money is not in doubt: web scraping and data extraction has 213 gigs and 37,685 buyer reviews with a $55 starting price, and setting up or fixing Meta Pixel and the Conversions API has 110 gigs and 5,647 buyer reviews starting at $32. Those two clusters describe jobs that recur constantly and that nobody wants to learn. Start here when you need revenue before you need a product.

A productized service

Same work, fixed scope, fixed price, no discovery call. This type exists because the fifth time you do a job, you already know what varies and what does not. Building Excel dashboards and reports has 7 gigs and 3,622 buyer reviews at a $47 start, while a neighbouring cluster covering Excel and Google Sheets dashboards has 10 gigs, 993 buyer reviews and a $199 start. Setting up and customizing a Shopify store starts at $209 across 8 gigs and 1,325 buyer reviews. Airtable consulting has 4 gigs, 1,107 buyer reviews and an $89 start. Delivery scales through templates and subcontractors rather than software, and the price spread between $47 and $199 for the same category of dashboard work is where the margin decision sits.

Software inside a marketplace

The buyer is a customer of a host platform who arrived with the problem already in mind and a payment method on file. You build once and the host sells for you, which is the only type where distribution gets cheaper over time instead of more expensive. The wait is longer, and you inherit the host’s rules. Developing a Shopify app is a thin cluster in the catalogue, 1 gig and 126 buyer reviews at a $143 start, and that thinness is the point: the recurring revenue sits in the app, not in the build service. The model itself is covered in the micro SaaS guide.

A browser extension or an automation

This type is the shortest path from a service to a product, because the repetition shows up in your own invoices first. Building n8n automations and workflows has 8 gigs, 1,147 buyer reviews and a $71 start. Custom browser extension development has 45 gigs, 2,095 buyer reviews and a $114 start. We also scored 8,558 feature gaps mined from 1-star extension reviews, listed under Chrome opportunities, with accessibility at 1,643 gaps and privacy and security at 1,177. Distribution is the catch. A Chrome listing is a channel, but store search is a weaker channel than a marketplace app store attached to a paying customer base.

A table that matches type to your situation

If you are asking what kind of small business should I start with no capital and no staff, read the first two columns and ignore the rest for now. If you have delivery capacity already, the last column is the one that decides.

Type	What you need to start	How fast the first payment arrives	Where it stops scaling
Service you deliver	A skill and one buyer	Days. Web scraping gigs hold 37,685 buyer reviews at a $55 start	Your calendar. Revenue tracks hours
Productized service	The same job done five times, then a fixed scope and price	Weeks. Shopify store setup starts at $209 across 1,325 buyer reviews	Delivery people. Margin improves, headcount comes back
Software in a marketplace	A documented pain, a host API that allows the operation, a few months	Months. The host app store is the channel, and Shopify has 298 published pain cards	Platform rules and the host review queue
Extension or automation	One narrow repetitive task and a channel of your own	Weeks to months. Custom extension gigs hold 2,095 buyer reviews at $114	Discovery. Store search, not the product, becomes the limit

Read the third column as a cash flow constraint rather than a ranking. Someone with six months of savings can take the marketplace route. Someone who needs money in November cannot, and pretending otherwise is how people end up quitting a good idea in month three.

Why the marketplace type has an unfair advantage

A host platform solves the problem that takes longest to solve from zero, which is being found by someone who already has the pain. Our corpus covers 75 software marketplaces, and complaint volume is a rough proxy for how much unpaid work sits inside each one. Shopify leads with 298 published pain cards, then Autodesk at 254, Grafana at 163, Cursor at 160, SketchUp at 157, OpenAI at 156, Asana at 134 and Dropbox at 121.

Two things make those numbers different from ordinary market size figures. The complaints are attached to a specific product with a specific customer list, so the buyer is a role you can name rather than a segment. And the same place that produced the complaint also sells your fix, which collapses research and distribution into one surface. Per marketplace views such as Shopify show which pains cluster where.

The cost is real. You inherit the host’s API limits, its review queue and its pricing conventions, and a feature the platform decides to ship itself can end your product in one release note. That risk is priced into the type, not into the idea.

The evidence check that applies to every type

The check does not change when the type changes. A pain card is published only after at least 3 distinct authors describe the same problem, and 804 of our 3,261 published cards carry an explicit monetization signal, meaning someone in the evidence says they pay, would pay, or already switched to a paid tool. That last subset is where the software types belong, because a subscription needs a buyer who has already decided the problem costs money.

Evidence tells you which problems are real. It does not tell you what kind of business to start on its own, which is why the type decision comes first and the full four-question filter in what business should I start comes second. How the clustering, author counts and monetization flags are produced is documented on the how it works page, and the field definitions are in the glossary.

Two worked type decisions

Two cards from the corpus, similar evidence strength, different answers on type.

Card 4136 covers profit reconciliation on Shopify, hidden COGS and shipping. It has 43 posts from 41 distinct authors, a score of 0.986, and no direct competitor listed. 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). An empty competitor column means the market is unserved or unservable, and you do not yet know which. That points at a productized service first: a fixed price monthly reconciliation, done by hand for ten merchants, tells you in six weeks whether the numbers can even be pulled reliably. If they can, you have a spec and ten paying references for the software version.

Card 4828 covers scheduling sold-out modifiers for food shops on Square. It has 55 posts from 48 distinct authors and is typed as an add-on opportunity. At a bagel shop, certain flavors sell out and staff “manually track on a paper note” (source). No service version of this exists, because the job repeats every morning at opening and nobody will pay a freelancer to log in daily. A job that recurs on a daily cycle inside one product is a software job, and it belongs inside the host platform where the counter staff already work. Same evidence bar, opposite type, and the deciding factor is the frequency of the job rather than the size of the complaint.

So, what kind of business should I start

Take the three checks in order. If you can deliver the work yourself and need money soon, start with the service and productize it once the scope stops changing. If the job repeats on a schedule inside somebody else’s product, build the app and take the host’s distribution. If the repetition lives in your own workflow rather than a buyer’s, an extension or an automation is the cheaper first version. The idea comes after that, and the shortlist of concrete candidates sits in 12 unique business ideas and profitable business ideas.

You can test a type against real evidence rather than memory. The validation workflow runs a candidate through the same author, monetization and competitor checks against 3,261 published pain cards, and the pain card index lets you sort by monetization signal to see which problems already have money attached.