Which SaaS should you build? How to find an idea people will actually pay for

Almost every "SaaS ideas" article is a list. Twenty niches, a tech stack for each, a pricing model. They are enjoyable to read and they help almost nobody, because the list is not where the difficulty lives.

The global SaaS market reached roughly $408 billion in 2025 and is projected near $465 billion in 2026. There are more than 42,000 SaaS companies already competing. Ideas are not scarce. What is scarce is an idea with a buyer already attached to it.

Start with the failure data

Finding	Source
42% of startups fail because there was no market need	CB Insights post-mortem analysis
92% of SaaS startups never reach $1M ARR within three years	2026 industry data
Only about 40% run formal validation before launch	2026 startup research
Roughly 45% of failures occur 18–24 months after launch	2026 startup research

Read those together and a pattern appears. The problem is rarely a bad idea. It is a reasonable idea that nobody confirmed anyone would pay for, discovered eighteen months and a lot of money later.

The three questions that filter almost everything

If you cannot answer all three before building, you are betting on discovery rather than validating a hypothesis.

1. Who exactly has this problem? Not "small businesses". A job title, an industry and a company size. "Practice managers at dental clinics with two to six chairs."
2. What are they doing about it right now? There is always a current solution, even if it is a spreadsheet, a WhatsApp group or an assistant doing it manually.
3. What is that costing them? In money or hours. If the answer is "not much", you have found an annoyance, not a problem.

That second question is the one that separates real opportunities from imagined ones. If nobody is doing anything about it, that is usually because it does not hurt enough to act on.

Where good ideas actually come from

1. Work you have already done

The most reliable source is a job you have held. You know the workflow, you know who signs off on purchases, and you know which parts are genuinely painful rather than theoretically inefficient. That knowledge takes years to acquire and cannot be researched quickly.

If you have spent five years in logistics, your unfair advantage is in logistics, not in the AI productivity tool everyone is currently discussing.

2. Workarounds you can see people paying for

Look for spreadsheets that have grown into systems. A shared sheet with twelve tabs, colour-coded, that three people maintain, is a product waiting to exist. Somebody is already paying for it in hours.

3. Frustration in a crowded market

Counter-intuitively, a market with strong competitors and unhappy customers is often more promising than a market with no competitors at all. No competition usually means no demand. Frustrated customers in a busy market means the incumbents are failing at something specific — and specific is something you can build against.

Read the one and two-star reviews of established tools on G2 and Capterra. People describe exactly what is wrong, in their own words, for free.

4. A niche too small for the incumbents

The rise of micro-SaaS reflects a real gap: vertical products serving one industry's specific workflow. A general-purpose tool serving everybody serves nobody particularly well, and a large company cannot justify building for a market of 4,000 businesses. You can.

Signals that an idea is worth pursuing

• Someone offers to pay before it exists. The strongest signal there is.
• People describe the problem unprompted. If you have to explain why it is a problem, it is not one.
• An expensive manual workaround exists. Time is money and somebody is already spending it.
• The buyer has a budget and authority. Selling to someone who must ask permission triples your sales cycle.
• You can reach them. If your target customer is not in any community, list, event or search you can access, distribution will kill you regardless of the product.

Signals to walk away

• "Everyone could use this." Nobody markets to everyone. This usually means you have not identified the buyer.
• Only friends and family like it. They are being kind, not making a purchase decision.
• Interest without money. "That sounds useful" and "I'd sign up" are not validation. Only a payment or a demonstrated willingness to pay counts.
• The problem is real but occasional. Software people use twice a year does not sustain a subscription.
• You need thousands of users to make it work. Consumer scale is a different and much harder game than B2B.

The categories with genuine pull right now

Stated as directions rather than as ideas to copy, because the specific opportunity is always inside a niche you understand.

Direction	Why there is demand
Vertical tools for regulated industries	General tools cannot handle compliance specifics; buyers pay premiums for tools that do
Workflow tools replacing spreadsheet systems	The workaround is visible, quantifiable and already costing hours
Compliance and reporting for new regulation	New rules create deadlines, and deadlines create budgets
Integration between tools a specific industry uses	Nobody builds the boring connector, and everybody needs it
Operational tools for trades and local services	Historically underserved, increasingly comfortable with software

Notice that none of these is a product. They are places to look. The product is whatever specific thing the people in that space keep complaining about.

What to do this week

1. Write down the industry you know best. Not the one that sounds most exciting.
2. List five things that were genuinely annoying about that work. Not inefficiencies you have read about — things that made your own week worse.
3. For each, ask what people do about it now and roughly what that costs.
4. Pick the one with the most expensive workaround.
5. Talk to ten people who have that problem before writing any code. Not to pitch — to ask what they currently do.

That last step is where most founders skip ahead, and it is the step the failure data keeps pointing at. Ten conversations cost you a fortnight. Building the wrong thing costs somewhere between $50,000 and $150,000 and twelve to eighteen months.

The good news in 2026 is that failing fast has become genuinely cheap. AI-assisted development has compressed build timelines significantly, which means the cost of testing a hypothesis has fallen faster than the cost of being wrong. That only helps if you test.