How to Start a SaaS Business with AI in 2026 and Make It Pay

What a SaaS is in 2026

The acronym stands for software as a service: your customers pay as long as they use it, usually every month, sometimes by buying credits in advance that they use up. They open it in a browser or in a phone app. You host it: you keep it running on a server, a computer online that answers your users. You also fix it and keep improving it.

Google Workspace (Gmail, Docs) and Slack, the team messaging app, are widely used examples. The SaaS businesses in this guide are far smaller: one tool for one line of work, often sold by one person.

It's often confused with renting servers, the part of cloud computing where companies such as Amazon Web Services rent computers to developers so their software can run. A SaaS puts finished software in its users' hands.

As long as you're the only one using your software, it's a tool. Paying users and their data change what you're responsible for: each customer's account has to be kept apart from the others, their data protected, problems answered, payments taken and refunded, and tax rules followed that didn't concern you before.

For people to keep paying, the problem has to keep coming back. Ben Bozzay, a software developer, first sold a paid membership with guides that showed adults how to lock down their own phones and computers to curb their internet use.

Most members had their setup done within a month, and then, as he puts it, "why do they need to keep paying for the membership?" So he built a website-blocking service that works every day and sold it by subscription: his customers kept paying every month.

A SaaS isn't passive income. Zigpoll, survey software for online stores, brought in about $142,000 a month in subscriptions in September 2026. Jason Zigelbaum, who runs it alone, still answers every support email himself: "I don't like the way AI writes emails."

His growth came from small improvements, made again and again, "growth by a thousand cuts" in his words: to onboarding (the first days a new customer spends in the product), to prices, and to whatever made customers leave. A SaaS can serve many more customers without much more work, but it doesn't run on its own.

How to start a SaaS company on your own: the steps, in order

Here is the whole path, from the idea to charging every month. Each step has its proof, the sign that it's done. The proof that decides whether to build is someone paying before your software exists.

Two founders here built first and sold within weeks, because the demand was visible before any code. Denis Yurchak spent a weekend building Yadaphone, an app for calling phones abroad from a browser, once Microsoft had announced that Skype, used for such calls, would close, leaving its users looking for another way to call abroad. Hasaam Bhatti saw that the members of a coaching program were looking for tools. He spent 48 hours on his after offering the program's owners a partnership if they liked it.

Others who built first waited months for their first payment. Vadim, who sells an app that cuts long videos into short clips, built for four months, and the founder of Copyboost, which scores marketing texts, for six. Without a sign as strong as Yurchak's or Bhatti's, take pre-orders first.

Choosing a problem, talking with the people who have it, getting set up to sell, taking their pre-orders, launching and finding customers take no code at all. Charging every month takes four requests to your agent, each one given in this guide. A first version you can show, often called an MVP (minimum viable product), takes days. One that people pay for took the founders here from two days to six months. Finding customers starts with your first conversations and never stops.

Finding an idea that sells

Among the founders here, the ideas that sold came from problems seen up close. Two examples:

• Freddy Galindo Jr. built Sprint Suite, software that now serves some of Australia's largest mining contractors, six years after he started it (Revenue Leaders podcast, June 2026). Mining contractors are firms that send mechanics, called fitters, and other workers to mine sites. There they maintain the big machines of mining companies such as Rio Tinto and BHP. His family ran one, and managed its work in manila folders and spreadsheets.

  When he looked for software to run that company, nothing was made for mining contractors. The general tools his family's company tried forced it "to change our business to suit the software which doesn't really work in mining."

• Hasaam Bhatti, the founder from the opening, sold his own products on Amazon under two brands, alongside a corporate job. Before his tool for Amazon sellers, he had tried to build ten or twelve software products, from AI video tools to job-automation apps. None of those first products was ever finished and released, "because I was building for audiences I didn't belong to": he didn't know those people's problems from the inside.

  He had the Amazon sellers' problem himself. For each item he considered selling on Amazon, he spent 20 to 30 hours of research, copying data into Google Sheets: "I was frustrated, and I knew exactly what was broken. So I built the tool I needed" (in a post on Indie Hackers, a site where founders share their numbers, June 2026). His tool does that product research for Amazon sellers.

You can also start from a problem in your own daily work, without waiting for anyone else to point it out. With my own products, ReadyToPost and then contenttosocial, both projects of this site, I work this way, though not everyone agrees with the method: you know a problem you have every day, and you can check every day whether your solution fixes it.

My first SaaS, ReadyToPost, writes and publishes a brand's social media posts, a job I needed done myself. My second, contenttosocial, which turns articles into social media posts, came from the way I now work, almost entirely through AI agents over MCP, a standard way for an agent to use a piece of software for you.

I needed software I could run entirely that way, and other people who work like me need it too, so I designed contenttosocial for agents from the start.

Five signs that an idea, and the audience you pick for it, will sell:

• Your customer needs it now. The right problem is a sharp pain: your software must pay off from the first use, enough for your customer to feel it.

  Some tools might help one day, but your customer doesn't quite believe in them, because the result is distant or uncertain: those tools stay nice-to-haves, put off for later. Customers take a tool they're stuck without today right away.

• The problem comes back often enough to be worth paying for every month. A subscription is paid every month, so the problem has to come back at least that often, ideally every week. A problem people solve once and then forget stops them paying.

• Competitors already exist. In October 2026, Links brought in $31,500 a month, though a much bigger rival came first. Around 2020, Charlie Clark, a software engineer, gave himself two weeks to build Links as an alternative to Linktree, a service creators use to gather all their links on one page, linked from their social media profiles. Linktree was already "the big one". He found its pages didn't look good and made his look better.

  Competitors prove that people already pay to solve the problem.

• Your customer already spends money on the problem, ideally their business's money. Vadim, 19, sells VGola, which cuts long videos into short clips with AI, to people who earn a living from those clips: "clippers" paid to cut other people's videos, content creators, agencies. They already paid for this work, in editors they hired or in other clipping tools. For them, the subscription is a work expense that earns its keep, which makes it easier to pay.

• You know who you're talking to. With ReadyToPost, I launched without choosing a precise audience, and that was a mistake: with no target, I didn't know what message to write, whom to contact or how.

  I aimed contenttosocial only at people who work in SEO and GEO, the work of getting websites to show up in Google and in AI assistants' answers. These experts write articles for their clients. With contenttosocial, they also sell those clients social media posts drawn from the same articles.

  Each post carries the expert's own brand, and the client can see and approve it before it goes out. The expert bills the client for these posts every month. I know this audience well and I enjoy talking with it.

  A precise audience, in a field you know, tells you where to find your customers and what their real problems are, and lets you speak to them in their own terms. Written down (who your buyer is, in which line of work, and why they would pay), that audience is called your ideal customer profile (ICP).

Check that people will pay, before you build

Validating means checking that people will pay for your software before you build it. Being able to build in a day doesn't spare you that check.

Rob Walling runs TinySeed, a fund that has invested in 241 software-subscription companies. He also sells a course for SaaS founders.

In October 2026 he told the Practical Founders podcast that a buyer of the course asked for a refund, "because I can just build a product in a day and launch it". Walling refunded him but disagreed: building fast doesn't tell you whether anyone will pay.

Walling revised the course to answer that exact objection. He heard the same from Steve Blank, who teaches entrepreneurship at Stanford: students now arrive with a prototype built by AI, having talked to no customer, "and it was to their detriment".

Walling sums up what experienced founders say about their first company: "I wish I would have had more conversations before I started building and I wish I would have sold something before I quit my day job."

To check that people will pay, make these four moves in order:

1. Find people who have the problem. Look where they already talk about it, or where they gather: the comments under posts about the problem, the online groups and forums of their line of work, a trade fair, a community you belong to. Several founders in this guide found the first people they talked to in exactly those places.

2. Ask about their life, not your idea. Rob Fitzpatrick, author of The Mom Test, the standard book on customer interviews, advises asking for facts: what did they do the last time the problem happened? If you pitch your idea, people compliment it out of politeness. A compliment, he says, "means nothing. It's like the fake phone number of business feedback."

   In a 2018 talk, he gives a rough number: talk to about six people in the same situation, with the same job or the same kind of business. If your guess about the problem is right, all six describe the same problem and some ask how they can buy it. If you hear three different stories, you're talking to three different kinds of customer, with different problems and budgets.

3. Look at what the problem costs them today. People are bad at predicting what they'd pay. What they already spend sets the ceiling of what they'll pay you: the price of the software they use now, or the hours they lose multiplied by what an hour of their work is worth.

4. Ask for money. Take pre-orders: a payment before your software exists, which proves the problem is worth money to them. None of the founders here names a number of pre-orders to aim for. Vic Levitin, a serial founder who sold software to eBay sellers for years, judges a new product instead by its "time to money": how fast it reaches its first $100 of monthly revenue. At $29 a month, that is four customers.

   Levitin now builds with a technical partner under one rule: no code before proof that people will pay. In Levitin's words, his partner "doesn't write a single line of code before I prove to both of us that I can generate demand."

   That first $100 matters, he says, because "there's a huge difference between a product that does nothing and a product that does even $100". Levitin knows the other path, the one most first-time founders take: "You lock yourself in a garage, you build for two years, you ship, and you discover that nobody actually needs what you want" (Practical Founders, August 2026).

   Take a simple pre-order: you charge the buyer's first month, $29, today. Your terms of sale say three things. Access opens by a date you set, 1 March for example. If it doesn't, you refund the $29 in full. From one month after access opens, 1 April here, the card is charged $29 a month until the buyer cancels.

Clar, a study assistant built by three students in Stockholm, shows these steps in practice. A month after its launch at the end of January 2026, its subscriptions were running at about $10,800 a month, about $130,000 a year (Foundore podcast, May 2026).

Its founders, Isabelle, Eric and Andreas, had no technical background.

Isabelle had already learned what a waitlist is worth. Her earlier study app had filled one with students who liked the pitch, because signing up for a waitlist costs nothing. The list said nothing about whether the app would help at exam time, and it didn't: "when the exam came the usage dropped completely".

So for Clar, the three set up a table at their university's career fair, with a sign telling students who already paid for ChatGPT Plus, ChatGPT's paid plan, how much they would save by switching. They didn't collect emails: they asked students to pay before Clar existed, by card or Apple Pay.

They don't say how many paid that day, but it was enough for all three to choose to leave university and build it with Lovable, an app builder that writes a web application from a conversation.

You don't have to do the same: checking an idea doesn't require leaving your job. Each conversation fits in a lunch break. What takes time is finding a slot with each person: allow a few weeks for six or so conversations, and keep going while new people still tell you something new.

Building with an AI agent

Building is now the shorter part: days to months, compared with months or years to find customers. Boris Cherny runs Claude Code, the coding agent of Anthropic, the company that makes the AI assistant Claude. "I don't write code. I prompt Claude", he says: he types requests, called prompts, and Claude writes the code.

He said it on the Big Technology Podcast in May 2026. In October 2026, his team put the share of Claude Code's own code written by Claude itself at about 90%.

In May, at a talk at Y Combinator, a Silicon Valley program that funds and coaches new startups, Cherny asked the hundred or so founders in the room which of them had all their code written by Claude Code: about half raised their hands, he recounted on the same podcast.

At MicroConf Europe, Rob Walling's conference for software founders, in October 2026, Derek Reimer, founder of the scheduling software SavvyCal, met several founders who "don't even read code anymore". AI can build your SaaS: the question now is what you ask it for.

Two kinds of services build for you:

• App builders (Lovable, Bolt, Base44) build a whole web application, software you use in a browser, from a conversation. They include the database, the sign-in for your users, and the hosting. An app builder is the shortest path if you don't want to install anything, and the cheapest at the start.

  Base44's own guide to starting a SaaS calls this kind of service a "no-code app builder": you build without writing code, by describing what you want.

• Coding agents (Claude Code from Anthropic, Codex from OpenAI, Cursor) work on the files that make up your app's code, in a folder on your computer. Claude Code also lives in the Claude desktop app, under the Code tab, so you don't need the terminal, the window where you type commands. With a coding agent, you set up hosting and the database, by asking your agent to do it. Coding agents give you more control over each piece and what it costs.

Each plan comes with a monthly amount of use (credits or usage limits). For more, you move up a plan or buy credits.

What founders who don't code say

• Vadim spent about four months building VGola with Lovable, from October 2025, before he started selling it.

  As he learned more, he asked Claude how to move off Lovable. The answer: for a video app like his, which processes video for hundreds of paying users, separate services would do the same "for way cheaper": GitHub to keep the code online, Supabase for the database, Vercel for hosting. He now builds with Codex, OpenAI's coding agent.

• The Clar founders went from months to days. Before Lovable, Isabelle had paid a team of engineers in India for six months to build an app. "With Lovable I could… build like much better version… in like a couple of days."

  Their hard part was elsewhere: cutting a thousand possible features down to one workflow students already had with ChatGPT, made slightly better for studying. That took about four months of tests before launch.

• The founder of Copyboost, a marketing student who couldn't write the simplest program, built a SaaS that scores marketing texts on his own, in six months, with Claude Code as his "technical co-pilot". Copyboost now has paying users (Indie Hackers, 2026). His first lesson: "Ship with 3 features, not 7. I over-scoped": he planned too much for a first version.

With my own products, ReadyToPost and contenttosocial, coding fast is no longer the issue, marketing is. The product has to work and meet the need, but it doesn't have to be perfect to find its first customers: what counts is a first customer, then a second, then a third.

The thinking at the start matters more than the building: who you're talking to, and how. Once those choices are clear, AI can take on a good part of the execution.

What stays hard, and what to ask your agent

1. Knowing what you want. First write, in a few paragraphs, what your software does, for whom, and the one task it must get right. Then ask your agent to propose its plan of work before any code: Lovable and Claude Code both have a plan mode, which you switch on before sending your request. In plan mode, your agent describes what it will do without changing anything.

2. Connecting the database and the hosting. This is where people who don't code often get stuck. An app builder that includes the database and hosting spares you this step. Otherwise, ask your agent how to do it every time you're stuck.

3. Keeping every version. With a coding agent, ask it to save each working step with Git, the system that keeps the history of your code, and to send that history to GitHub, a service that keeps a copy online. If a change breaks everything, ask your agent to go back to the previous version.

4. Giving instructions once and for all. Ask your agent to keep a file of standing instructions it rereads with every request (CLAUDE.md in Claude Code): the project's rules, the decisions you've made, and for example that "save" means saving this step with Git and sending it to GitHub. Claude Code reads the CLAUDE.md files in your project on its own.

5. Small steps, checked as you go. Launch with few features, and ask your agent to write automated tests, small programs that check each feature after every change. As long as no customer can use your software, every addition delays the day your first customers can tell you whether it does the job they paid for.

   With my own products, I also ask my agent to test each change itself in a browser, the way a user would. And my automated tests run on their own each time the code changes, before the change goes live. This is called continuous integration (CI). It protects me from a good many mistakes. You can ask your agent to set up both.

What it costs

Building costs little money. What costs you is your time, then the bills that grow with each customer. Prices read on the services' sites on 29 October 2026:

A common trap concerns Vercel, often described as free: its rules reserve the Hobby plan for "non-commercial personal use only" and count as commercial "any method of requesting or processing payment from visitors of the site". So a SaaS that takes payments needs a paid plan, such as Vercel Pro at $20 a month.

Two typical budgets at the start:

• Everything on Lovable: the Pro plan costs $25 a month and lets you use your own domain name (bought separately). One credit balance pays for building, hosting, the built-in database and your app's AI, so it runs down faster as your customers use your app: then you move up a plan or buy credits.
• A coding agent and services you pick yourself: Claude Pro ($20), Vercel Pro ($20) and Supabase Pro ($25) make $65 a month, before the AI your customers use.

The first is cheaper at the start. With the second, each service's price is set separately, so you see what grows with your customers.

The AI in your product is billed by how much your app uses it. With my own products, that bill climbs very fast with the best models, so I set a spending cap wherever a service offers one. Anthropic, for example, lets you set a monthly spending cap on its API, the doorway your app goes through to use Claude.

Once your app reaches that spending cap, Anthropic refuses your app's requests until you raise the cap or the month ends (its documentation, read on 3 November 2026). Your app's AI then stops working for your customers, so set the cap above their usual use.

Security, in proportion

For most apps, security doesn't take months: six checks are enough, each one a request to your agent or a setting to switch on. Do them before your pre-order buyers' access date, the day your first customers get in.

The Copyboost founder found two flaws by checking his own product. His database access rules "technically worked", but a user who sent requests straight to the database, past the app's screens, could have read other users' data. He rewrote them all.

He also rewrote the part of his app that receives Stripe's payment messages three times "before it stopped losing events". Until then, a payment or a cancellation could go unrecorded, so a customer could pay and not get access, or cancel and keep access.

Security can't wait because attackers use AI too. Google Threat Intelligence Group, Google's team that tracks attacks, counted the software flaws that attackers actually used, worldwide. It found 10.5 a month in 2025, and 18 a month from January to August 2026.

Google offers an explanation. Your app rests on building blocks of code written by others, which their authors fix regularly. When an author publishes a fix for a flaw, the fix also shows where the flaw was.

Comparing the old and new versions to build an attack used to take work. AI lets attackers automate that work, so apps still missing the fix get attacked very quickly.

One example: in early 2026, a flaw was announced in BeyondTrust's software, which companies use to take remote control of their computers. Within four days, attackers were using it to get into companies that hadn't installed the fix (Google's report of 30 September 2026).

Once a fix for a widely used building block is out, you have only days to install it, which is why updates need to reach you fast. AI can also check your app before attackers do. What to ask for and what to run:

1. Each customer sees only their own data. This is what AI-built apps can get wrong. In many of them, the browser talks directly to the database. Each of the database's tables holds one kind of data, such as users or orders. Only rules written in the database, called row-level security, decide who can read what: without them, any user can read every customer's data.

   Ask your agent: "Check that every table has access rules and that a user can only read or change their own account's data. Explain the rules to me in plain English." Then test it yourself in five minutes: create two accounts and check that the second sees nothing of the first.

   That test only shows what your app's screens let through. To check a request sent straight to the database, as in the Copyboost founder's case, ask your agent: "From the second account, try to read the first account's data by querying the database directly, and show me the result."

2. Run the built-in security checks. In Lovable, open your project's Security view: a quick scan runs by itself each time you publish, and you start the deep scan there by hand.

   If your database is on Supabase, open its dashboard's Advisors tab, which flags tables without rules. In Claude Code, type /security-review in your conversation: it checks your latest changes for flaws. Before your first customers get in, ask your agent to check your whole app for flaws.

3. The payment opens access, not the thank-you page. When a customer pays, Stripe sends your app a signed message, called a webhook, that proves the payment went through. That message is what should open access. If access opens as soon as someone reaches the "thanks for your purchase" page, anyone can type that page's address and get in without paying.

   This matters for your pre-order buyers too: their accounts are created from the payments Stripe confirmed. So do this check before their access date.

   Ask your agent: "Only open access when Stripe confirms the payment with a signed webhook, never from the thank-you page. Handle renewals, failed payments and cancellations too, and show me how to check that none is lost." Stripe sends a message at each stage of a subscription, and handling all of them can take several tries, as the Copyboost founder found.

4. Secret keys stay secret. A Stripe secret key is the password your app uses to act on your Stripe account: whoever finds it can issue refunds or read your customers' data, for example. Ask your agent to keep every key on the server side, in environment variables, a place the browser never sees, and to check that none appears in what the browser receives.

5. Security updates reach you within days. Since attackers exploit known flaws quickly, don't let updates to your app's building blocks wait. If your code is on GitHub, switch on Dependabot, GitHub's service that proposes security fixes: on your project's page, open Settings, then Advanced Security, and click Enable next to "Dependabot security updates".

   Dependabot doesn't install anything by itself: it proposes each security update as a change ready to accept. Ask your agent to apply them as they come.

6. You know when it breaks, and you can restore it. Ask your agent to connect an error-tracking service such as Sentry or PostHog, which alerts you when a user hits an error. Ask it once, too, to show you how to restore your database to yesterday's state from a backup, and to try it on a copy of the database it makes for you, never on your live database. On Supabase, daily backups come with the Pro plan.

   I tried such a restore on one of my products and found that Supabase's backups leave out your users' files. Supabase keeps the photos or documents your users upload in Supabase Storage, a separate service. A database backup holds only information about those files, not the files themselves (Supabase's documentation, read on 3 November 2026). So I now back up those files separately. If your users upload files, ask your agent to back them up too.

If your app holds sensitive data (health records, bank details, data about children), have a developer review it before inviting customers. Pawel Reszka, who reviews apps built with AI for a living, charged about $500 for a two-to-three-day audit in April 2026 (Indie Hackers).

Launching your SaaS

Launching your SaaS means opening it to paying customers on a date you announce, with a clear offer: a price and what it includes. Putting your software online brings no customer by itself: if nobody knows you or your product yet, nobody comes on the day it goes live.

Your launch brings your first customers beyond those who pre-ordered. For the next ones you need a channel, a way of reaching new customers that you can repeat every week, such as outreach (contacting people directly), content (articles, videos) or an app store. The trap at this step is counting on launch day alone.

Two dates come in order. The first is the access date your pre-order terms promise. On that date, your agent creates your pre-order buyers' accounts from the payments Stripe confirmed and starts their subscriptions, with the first charge a month later. You open your first version to them, often called an alpha or a beta: they use it for real work while you fix what breaks.

When one of them uses it without your help, building is done, and you set the second date: your launch day, when you open to everyone. Stripe's guide to starting a SaaS gives the same advice: a beta lets you fix problems, and hear what users think, before you open to everyone.

Alex Kracov, who co-founded Dock, software for sales teams, describes that phase in his account of its first 90 days. He invited users into Dock's "alpha program" and helped each of them get started, even building their first templates himself: "the classic 'do things that don't scale' move", work done by hand for each user that would be impossible with thousands.

The users who kept using Dock started inviting their teammates, which gave him the first signs that "it's working".

Meanwhile, write down who you'll tell first: the people who described the problem to you when you checked that people would pay, then those who already know you. Those who described it already know the problem your software solves. What your launch brings depends on that list, so build it in the weeks before launch day, not on the day itself.

Within a week of launching Swipefast in October 2026, Jacob Pegs had his first $49-a-month payments. He is a consultant who teaches sales and marketing to freelancers and coaches, and he built a waitlist for a month before the launch.

Swipefast makes checkout pages, the pages where a buyer pays for an offer, and lets an AI assistant such as Claude draft them for you. Before the launch, about 16 of his consulting clients already used Swipefast, as part of the consulting they paid him for.

He has written a daily email for four years, to a list of about 6,500 people. For a month, the bottom of each email carried a link to a waitlist and one line: it launches in October. 172 people joined that waitlist. Signing up costs nothing, so a waitlist doesn't prove that people will pay. His waitlist served another purpose: on launch day, he knew whom to write to.

Launch week, he sent seven emails to the 172 on his waitlist and published seven posts on LinkedIn, where 56,000 people follow him. His checkout page said plainly that the first 100 users would pay $49 a month and that the price would rise afterwards, with no countdown. Within the week he counted more than a dozen paying users, he said in a video recorded that week (YouTube, October 2026).

His launch rested on people who already followed him, an email list and an audience built over years, as he says himself. His advice: "pick the channels you already know. Don't spread yourself too thin."

A launch price for the first customers, like his, gives a reason to buy now without inventing urgency. A lifetime deal, one payment for access forever, is another launch offer, and it works only in certain conditions. Davis Baer, co-founder of Youform, a form builder, says Youform's lifetime deal brought in about $30,000 in its first few months.

He already had three things: a following on X, a large email list and enough revenue to keep Youform running. The list and the revenue both came from OneUp, his social media scheduler. So buyers could trust that Youform would last. Each lifetime customer also cost him nothing more to serve than a free user.

He dropped the deal once Youform's subscriptions reached $5,000 a month, and says nearly all the Youform copycats with lifetime deals have closed. In his words, a lifetime deal "only works in certain circumstances": buyers who trust the product will last, and customers who cost little to serve (his post, read on the Startups for the Rest of Us podcast, August 2026).

On launch day, tell your list first, then any community or audience that already trusts you, the way Hasaam Bhatti launched inside a coaching program for Amazon sellers whose owners had become his partners.

Launch sites come next: websites where makers present a new product to visitors for a day, Product Hunt being the best known. A launch site sends few visitors to each product, on average. Uneed, a launch site for tech products, says on its pricing page that it sends 22,000 visitors a month to the sites of the products it lists (read on 2 November 2026).

Its founder, Thomas Sanlis, said on Le Board, a French podcast for solo founders, in September 2026 that Uneed publishes 60 to 80 launches a day, about 2,100 a month.

That makes about ten visitors per product launched, on average: 22,000 ÷ 2,100 ≈ 10. And some of those visitors go to products launched in earlier months, so a new launch gets fewer.

With no one lined up, launch day can bring nothing at all. Gionatha, a solo founder, built Tryreplify, which uses AI to write replies to the reviews that customers leave local businesses on Google, and launched it at the end of April 2026 at $15 a month, with no audience. His first week brought no paying customer, and two votes on Product Hunt, one of them his own, he wrote on Indie Hackers (May 2026).

He hadn't checked whether anyone would pay before building. The shop owners he talked to afterwards, at least ten of them, all found the tool useful, yet none started using it. Founders who replied to his post pointed out that his buyers, owners of local businesses, don't spend their days on Product Hunt: they are in their own groups and forums, where he would have to go and find them.

People lined up before launch day make the difference on a launch site too. Plug This finished third of the day on Product Hunt in the summer of 2026, with 36 paying customers on launch day, though its founder, Uday Prakash, had no big email list. Plug This builds Chrome extensions, small add-ons for the Chrome browser, from a description you type.

For two weeks before launch day, he prepared the listing, the product's page on Product Hunt, and lined up supporters one by one. He posted on LinkedIn, Instagram and X that he was launching, and messaged everyone who reacted to ask them to vote on launch day.

Product Hunt forbids that last part: its rules let you ask people to visit your product and comment, never to upvote it, that is, to vote for it (its launch page, in the Internet Archive copy of 21 October 2026). Ask your supporters for a visit and a comment.

On launch day, his team answered every comment within minutes, for about 20 of the 24 hours. Plug This drew about 558 votes and 2,600 visits. A launch offer turned 36 of those visitors into paying customers then and there: for the day, its most expensive plan cost the price of its cheapest (YouTube, August 2026).

So don't skip launch sites: they won't be your main channel, but they can bring you a few customers. My own first customers came from a launch site.

One post, or a few, where your buyers read can do the same job as a launch site: the first buyers of Denis Yurchak, who built an app for calling phones abroad, came from a Reddit post, and Vadim's first revenue from posts on X.

In the weeks after, watch your new users get started: fix wherever they get stuck during onboarding, their first days in your product. Then work on a channel you can repeat every week.

Finding your first paying customers

Finding customers decides whether your SaaS survives and what it earns, and that work starts with your first conversations.

Vic Levitin sums up the difference between founders: "First-time founders are obsessed about the product… serial entrepreneurs are obsessed about the distribution", distribution meaning how the product reaches its buyers. He now starts from it.

His latest company reached $90,000 a month of revenue within about 18 months, in a field he knew nothing about: AI influencers, made-up characters run by AI that build a following on social media. He got there through a partner who knew that field up close (Practical Founders, August 2026).

That partner was an expert who taught his own followers how to create and run AI influencers. The expert recommended a competing tool to them and earned a commission on each sale. Levitin proposed that the expert become his co-founder instead, with an equal share of a new company. Levitin and his technical partner would build and run the software, and the expert would bring his knowledge of the field and his audience. Their company sells a platform that lets anyone create such a character.

The founders quoted here found their first customers through the channels below, starting with the fastest.

People who already trust someone

The fastest way is to sell to people who already trust you, or who trust someone who vouches for you.

• An audience of your own. It's the fastest channel once you have one, and slow to build: Ben Bozzay needed six years to reach $20,000 a month. In 2020, he wrote free guides on Medium, Reddit and YouTube about how he had locked down his own devices with tools companies use on their employees' computers.

  Readers started emailing him to hire him as a consultant. He couldn't help them one by one, so he sold the answer: the paid membership first, then the subscription service. When he was laid off in 2023, he had about 300 paying customers, bringing less than $2,000 a month of profit. He then worked on it full time, and for his first months revenue barely moved.

• An audience you borrow. Hasaam Bhatti had no audience: "Zero audience means zero customers." But two years before building his tool, he had joined a paid coaching program for Amazon sellers, Legacy X, whose thousands of members were looking for tools. He offered its owners: "Give me 48 hours to build the solution. If you like it, we partner."

  He built it in those 48 hours with Cursor, and the program's owners called him the next morning to go ahead. The product reached $10,000 a month 30 days after launch, $21,800 at 90 days and $30,000 a month a few months later, a business he runs in partnership with the program's owners.

  He still can't read his own app's error reports without AI's help: he runs the product without being able to code.

  To copy his approach, go to whoever already gathers your future customers, and offer them a tool their members need, which you build and run for them.

No network: contact customers one by one

Without an audience or a network, what's left is direct contact, one person at a time, called cold outreach when the people don't know you yet.

• Freddy Galindo's customers aren't online, so his team finds them on the road. Prospecting is the search for possible customers, and his team does it by car, photographing the contractors' premises it drives past to note names it can't find online: "we drive around… taking photos of like companies cuz you can't find them on like Google Maps or like Sales Navigator". Sales Navigator is LinkedIn's search service for salespeople. His team then walks through the door, because "we align… the way we sell to how our customers buy".

  His company now works with SDRs, salespeople whose job is to book meetings, because one person can't make all those calls and visits: each new customer needs new ones. Working alone, the same idea, selling the way your customers buy, means meeting your buyers where they actually are.

• Denis Yurchak, a solo founder, built Yadaphone, which lets people call phones and landlines abroad from a browser. By May 2026 it brought in $17,500 a month, from 20,000 users who buy call credits. He built it in a weekend after Microsoft announced in February 2025 that it would close Skype. A Reddit post brought his first buyers, about $1,500 in the first month.

  His main channel came next. Many blog articles listing the best ways to make international calls still featured Skype, and some sat at the top of Google's results. He wrote to their authors, one by one: "you're still featuring Skype… there might be an alternative you might like", then explained why Yadaphone could replace Skype in their article.

  "About 90% of your cold outreach is going to get ignored", he warns, and each article takes two to four weeks of back and forth. But when one article high in Google's results swapped Skype for Yadaphone, new sign-ups rose by about 50 a day.

  He then asked happy users for reviews on Trustpilot, a review site, so that a search for his product's reviews would find some. He still starts every day with marketing, then support, then code, "because the hard thing is actually marketing and doing the distribution for your product" (Niche Pursuits podcast, May 2026).

You can also write to possible customers by email. In the US, the Federal Trade Commission, the consumer protection agency, explains the CAN-SPAM Act, the law on commercial email, in its compliance guide. No permission is needed to send sales emails, including to businesses. Each email must give accurate sender details and a subject line that doesn't mislead, and say it's an ad. It must also include a postal address, and you must honor an unsubscribe request within 10 business days.

In the UK, the Information Commissioner's Office, the data protection regulator, says the rule requiring prior consent for marketing email doesn't apply when you email a company, but sole traders count as individuals: you need their consent first. The rules and an example are in this guide to cold email.

Show your product at work

With short demonstration videos, Vadim took VGola to $11,200 a month, up from $2,000 to $3,000. That figure comes from about 400 paying users. He measured it over the four weeks before his interview on Starter Story, a YouTube channel of founder interviews, in October 2026.

Vadim had never coded when he started building VGola. His first $2,000 to $3,000 a month came from posts on X. Then he opened a new Instagram account and posted short demonstrations of his product, three to six a day. He copied what already worked in his competitors' videos: the opening line that makes a viewer stop, the length, the sound. It takes him about 45 minutes a day.

Videos show your product to many people at once. A live demo does that job with possible customers you talk to one by one. I gave many demos of my own products, and I let each person run the demo: each one got a demo account, a login set up just for trying the product. I watched their screen on a video call while they used it.

Using the product themselves, they made up their minds about it much faster than by watching me use it. These early demos also tested my screens: I could see whether people found a button and clicked in the right place without help.

Give users a reason to talk about you

On TrustMRR, a site built by Marc Lou, a solo developer who lives off his software, founders display their monthly revenue and can sell their business. Each founder gives the site a key to their Stripe account, and TrustMRR reads the revenue there. It isn't a subscription business: in August 2026, it brought in $44,000 a month without advertising, from sponsors and a fee on each sale.

Marc Lou adds features his users want to show off, such as their revenue chart: by sharing it on social media, they promote the site for him. In August 2026, TrustMRR had 200,000 visitors a month. The idea works for any software whose users have a result they're proud to show.

The slow channels: Google, app stores and AI assistants

Being found on Google takes months to pay off. Ben Bozzay bet on it because he knew how, from the agency he once ran. At its peak in 2024 and 2025, his site drew about 70,000 visitors a month from Google, with guides tested on real devices, "something that you couldn't even ask AI to answer".

Many people now search for his brand name directly. People who type his brand's name find his site whatever Google changes in how it ranks other pages, so those visitors don't depend on his rankings.

His choice shows the test for picking a channel. None is easy and you can't work them all: pick the one you're good at and that you like, then stick to it.

For my own products, I chose SEO and GEO, the work of showing up in Google and in AI assistants' answers. They have two advantages when you work alone. You don't pay for the visitors they bring: that's called organic traffic. And their results add up: a page that ranks well keeps bringing visitors in the months that follow, while an ad brings no one once you stop paying for it.

When your buyers work in SEO, as contenttosocial's do, SEO also lets you go where they spend their day. People who work in SEO check, every morning, tools that show where their pages stand in Google for the searches they care about. An agency tracks, for example, the search "SEO agency" plus the name of its city.

So for contenttosocial, I published lists of the best SEO agencies, country by country and city by city, built to appear in Google for those same searches. When one of these pages ranks, the agency's tool shows it every morning among the results for the agency's own search, next to the agency's site. The agency sees it, opens it, and finds contenttosocial there.

App stores are another slow channel whose effect builds up over time. Zigpoll gets 90% of its business from the Shopify App Store, where online stores built on Shopify look for add-ons. For his first two or three years, Jason Zigelbaum's single goal was reviews on that store: "Like that was the North Star. Everything else didn't matter."

It took him about two years to get going, and since 2022 its revenue has doubled every year.

AI assistants are starting to send customers too. Several first-time attendees told Rob Walling in October 2026 that they had found MicroConf, his conference, by asking ChatGPT where to meet other founders. What makes an assistant cite a product is in this guide to GEO vs SEO.

Paid ads don't replace these channels. Simon Høiberg, who runs several software products from Switzerland, says ads "should amplify" an audience you've built "instead of being the only way you can reach a customer".

I tried Meta's ads, those shown on Facebook and Instagram, for one of my products. An ad works only with a precise target: it has to speak to people who recognize their own problem in it.

Ads also took a lot of time and a large budget at the start, which is hard when you work alone. Meta's program, which decides who sees each ad, first had to learn who my audience was. It mostly ran tests until my ads had been shown enough times for that, and my budget paid for those tests.

Keeping your customers, and what you depend on

Keeping customers matters too: a customer who stays is one less to find.

• Let your customers help themselves, and answer fast. Denis Yurchak runs two products alone: Yadaphone and eSIMPal, which sells eSIMs, phone data plans for travelers. Support takes him about an hour on a bad day. His sites tell customers what to do when something goes wrong, and he automates any support task he's had to do by hand more than three times, which you can ask your agent to do.

  For the rest, AI drafts a reply to each message in his app's admin page, which only he can see, and he reads each draft before sending it.

  Answering yourself has another value, beyond speed: a reported bug often starts a real exchange with a customer. When they see you take their request seriously, a relationship forms, which is why, with my own products, I still answer them myself.

• Watch your customers use your software. Sit next to a user trying your product, or watch them share their screen on a video call, and note where they look for a button, without helping. An analytics service such as PostHog also shows, for each user, where they click and where they give up.

• Guide the first days, and find out why customers leave. Jason Zigelbaum asks each new Zigpoll user what they came for, then sends one of 24 onboarding emails, one for each use he has identified. Before AI, Zigelbaum says, writing his 24 "would have taken weeks". My own products work the same way: their emails depend on what each new user has already tried.

  When a Zigpoll customer moves down a plan, including to the free one, a survey asks why. The main reason: customers run one survey campaign, get their answers and stop. So Zigelbaum shows them a use they can leave running for good, such as a question after each purchase asking buyers how they heard of the store.

  Churn is the share of customers who leave each month. "Lowering churn by like one to two percent", Zigelbaum says, "can enable a lot more growth". If 5 in 100 customers leave each month, the average customer stays 100 ÷ 5 = 20 months. At 4 in 100, 100 ÷ 4 = 25 months: the same customers bring in a quarter more.

  He also offers yearly plans at 25% off, because a customer who pays for the year is far less likely to leave. He tried 50% off first, but "it came off as perhaps desperate."

• Know which platform you depend on, and for what. Getting most of your customers from one place, such as an app store, is a risk you reduce over time by adding channels. Renting standard services, such as hosting or a database, is a small risk if your code and data can move to another provider.

  The deepest risk is software that only works on top of one company's platform, because a change on their side breaks your software. Zigpoll, which runs inside Shopify, carries some of that risk. Vic Levitin built his first software company on eBay's API, the doorway eBay opens to other programs: "it was very painful when eBay make a tiny API change… months of fixing your app". He now avoids products that depend on a single company's platform.

• Plan for the AI model your product uses to fail. An AI model is one version of an AI, such as one of Claude's. With my own products, AI models have let me down, though less often today: a model unavailable for a while, an answer that wasn't the one expected, or no answer at all. Each failure then reaches your customer as an error, unless something takes over.

  So I set up fallbacks, backup models that take over one after another: if model A fails, the request goes to model B, and if model B fails, to model C. You can ask your agent to set up the same.

  AI providers also set a request limit: how many requests your app can send in a given time. Anthropic can start a new account with lower request limits, which rise as the account builds a history of use (its documentation, read on 3 November 2026). Above its request limit, your app's requests are refused for a while, so check your provider's request limits before your customers' use grows.

Pricing: how many customers make $1,000 a month

Your price sets how many customers you need: at $20 a month, 50 customers bring in $1,000. At $100 a month, ten are enough.

Charge from the start, and start from your competitors' price and from what the problem costs your customer today.

Charlie Clark's Product Hunt launch brought Links traffic but no revenue, since Links was free. The money came when he added a premium plan, at about $3 a month, which brought in $100 a month within a couple of months. He sets his price by looking at Linktree's and staying below it, a choice he made as the small rival of a much bigger service. The premium plan later went to $4, and in late June 2026 to $5.

To measure such a change, founders use MRR, monthly recurring revenue: what subscriptions bring in on average each month. A customer at $29 a month counts for $29. A customer who pays $290 for a year counts for $290 divided by 12, about $24 a month.

Links has thousands of subscribers who each pay a little, so the extra dollar "translated to nearly 5,000 in additional MRR": about 5,000 premium subscribers, each paying $1 more. Links also has a Pro plan from $12 a month for customers who need several pages, and a few of them pay over $100 a month.

The standard pricing models are a flat price, tiers, per user, by use, or freemium. Tiers are several plans at rising prices, each with more features, as at Links. A price by use means paying for what you consume. A price per user is common when people work together in the software. Freemium means a free version with limits, next to the paid plans.

Two other founders' choices, each suited to a different case:

• Credits instead of a subscription, when use is irregular. Yadaphone has no subscription: people call abroad now and then, so they buy call credits that never expire, from $5. Yurchak refunds on request, no questions asked, which reassures buyers who don't know him.

  Yadaphone's business customers top up a shared balance from $100, with no fee per user. In a year, only one of Yadaphone's 30 company customers left, and Denis Yurchak blames his own landing page, the page that presents the product, which led that customer to expect something else.

• A free trial, when people only see the value by using it. Once Clar was live, it also gave students 30 days free, and about 40% of the students who started the free month were still paying in month three (Foundore podcast, May 2026). A study assistant proves itself over weeks of revision.

With my own products, I went the other way: removing the free trials turned more visitors into customers. A customer who has paid gets involved: they want the software to work for them, and they tell you what's missing. That feedback is what teaches you what your customers really want. A free user moves on more easily.

When you do give something free, give it an end or a limit: a trial ends, and a free plan, like Zigpoll's or Links', does less than the paid ones, so a growing customer moves up. Jason Zigelbaum keeps Zigpoll's free plan because its customers sign up on their own, from the Shopify App Store, without talking to him.

A one-time purchase brings money once, while every customer costs you hosting and AI every month. Simon Høiberg sells a lifetime bundle of his products that "people kept buying… consistently", and says it works because he spent a lot of time making his servers cheap to run. Whatever your price, it has to cover what each customer costs you.

Business structure and tax in the US: what you keep

Keep the legal and tax side in proportion: as long as your SaaS earns nothing, you owe almost nothing. Your structure and income tax can wait until after your first sale, and a pre-order already counts as a sale.

If you have a job, check your employment contract for clauses on side work and on who owns what you build, in the US as in the UK.

1. You're a sole proprietor from your first sale. "You're automatically considered to be a sole proprietorship if you do business activities but don't register as any other kind of business", says the Small Business Administration, the federal agency for small businesses. There's nothing to file to create it.

   You and the business are the same person, responsible for its debts. If you sell under a name other than your own, most states ask you to register that name (a DBA, for "doing business as"). Before your first sale, check your city's and county's websites too, since local governments set their own license rules (Small Business Administration, read on 29 October 2026).

2. An EIN is optional at first. It's a tax number for the business, free and issued in minutes on the website of the IRS, the federal tax agency. Without employees, you can use your own Social Security number instead. Some banks ask for an EIN to open a business account.

3. You pay tax on the profit. In general, self-employment tax, 15.3% for Social Security and Medicare, is due once your net earnings reach $400 in a year. It's worked out on 92.35% of your profit, on a form called Schedule SE.

   The 12.4% Social Security part of that tax stops once your salary and your self-employment earnings together pass $184,500 in 2026 (IRS, read on 2 November 2026).

   Income tax comes on top, at your own rate, with your business profit reported on Schedule C, the part of your return for a sole proprietor's profit or loss. Most states add their own income tax.

4. You pay during the year. If you expect to owe $1,000 or more in tax when you file your return, the IRS expects estimated payments each quarter. With a salary, there's a simpler route: ask your employer to withhold more tax from your pay, with a new Form W-4, the form that tells your employer how much to withhold.

   Generally, you avoid a penalty if what you paid during the year covers at least 90% of this year's tax, or 100% of last year's, whichever is smaller (IRS, read on 2 November 2026).

What's really left on $1,000 a month, from 50 subscribers at $20, with a coding agent and services you pick yourself:

An LLC protects your personal assets, not your tax bill. A limited liability company, formed with your state, shields your house, car and savings if the business is sued or goes bankrupt. For federal tax, a one-person LLC is "disregarded" by default: its profit goes on your return exactly like a sole proprietor's (IRS).

An LLC costs a state fee every year, for example $800 in California and $400 in Delaware. The Small Business Administration describes it as a good choice for medium- or higher-risk businesses and for owners with significant personal assets to protect.

Two things older guides still mention. Stripe Atlas, a Stripe service, forms a Delaware company for you for $500, and Stripe presents it as a way to create a US company from wherever you live and to raise money from investors: US founders of a small SaaS don't need one.

And companies created in the US no longer have to file a beneficial ownership report (BOI, a list of who owns the company): FinCEN, the Treasury's financial crimes bureau, made that exemption final on 11 August 2026.

Business structure and tax in the UK: what you keep

UK founders face the same choice of structure. "Most people set up as a sole trader when first starting out as a business", says GOV.UK, the UK government's website: it's "the simplest business structure to set up", and you can be "employed and self-employed at the same time". You and the business are the same person, responsible for its debts. In order:

1. Start selling. You can trade without registering. In each tax year (6 April to 5 April), your first £1,000 of self-employment income is tax-free: this is the trading allowance. Once your income passes £1,000, you take off either that £1,000 or your real costs, such as fees and tools, but not both (GOV.UK).

2. Register after a tax year in which you pass £1,000. Register with HMRC, the UK tax office, for Self Assessment, the yearly online tax return, by 5 October after the end of that tax year. If you start in the tax year that ends on 5 April 2027, for example, you register by 5 October 2027.

   The year's tax is paid by 31 January after it ends. Once your tax bill reaches £1,000, HMRC also asks for two payments on account: advance payments toward the next year's bill, each half of last year's, due by 31 January and 31 July. It doesn't ask for them if more than 80% of your tax was already taken outside Self Assessment, from your salary for example (GOV.UK, read on 2 November 2026).

3. Pay on the profit. Income Tax, at 2026–27 rates, is nothing on the first £12,570 of all your income: your Personal Allowance, which shrinks once your income passes £100,000. Then it's 20% up to £50,270, 40% up to £125,140 and 45% above (Scotland's bands differ). National Insurance, the UK's social contributions, is 6% on self-employed profits between £12,570 and £50,270, and 2% above.

4. Keep digital records later. Making Tax Digital, HMRC's rules for keeping records and filing through software, requires sole traders to report through compatible software once their self-employment and property income passes £50,000 a year. That threshold has applied since April 2026. It falls to £30,000 in April 2027 and to £20,000 in April 2028.

What's really left. Take a side project alongside a job, with £11,000 of profit a year after your real costs, such as fees and tools. Because you took off your real costs, the £1,000 trading allowance doesn't apply.

Say your salary already uses your Personal Allowance and stays under about £39,000, so that salary and profit together stay below £50,270, where the 40% rate starts. You pay 20% Income Tax on the profit, £2,200. You pay no National Insurance, since the profit stays under £12,570. That leaves about £8,800 of the £11,000.

If the SaaS is your only income, the same profit falls within the £12,570 Personal Allowance: no Income Tax and no National Insurance.

A limited company is a separate legal person, so if it fails, you risk only what you put in. It costs £100 to register online with Companies House, the UK's register of companies, and £50 a year for its yearly confirmation of details. It pays Corporation Tax on its profit, 19% up to £50,000 and 25% above £250,000, with the rate rising gradually in between.

In return for that protection, a limited company must keep and file company accounts, and taking money out of it, as a salary or dividends, has its own rules and taxes.

Sales tax, VAT and taking payments

These taxes matter from your first pre-order, then every month you charge. Two taxes can apply to what you sell: US sales tax, set by each state, and VAT, the tax the UK, the EU and many other countries add to the price of most purchases.

US sales tax: your own state first. Each state decides whether software sold online is taxed. New York taxes it. Texas taxes software as a service as data processing. It exempts 20% of the price and taxes the other 80% at the state rate of 6.25%, so the state tax comes to 5% of the price. California generally doesn't, since no physical item changes hands. Your state tax department's website says whether it taxes software: search it for "software as a service". Then it depends on where your customer is:

• In your own state, if it taxes software, you register before your first taxable sale, a pre-order included, and collect the tax from then on. New York asks every seller of taxable goods or services, "even if you make sales from your home", to register "at least 20 days before you begin business" (read on 2 November 2026). Texas requires a sales tax permit from any business with a location in the state, such as an office.

• In other states, you collect only once your sales into that state pass the threshold it sets for sellers based elsewhere: more than $500,000 a year in Texas, a high one.

  In 2018, in a case called South Dakota v. Wayfair, the Supreme Court ruled that sales alone can oblige a seller with no presence in a state to collect its tax. That ruling is why other states can make you collect their tax once your sales there pass their threshold. The Court upheld South Dakota's much lower threshold: $100,000 or 200 sales a year. South Dakota dropped the 200-sales test in July 2023 (its Department of Revenue, read on 2 November 2026). Check each state where you sell.

For example, a founder based in New York charges New York sales tax to a customer in Buffalo from the first sale, but charges no Texas tax to a customer in Houston until sales into Texas pass $500,000 a year. In its early months, a small SaaS rarely reaches a threshold set on sales amounts. Where a state counts sales by number, each monthly charge may count as one: check its rule.

Stripe Tax, a paid option, warns you as your sales near a state's threshold: it cost 0.5% per transaction where you're registered, on 29 October 2026. A merchant of record, which sells your software on your behalf, handles the tax for you.

Who charges VAT, case by case.

• To individuals in the EU: the €10,000 yearly threshold you may read about only applies to sellers based in the EU. US and UK businesses owe EU VAT from the first sale, at each customer's country's rate. They pay it through the EU's One Stop Shop, a single online service for VAT on sales across the EU: you register once, in one EU country of your choice, then file one return a quarter for all your EU sales.
• To a business in the EU: the business customer usually declares and pays the VAT itself (the "reverse charge"), so you invoice without VAT. A UK seller charges no UK VAT on that sale either (GOV.UK, VAT Notice 741A).
• To individuals in the UK, from outside the UK: you must register for UK VAT "regardless of taxable turnover", meaning whatever your sales (GOV.UK).
• UK sellers at home: VAT registration becomes compulsory once sales subject to VAT pass £90,000 over 12 months.
• To a business in the UK, from abroad: the UK customer pays the VAT itself (the "reverse charge"), so you invoice without UK VAT.

Taking payments: payment processor or merchant of record

Before your first pre-order, choose how to take payments. There are two ways. With a payment processor, you stay the seller: it charges your customer's card, and you handle invoices and any sales tax or VAT.

A merchant of record sells your software for you: it takes the payment, works out and pays the sales tax or VAT of each country, handles fraud and disputes (customers asking their bank to reverse a payment), then pays you the rest. Stripe offers both: its standard offer is a payment processor, Stripe Managed Payments is a merchant of record. Stripe reviews each business before letting it use Managed Payments.

Stripe makes a payment page you send by email or message, with no code. For a pre-order, that page is a one-time payment link, not a subscription, so the monthly charge starts from the access date, not from the day they pre-order. When you create it, tick "Save payment details for future use", under Advanced options, so the buyer's card is kept for the monthly charge.

Lemon Squeezy is being folded into Stripe: its team now builds Stripe Managed Payments, and in 2026 its co-founder wrote that the goal is to give Lemon Squeezy users "an easy way to migrate" there.

What you keep on a $20 monthly plan, paid with a US card, with no sales tax due:

• Stripe standard: 2.9% of $20 ($0.58), plus 30¢, plus 0.7% for subscriptions ($0.14), makes $1.02 in fees, so $18.98 left.
• Paddle, or Polar's free plan: 5% ($1.00) plus 50¢ makes $1.50, so $18.50 left.
• Stripe Managed Payments: the $1.02 above plus 3.5% ($0.70) makes $1.72, so $18.28 left.

The merchant-of-record trap. A merchant of record is the legal seller, so it charges sales tax or VAT wherever the sale is taxed, from the first sale, whatever your own thresholds. A UK sole trader far below £90,000 charges no VAT to a UK customer when selling directly.

Through a merchant of record, the customer's 20% VAT is charged: on a £20 price that includes it, you keep £20 ÷ 1.2, about £16.67, before fees, and if the VAT is added on top of the £20, your customer pays £24.

A US founder likewise sees sales tax collected for customers in New York or Texas from the first sale, even when based in another state and far below those states' thresholds.

A merchant of record is worth its fees when individuals abroad can buy: in the EU, or in the UK when you're based elsewhere, you owe their VAT from the first sale anyway. Where you'd owe no tax, it costs you the tax it charges, plus 48¢ to 70¢ more on a $20 payment than Stripe alone (Paddle or Polar, then Stripe Managed Payments).

My second SaaS, contenttosocial, has customers in other countries, and I didn't want to deal with each country's tax rules. That's why I use Stripe Managed Payments, at that price: it handles the tax in each customer's country for me.

Charging every month: what to ask your agent

Four requests to your agent set up the monthly charge (Stripe's documentation, read on 2 November 2026). Give all four before your pre-order buyers' access date: your agent runs the first two on that date.

1. Give your pre-order buyers their accounts. They paid before your app existed, so no payment message reached it. Ask your agent: "Create an account in my app for every email address whose payment Stripe confirmed through my pre-order link, and send each one an invitation to sign in." Access still comes from Stripe's confirmation, never from the thank-you page.

2. Start their subscriptions. Ask your agent: "On the access date, start a monthly subscription for each pre-order buyer, on the card they saved, with the billing cycle anchor one month after access opens and proration set to none, so the first month they paid isn't charged again."

   The billing cycle anchor is the date that sets each monthly charge: 1 April, if access opens on 1 March. Proration means charging for part of a month. Set to none, it charges nothing for the days before the anchor.

3. Let new customers pay in your app. For customers who come after the launch, ask your agent to add a Stripe checkout for your monthly plan to your app's sign-up page.

4. Let your customers cancel online. Ask your agent to add Stripe's customer portal, a page where each customer cancels their own subscription.

   In California, a business that lets individuals sign up online for a payment that renews by itself must let them cancel online, through a button in their account or a ready-to-send cancellation email that the business provides and the customer sends without adding anything (California's automatic renewal law, read on 2 November 2026). Until your app is live, link such a pre-written email from your terms.

The rules for selling online

Before your first pre-order, two documents go online next to your payment link, each on a public web page. Your agent can draft them from what you're about to sell and put them online, on a simple web page it makes for you, or in your app builder's project if you've started one:

• Terms of sale: who you are, what you sell, the price, the date access opens and the monthly charge that follows, and how to cancel and get a refund. For individuals in the UK, this information is required before they pay. So is their right to cancel within 14 days without giving a reason, and whether they'd still owe reasonable costs for the service used before cancelling (GOV.UK).
• A privacy notice: in the UK, you "must provide privacy information to individuals at the time you collect their personal data": why you collect it, how long you keep it, and who you share it with (ICO).

Stripe asks only for each page's address, in your account's Public details, then links to both from the payment page.

When you create the payment link, also tick the option that asks buyers to accept your terms of service, Stripe's name for your terms of sale: it appears once you've given Stripe the address of your terms.

One contract comes later, when you sell to businesses.

If a business customer puts its own clients' personal data in your software, you hold that data for it. UK data protection law (UK GDPR) then requires a written contract between you, usually called a data processing agreement (DPA), saying you use the data only "in line with the controller's documented instructions", the controller being your customer (ICO). Sign it when a business customer asks, or before it puts that data in.

Is SaaS growing or dying?

One question follows every founder from the start: is SaaS dead? "SaaS is dead", the "SaaSpocalypse" as founders call it, has been the talk since early 2026. The reasoning: if an AI agent can rebuild software in a few hours, why pay a subscription every month?

Buyers of companies took it seriously. Private-equity funds are investment firms that buy companies: by Rob Walling's count, they buy about 70% of the software companies sold with $2 million to $20 million of yearly revenue. In 2026 they nearly stopped buying: "like nuclear winter again", he says.

Yet he sees revenue going the other way. "I've never seen more of a dissonance between sentiment and reality", he said in October 2026: the combined revenue of the companies his fund backs keeps rising. For Rob Walling, AI won't kill SaaS.

He estimates that 10 to 20% of software companies face "real struggle, real headwinds", while the others will survive, and many will thrive. He bets on software that customers trust enough to let it do the job itself, rather than only advise them on it.

The sharpest dissent comes from a founder. Simon Høiberg says that "the micro SaaS playbook", the recipe of the small one-person SaaS, "where you wrap a few APIs, put a monthly subscription on top and rent the entire operation is dying".

The APIs are other companies' services your software plugs into. Hosting, database, AI, payments and advertising each take their cut before the founder does.

That doesn't mean avoiding rented services at the start, the setup this guide describes. Høiberg's own advice is to "make every important dependency replaceable" and to own your distribution, an audience and an email list of your own, so that no single company stands between you and your customers.

What's threatened is software anyone can now rebuild, whose value lies only in its code or its screens. "Development now is a commodity", something anyone can buy cheaply, says Freddy Galindo, whose company now spends on flying its people out to learn how its customers work rather than on code.

His customers, mining contractors, are "12 to 24 months behind" on technology, he says. What makes Sprint Suite hard to replace is knowing how a mine site runs, which a competitor can't copy with a request to an AI.

And switching software gets easier when, as Boris Cherny notes, customers can ask an agent to move their data to a competitor: leaving you becomes as easy as copying you.

What still sells, according to these founders and investors:

• Software that does one precise task for one line of work, which only someone who knows that work can build.
• Your own way to reach buyers, such as an audience and an email list of your own.
• Software people use together, which is harder to replace. Cherny can build a messaging app in a few hours with Claude Code, but "it's still not useful because I can't talk to my friends."
• Software your customers trust to do a whole job for them, not just to advise them, a trust built over months of doing it right.

For you, two things really change. There are far more competitors, since anyone can build. And a SaaS business fetches little when you sell the company itself: about 1.5 times a year's revenue on TrustMRR in July 2026. SaaS isn't dead. What has gone is the advantage of being able to build: everyone can now, and the way you find customers makes the difference.

How much a SaaS makes, and how long it takes

Among the founders quoted in this guide, revenue in 2026 ran from nothing to about $165,000 a month. At the bottom, Tryreplify's first week brought nothing, and Swipefast's brought its first $49 payments.

At the top, DabbleWriter, writing software run by one person, brings in about $2 million a year, $165,000 a month. Between them, the founders' latest monthly figures run from about $10,800 to about $142,000, and the highest took years. Published figures leave out the projects that make nothing.

Three numbers tell you whether your SaaS is moving:

• Monthly recurring revenue (MRR): a customer who pays $290 for a year counts for about $24 a month. MRR isn't what you collect: the month a customer pays $290 for the year, you collect $290, then nothing from them for eleven months.
• Churn, the share of your subscribers who leave each month.
• Margin, the share of your revenue left once hosting, AI, tools and payment fees are paid.

From the price each customer pays and churn, you work out what a customer brings in over their whole subscription, called lifetime value (LTV). The average subscription length comes from churn: it's 100 divided by the churn rate.

If 5% of your subscribers leave each month, a subscriber stays 100 ÷ 5 = 20 months on average. At $30 a month, they bring in about 20 × $30 = $600.

Take away what that customer costs you (hosting, AI, payment fees, taxes): what's left is the most you can spend to win a new customer without losing money. What you actually spend to win one is called the customer acquisition cost (CAC). Beyond that ceiling, each customer costs you more than they bring in.

What to expect, from those who've been through it:

• Several projects before the right one. Marc Lou says that "90% of what I work on is basically going to end up nowhere because nobody wants to use them" (video of 31 August 2026).
• Often years, not months. Charlie Clark went back to a job at Squarespace, the website builder, in 2020, after a year on his own projects, and grew Links on nights and weekends alongside it. He only left that job in August 2024, once Links' revenue roughly matched his salary.
• Rare highs. Jacob Wright runs DabbleWriter, writing software for novelists, alone. It brings in about $2 million a year, with churn around 4% a month, very low for software sold to individuals, according to his talk at MicroConf in October 2026.

Selling your SaaS business. By August 2026, 144 founders had sold their business through TrustMRR, Marc Lou's marketplace. In July 2026, when 115 had sold, he said the average price was 1.5 times annual revenue: a SaaS bringing in $12,000 a year sold for about $18,000.

Most of these are small, young products.

If you might sell one day, use a domain name and an email address dedicated to your SaaS from the start: the buyer takes both with it.

Your action plan

Every action in this guide, in the order of the steps, to tick as you go.