Pixel Street :: How to Start a Web Design Company, From Someone Who Lost Two First

Pick one kind of client you can credibly claim, sell judgement rather than production hours, register properly and get your GST position right before you invoice anyone, price against your own cost of delivery instead of against marketplace listings, and expect your first clients to come from people who already know you. Everything below is the detail behind that sentence, plus the parts of the standard advice I think are now wrong.

Most guides to starting a web design company are written by people still running their first one. I lost two companies before Pixel Street, so read this as a list of the things that actually kill you rather than a list of steps that make the process look orderly.

It was not orderly. The order below is the order I would use if I were starting again in 2026, which is not the order most of the internet still uses.

One warning before any of it. The market you are entering is not the market these guides describe. The bottom of it has been taken by software, and pretending otherwise is the most expensive mistake available to you right now.

The short answer

Pick one kind of client you can credibly claim, sell judgement rather than production hours, register properly and get your GST position right before you invoice anyone, price against your own cost of delivery instead of against marketplace listings, and expect your first clients to come from people who already know you. Everything below is the detail behind that sentence, plus the parts of the standard advice I think are now wrong.

The reason you were given to start is not evidence

Nearly every guide to this opens by telling you that employment for web designers is expected to grow significantly by 2030. It is worth spelling out why that number should not move you, because you will meet it on a lot of pages.

It is a United States Bureau of Labor Statistics occupational projection. Its horizon has effectively arrived. It counts people employed with the job title of web designer, which is not the same thing as demand for a web design business, and it describes a labour market on the other side of the world from most people reading this page. Three problems, any one of which disqualifies it. Using it as encouragement to start a company in Kolkata is an argument dressed as evidence.

The evidence that does exist points the other way at the bottom of the market, and it is worth knowing before you commit a year of your life.

Xiang Hui, Oren Reshef and Luofeng Zhou tracked freelancers through the release of ChatGPT, DALL-E 2 and Midjourney using Upwork's own data, and published the result in Organization Science in November 2024 (The Short-Term Effects of Generative Artificial Intelligence on Employment). Freelancers in the most exposed occupations lost work and lost earnings. The finding that should concern you most, if your plan is to be better than the competition, is that being better did not help. The authors found no evidence that strong past performance moderated the effect, and suggestive evidence that top freelancers were hit disproportionately.

The platforms are not arguing. Fiverr cut 250 people in September 2025, roughly 30% of its own staff, and its chief executive Micha Kaufman described the goal as an AI-first company that is leaner and faster with a smaller team. The marketplace built on cheap human production hours stopped believing in cheap human production hours.

I want to be precise about what that means, because the doom version is also wrong. It does not mean there is no business here. It means the entry-level version of this business, the one where you undercut everybody and produce brochure sites quickly, is now competing with a subscription. I have argued the full version of that case in AI website builders versus custom development, including where the builders genuinely win and why we still rebuild their output most months. Read it before you decide what to sell, not after.

What actually kills a web design company

Not competition. In my experience the three that do the damage are cash, concentration and dependence.

Cash. Design work is delivered before it is paid. A company can be profitable on paper and dead in the bank on the same Tuesday. Advance payment is not a rudeness, it is the difference between a company and a hobby that owes money.

Concentration. One client at 60% of revenue is not a client, it is an employer who can fire you without notice and without severance. The month that client restructures its marketing team is the month you find out.

Dependence. I have written before about rented land. Never build your main asset on ground someone else controls. For a studio that means your leads, your portfolio and your client relationships should not live entirely inside a platform that can change its rules on a Tuesday, whether that platform is a marketplace, a social network or a search engine.

None of the three are design problems. That is the point. Nobody has ever lost a web design company because their kerning was poor.

[IMG: Three things kill a web design company. Cash: the work is delivered before it is paid, so you can be profitable on paper and dead in the bank on the same Tuesday. Concentration: one client at 60% of revenue is an employer who can fire you without notice. Dependence: your leads and relationships living on a platform that can change its rules.]

Pick something to be known for

The standard advice is to choose a niche, and it is right, but the reason usually given for it is wrong. A niche is not about focus or passion. It is about being describable.

Let me tell you what specificity looks like. Four years ago I made a cold call to ITC with a fancy deck and zero credibility. Today we design for them, and for Coca-Cola and Marico. That sentence describes exactly one agency. That is the kind of fact a machine can attach to a name, and it is the kind of fact a human remembers well enough to repeat to somebody else in a lift.

"We build beautiful, responsive websites for businesses of all sizes" describes forty thousand agencies and is therefore worth nothing to a person and nothing to a language model. If you are wondering why an AI assistant recommends a competitor instead of you, the answer is usually that the competitor is describable and you are not, and I have set out the diagnostic in why ChatGPT recommends your competitor.

You do not have to be right first time. You have to be specific first time, then correct it. A niche you can defend in one sentence beats a positioning statement you have to explain.

Decide what you sell, and what you refuse to sell

Every guide gives you the same service list: web design, UX, SEO, content, maintenance. The list is fine. The decision it hides is which of those you will do badly because a client asked nicely.

My own rule, arrived at expensively, is that a studio should sell the thing it can be held to. We sell design and build, and the thinking that goes in front of them. What we do not sell is a service we would have to subcontract while pretending we had not, because the day something breaks, the client is talking to us and we are talking to somebody who has moved on. There is nothing wrong with subcontracting. There is something wrong with concealing it, and I have written about the honest version of that arrangement in the white label web development guide.

One service is worth more attention than it usually gets. Maintenance is the least glamorous line on your rate card and the only one that pays you in a month when nobody signs anything. A studio with a maintenance base survives a bad quarter. A studio living on new projects finds out what a bad quarter is.

Pricing, and why the marketplaces are the wrong benchmark

The standard advice is to research market rates on Glassdoor and Upwork. It is worse than useless, and here is why.

Glassdoor reports what employees are paid. You are not selling a salary, you are selling a project that has to carry your software, your rework, your unbilled meetings and the months without a signature. And marketplace listings are gross of fees the listing does not mention. If you benchmark against a number that has not had its fees taken out, you have benchmarked against a number nobody receives.

What the marketplaces actually take, from their own documents

Two things follow from that table.

The first is arithmetic. A $5 gig costs the buyer about $8.78 once the small-order fee is applied, which is a 75% markup on the advertised price, and you still only see $4. The $5 floor is a doorway, not a price. Fiverr's own annual report on Form 20-F, filed with the SEC on 12 March 2026, reports annual spend per buyer of $342 and says 66% of marketplace revenue comes from buyers spending more than $500. The marketplace in the advertising and the marketplace in the filing are different places, and the filing is the one written for people who can sue.

[IMG: The advertised $5 gig costs the buyer about $8.78 once the 5.5% plus $3.50 small-order fee is applied, a 75% markup, while the seller receives $4 because Fiverr takes 20% of the order. The $5 floor is a doorway, not a price.]

The second is that the Upwork number half the internet still prints is wrong. The 20/10/5 sliding scale has been gone for years, and the flat 10% that replaced it is itself out of date. Upwork's FY2025 annual report on Form 10-K says the flat 10% talent service fee is maintained only for contracts formed prior to May 2025. Everything since sits on the variable rate, locked at the moment the contract is created, which means the fee you agreed to on your first contract is the fee you are stuck with on it.

Price against your own cost of delivery instead. Work out what a month of your studio costs to keep open, decide how many productive days that month actually contains once you subtract selling and admin, and let the client's budget tell you what scope fits rather than what discount applies. If you want the ground-level version of what the Indian market pays and what moves a quote, I have published our own numbers and the pages I read them off in what web design actually costs in Kolkata.

Registering it in India

Most guides to this tell you to choose between an LLC and a corporation, which are American structures. If you are reading this in India, your realistic options are a sole proprietorship, a partnership, a limited liability partnership, a one person company or a private limited company, and the choice is mostly about liability and about who you want to be able to invoice.

What I will not do is print a registration cost. mca.gov.in blocks automated retrieval, and every figure I could find in search came from a firm that sells incorporation services, which is not a source. A number I cannot read off a government page does not go on this page. Ask a chartered accountant in your own state, because stamp duty is a state matter and the total genuinely differs.

The number I can source is the one that catches people out.

Web design and development is taxed at 18% GST, under Heading 9983 of Notification No. 11/2017-Central Tax (Rate). Registration is compulsory once your aggregate turnover in a financial year crosses ₹20 lakh, or ₹10 lakh if you are supplying from one of the special category States. Section 22(1) of the CGST Act says so in terms. The ₹40 lakh threshold that comes up constantly in conversation is real, but it sits in a later proviso that applies only to a supplier engaged exclusively in the supply of goods, so it has nothing to do with you.

Two practical consequences. Below the threshold you cannot issue an invoice a GST-registered client can claim against, which quietly makes you the more expensive option for exactly the clients you want. Above it, 18% appears on every invoice and you need to have said so in the quote, because a ₹1,00,000 proposal inclusive of GST and a ₹1,00,000 proposal plus GST are written identically and differ by ₹18,000. The notification, the SAC codes and the arithmetic are set out in the Kolkata cost breakdown.

The software, and two recommendations you should ignore

Startup guides still routinely recommend Adobe XD and QuickBooks. Both recommendations are stale, and one of them is impossible.

Adobe XD is in maintenance mode. Adobe has said it has no plans to invest further in the product and that it is no longer sold as a single application to new customers, as reported in January 2024. Building a new studio's design workflow on it in 2026 would be a strange thing to do.

QuickBooks is not available in India. Intuit closed new sign-ups in July 2022 and ended subscriber access to QuickBooks Online, QuickBooks Online Accountant, the mobile app and QuickBooks Time on 30 April 2023. Telling an Indian reader to run their books on it is telling them to buy something they cannot buy. Zoho Books is sold here as GST-compliant accounting, with e-invoicing, e-way bills and return filing on the box, which is the actual requirement. Your accountant will have a preference that matters more than mine, and I would take theirs.

On the rest of it, my advice is duller than the standard list. Buy nothing in month one that you are not already billing against. A shared drive, an invoicing tool your accountant accepts, one place where tasks live, and whichever design tool you are already fast in. Trello is enough for a two-person studio and so is a spreadsheet. Tool choice has never been the constraint on a young studio. Selling has.

Where the first clients come from

Almost always from someone who already knows you, and almost never from the channels that get listed here.

Dribbble and Behance are portfolios, not lead sources. They are where a client confirms you are real after they have already heard your name, which makes them worth maintaining and not worth waiting on. The same is true of a shiny logo. Your identity work matters for the day someone checks you out, not for the day they decide to.

What actually generates the first ten conversations is asking specific people for specific work, publicly doing the kind of work you want more of, and being findable when someone searches for the thing you said you were. That last one has changed shape. Buyers now ask an assistant for a shortlist as often as they type a query, and the studios that get named are the ones with facts attached to them. The playbook we use for that is in how to get your brand mentioned by ChatGPT, and the search fundamentals underneath it are in the SEO checklist.

One thing worth doing before you have any clients at all: write down the questions you get asked in sales conversations and answer them in public, at length, with real numbers. It is slower than advertising and it compounds, which advertising does not.

When to stop being one person

The honest test is not revenue. It is whether the work in front of you has to happen at the same time or can happen in sequence. One person can do an enormous amount of work in sequence. One person cannot be in two conversations at once, and that is the wall, not the workload.

You have three routes past it and they are not equivalent. Hire, which is the most expensive and the only one that builds a company. Subcontract openly, which is fast and works until the day the subcontractor is busy. Or partner with a white label studio, which is the formal version of the same thing with contracts around it. I have written about the choice from the client's side in freelancer versus agency, and it reads differently once you are the one being chosen.

Questions I get asked

Do I need to register a company before I take my first project?

No, and most people do not. What you need before your first invoice is a clear answer on GST, because that determines whether a registered client can claim your tax and therefore whether your quote is really cheaper than the studio quoting more. Registration of a company and registration for GST are separate decisions and people conflate them constantly.

Is it still worth starting a web design company in 2026?

If you are selling production, no, and I would say the same about my own studio. Generating pages is cheap now. If you are selling judgement, positioning, and the systems around a website, then yes, and there is arguably less competition than there was, because the people who only knew how to produce are leaving.

Should I start on Upwork or Fiverr?

They are a reasonable place to find your first few projects and a bad place to build a business, for the same reason renting is a reasonable place to start and a bad place to end. Use them to learn what people will pay for, then move the relationship somewhere you control. Read the fee schedule first, so you know what your quote actually pays you.

What should I charge for my first project?

More than you want to, and against a written scope. The first quote most people give is set by nerves rather than by cost, and the client who accepts an underpriced quote is rarely the client who upgrades later. Work out your monthly cost to stay open before you name a number.

Do I need an office?

Not to do the work. Possibly to be believed, depending on who you sell to, and that is a marketing decision rather than an operational one. Decide which of the two you are buying before you sign a lease.

What I would tell you across a table

Start smaller than you think and be more specific than feels comfortable. The two things that will nearly finish you are running out of cash and depending on one client, and both are visible months in advance if you are willing to look. The design part is the part you already know how to do. The company part is the part nobody teaches you, and it is most of the job.

At Pixel Street we are a web design company in Salt Lake, Kolkata. We got here after two failures and one cold call, which is a less tidy story than the twelve-step version, and a more useful one.