Product, freelancer, agency or one engineer: who should build your system
Four options, and three of them are the right answer more often than the fourth. This page is written by the fourth, so it starts by making the case against itself.
By Eugene Romanov·Published
The short answer
Buy a product if your process resembles everyone else's in your trade, or if it has to work tomorrow with nothing adapted. Hire a freelancer if the job is one script and you can specify it precisely. Hire an agency if a committee approves the work, a security review is coming, or the system must outlive everyone who built it. Hire one engineer who does the whole thing when the calculation that decides your money is not in any product and you want to own the result outright. There are two speeds here, not one: an existing system can be started on now, from $99 a month, and shaped to how your company works in one to two weeks, and a system designed from your own process is from $2,000 over 30 to 45 days. Neither has a per-seat licence.
SHT-02What drives the number
What you are actually paying for in each case
With a product you are buying other people's debugging. It has broken and been fixed by thousands of companies doing what you do, and that is worth real money for a subscription. With a freelancer you are buying hours against a spec, which works when you can write the spec and fails when you cannot. With an agency you are buying coordination: a discovery phase, a designer, developers, a QA engineer and an account manager, and a meaningful share of that budget pays for those people to stay aligned with each other. That structure is not padding, it is what a committee-approved project needs. With one engineer you are buying the absence of coordination, which is the same money spent on the work instead.
The per-seat arithmetic, which nobody puts on a slide
Product pricing is quoted for one user. On Jobber's own pricing page on 2 September 2026, Grow lists at $149 a month billed annually with one user included and each additional user at $29, so a ten-person team is $410 a month, which is $4,920 a year. Housecall Pro lists Max at $299 a month with eight users and each additional at $75, so twelve people is $599 a month. Put the same ten people through our own subscription and it is the Team plan at $199 a month, each of the ten with their own access: owner, managers, crew, sales. The eleventh is $49. Working on your own, Pro is $99, and there is no per-user add-on on it: hiring a second person moves you to Team, because a crew needs separate access rather than a second login. That is the whole comparison in one line, $199 against $410 for the same ten desks. And a build is paid once and carries no seat licence of its own, so a team that grows from four people to nine does not change what you paid for it. Against $410 a month a build from $2,000 is roughly five months, and that comparison is honest only alongside the running costs below.
What a build still costs to run, said plainly
The database, the hosting, the mail sender, the phone number and the payment account are subscriptions, and they are on your own name rather than resold with a markup. They are much smaller than a per-seat plan and they are not zero. It also costs your own hours in the first week, answering questions about how your process actually works, and that part cannot be delegated to anyone, including me. A page that leaves both out is selling.
You talk to the person who writes the code
There is no account manager between the conversation and the commit, which is the main reason the prices and the timelines look the way they do. The cost of that is worth naming rather than hiding: three build slots a month, and when they are taken I say so instead of accepting a fourth and stretching everyone's dates. An agency does not have that limit, and for a company that needs guaranteed capacity that is a real advantage of theirs, not a flaw in their model.
A fix takes hours, because there is no queue in front of it
This is the part that is hard to believe until it happens. A bug, a rough edge or a small improvement is usually done the same day, in hours, because the person who wrote the code is the person reading the message. There is no ticket triage, no sprint boundary to wait for, no account manager relaying the description to somebody who has to reread the code first. With a product you file a request and it joins a roadmap you cannot see; with an agency a small change waits for capacity that was allocated last month. The same structure that limits how many builds happen in a month is the structure that makes a fix immediate.
Ownership is not a clause, it is where the accounts live
The repository, the database and the payment account are in your name from the first commit. Deposits land in your own Stripe account rather than passing through the studio. The provider subscriptions are yours, the invoices come to you, and nothing is switched off if we stop working together. With a subscription product your data exports but your workflow does not: the automations, the templates and the logic stay with the vendor, and rebuilding them elsewhere is the actual cost of leaving. That difference is worth more to some owners than to others, and it should be weighed rather than assumed.
The one calculation a product cannot model
Every trade has a number that decides whether a job earned anything, and a product built for every trade cannot model any one of them. In wallpaper it is the roll count, where the cut length rounds up to a whole number of pattern repeats and openings cannot simply be subtracted. In custom saunas it is a price that has to follow the drawing while the client is still watching. In car brokerage it is a profit rule that four screens must not each compute their own way. The test is blunt: if the number your money depends on is retyped into the CRM out of a spreadsheet, you are already paying for a custom system in labour, just badly.
The discipline is published, so you can check it rather than trust it
One fact has one owner: the rule for what a deal made exists once in the application and once in SQL, and the two move in one commit or neither moves, because a trigger and a screen computing different formulas is a bug that never announces itself. A guard test earns its place by failing first: every one of them was run against the code that shipped the original bug and had to catch it before being committed, and a test that has never failed is decoration. Empty and broken must never print the same sentence. Those are habits, not slogans, and each one exists because something went wrong quietly once.
Reuse is why the third system took weeks
Authentication, roles, documents, e-signature, payments, notifications and scheduled jobs are already written and already debugged across three systems in production. What gets built for you is the part that makes your trade different from the one next door, which is where the time actually goes. That is the mechanism behind a floor of $2,000 rather than a discount, and it is also the reason a fourth system in a fourth trade is a matter of weeks.
There are two speeds, and the fast one is not a compromise
A system designed from your own process takes 30 to 45 days, and that is the number on the ladder. It is not the only option, because the systems already running are real software rather than a portfolio: one of them can be started on now, and shaping it to how your company actually works is a matter of one to two weeks rather than a month and a half. That is the honest answer to "I need something before next month", and it is a different answer from "buy a product and bend to it": the adaptation is done by the person who wrote the thing being adapted. It is also priced by the work rather than by the floor. If the change is a day or two, it costs less than a build, and the $2,000 figure is for a system with pipelines and automations of its own. The trade is real. You are starting from somebody else's shape, which is a bargain when it is close to yours and a fight when it is not, and working out which of those it is takes one conversation.
Where this is the wrong choice, specifically
If it has to work tomorrow with nothing changed for you, buy the product: an adaptation is one to two weeks and a build is 30 to 45 days. If nobody inside your company will own how the system behaves after launch, buy the product: a custom system needs one person who will say how it should work. If a committee approves the purchase or a security review is coming, hire the agency. If you want enterprise SaaS, deep Shopify work, managed advertising, or a native app in the App Store, none of those are done here, and the mobile answer is an installable web app rather than a native build. Saying this costs a few enquiries a month and it is still the right way to write the page.
SHT-03The options, side by side
What each route actually costs you.
| Option | Price | Timeline | The trade-off |
|---|---|---|---|
| A product off the shelf | Per seat, per month, published by the vendor | Same day | Years of other people's debugging, support, upgrades handled for you. Your process bends to the product, and your trade's arithmetic lives in a spreadsheet beside it. |
| A freelancer | Hourly or per project | Days to weeks | Cheapest way to get one specified thing built. You supply the specification, the decisions and the integration, and the parts nobody scoped land back on you. |
| An agency | Agency budgets, quoted per project | Months | Capacity, process, and a project that survives its own staff turnover. A large share of the budget pays for internal alignment, which is worth it at committee scale and wasted below it. |
| A subscription to the existing system | Pro $99 a month working alone, Team $199 for ten, then $49 each | Start now, one to two weeks to fit | The fastest honest route when a system that already runs is close to how you work. You are adapting somebody else's shape, which is a bargain when it fits and a fight when it does not. |
| One engineer, end to end | From $2,000, no seat licence | 30-45 days | The conversation and the commit are the same person, so a fix lands in hours rather than in a queue, and you own the code, the database and the payment account. Three slots a month, so the calendar, not the price, is what runs out. |
SHT-04General notes
The questions people actually ask.
- N01Why would I not just buy Jobber or Housecall Pro?
- Usually you should. Buy the product if your process looks like your neighbour's in the same trade. The case for a build starts when the number that decides your money is not in the product, when per-seat cost at your headcount has passed what a build amortises to, or when you need to own the database outright.
- N02What is the real difference from an agency?
- Coordination. An agency's budget covers a designer, developers, QA and an account manager staying aligned, which is exactly what a committee-approved project needs and exactly what a five-person business is paying for and not using. The trade is capacity: an agency has it and one person does not.
- N03What happens if you are unavailable?
- The repository, the database and the payment account are in your name from the first commit, so nothing switches off and another developer can pick the code up. That is a design decision rather than a promise, and there is a page about exactly what ownership means here.
- N04Is a freelancer not cheaper?
- For one clearly specified script, yes, and that is often the right buy. The difference appears when nobody has decided what happens where two rules contradict each other: writing that decision down is most of the work, and it is not in an hourly quote.
- N05How is it faster and cheaper than the numbers I see quoted?
- Reuse and no coordination cost. Authentication, roles, documents, signatures, payments and scheduled jobs already exist and are already debugged in three production systems, so the custom part is the arithmetic of your trade. The third system took weeks rather than months for that reason.
- N06What does ten people cost on each option?
- On the Team plan, $199 a month, with each of the ten holding their own access. On Jobber's published Grow pricing, $410 a month. On Housecall Pro's published Max pricing, twelve people is $599 a month. A build of your own has no seat licence at all. Competitor figures are from their pricing pages on 2 September 2026 and do change.
- N07I am on Pro and I have hired someone. What now?
- You move to Team at $199 a month. Pro deliberately has no per-user add-on: it is the plan for working on your own, and the moment there is a crew the question stops being how many logins and becomes who is allowed to see what.
- N08How long does a fix or a small change take after launch?
- Usually hours, on the same day. The person reading your message is the person who wrote the code, so there is no triage step and no waiting for the next sprint. That is the other side of a studio that only takes three builds a month.
- N09How fast can something actually be running?
- An existing system can be started on now, and fitting it to how your company works is one to two weeks. A system designed from your own process is 30 to 45 days. Which of the two you want is usually obvious after one conversation about the arithmetic your trade runs on.
- N10How do I decide without paying anything?
- The free audit is a short recording, same day, no call: where enquiries are being lost and the one thing worth fixing first. If the answer is that a product fits you, that is what it will say.