Custom software vs off-the-shelf: how to decide

Off-the-shelf software wins when a process runs the same way in your business as in thousands of others: invoicing, payroll, bookings, a sales pipeline, a shop selling a standard range. The vendor spreads the cost of every tax change, security fix, browser update and phone app across all its customers, and your share arrives as a modest monthly fee.

The prices make the case. Xero's Grow plan in the UK is £39 a month before VAT, about €45, with no per-user licence fees (Xero). An invoicing system written from scratch would need decades of that fee to pay for itself, and somebody would still have to keep it in step with the tax rules. Shops work the same way: Shopify Basic is €19 a month billed yearly, or €27 month to month (Shopify), and the guide to building an online store compares it with the other platforms.

Before commissioning software, check whether an existing product covers the essential workflow and integrations. Compare the changes your team would need to make with the cost of building and maintaining a custom system.

Custom software is worth building when at least one of five things is true: the workflow is yours alone, the work is moving data between systems that don't talk, seat fees grow faster than the team's use of the tool, you need full control of your data, or the software is the product you sell.

• A workflow nobody else has: returns approved across three warehouses, or a pricing formula that took ten years to get right. Ready-made tools cope with custom fields until the workarounds become somebody's job.
• Orders arrive in Shopify, stock lives in a spreadsheet, invoices go out from Xero, and one person copies between the three every morning.
• Seat fees that follow headcount. They hurt most when the seats belong to people who use one screen of a large product.
• Control of the data: where it is stored, who can read it, and what happens to it the day you stop paying.
• The product itself. If customers pay for the software, renting its core makes a platform's price rises and rule changes part of your business model. For a phone app, what a mobile app costs covers the build.

Dekloro, which we built, does one job. It reads the report a Bulgarian resident downloads from Revolut, Trading 212 or Binance and turns it into the lines the Bulgarian annual return asks for: gains, dividends and crypto, each worked out in euros. The job is narrow and its rules come from one country's tax law, which is why it made sense as its own program.

The files are messy and the rules are exact. Each broker lays out its export in its own way; each sale is matched against the earliest purchases still held (first in, first out); and each transaction is converted at the euro rate for its own date, so any figure on the return traces back to one trade. Dekloro works from the downloaded file alone, with no broker login, and what it produces is a calculation rather than tax advice.

A careful person with a spreadsheet can do each of those steps. Doing all of them for every trade in a year, identically and without retyping a number, is where a small program earns its place. Fixed rules applied to untidy files, with one output at the end: that is the shape I look for first when a business asks what is worth building.

Over five years a subscription costs seat price × seats × 60 months, while a custom build costs the build plus 15–20% of it for each year after launch. On a €100 seat, a team of ten pays less on the subscription unless the build comes in under about €35,000; a team of fifty passes the top of the typical build range during year two.

The build range is the soft number, and it stays soft until you hold a quote: the published surveys stop at brackets. Most custom projects land in a medium band, $30,000–$100,000 and three to six months of work, according to a survey of more than 100 software companies worldwide run in September and October 2025, which puts Western firms at around $50,000–$250,000 for a medium project. In prices gathered from verified client reviews, projects typically cost $10,000–$49,999 but average $132,480, a gap that shows how long the expensive tail runs. A UK agency's guide from 30 July 2026 puts a standard business application at £30,000–£80,000, about €35,000–€93,000. For upkeep, the survey gives 10–20% a year and a software house's maintenance guide 15–20% of the original build. Dollar and pound figures are converted at the ECB's reference rates for 9 September 2026, and every price here was checked on 10 September 2026.

Read the table by rows. A €19 seat is hard to beat: forty people on monday.com's Pro plan, the largest team it prices without a quote (monday.com), cost €45,600 over five years, and only a build under about €28,500 comes in below that. A €100 seat is different. Fifty Salesforce seats cost €120,000 over two years, more than the dearest build in the range costs to make and run over the same two years (€103,200). At 25 seats, nearly the whole range breaks even by year five.

Holding seat prices flat flatters the subscription: Salesforce raised list prices for its Enterprise and Unlimited editions by an average of 6% on 1 August 2025, though it left Pro alone that time (Salesforce). What the table leaves out flatters the build. A subscription keeps adding features nobody would pay to have written for one company, and a custom system needs hours from your own staff to describe and test it, which no quote includes.

When cost savings are the main reason to build, I compare the subscription with development and maintenance over three years. I also calculate what happens if the team grows or the requirements change, because either can alter the result.

Yes: keep the off-the-shelf subscriptions and connect them, instead of replacing them with custom software. An automation service such as Zapier moves data between tools when something happens, a small custom program does the same through each tool's API, and a no-code builder such as Airtable or Microsoft Power Apps can hold a workflow that no ready-made tool covers.

Zapier's Professional plan starts at $19.99 a month billed annually for 750 tasks (Zapier). Airtable's Team plan is $20 per collaborator per month billed annually, with up to 50,000 records in a base, and Business is $45 with 125,000 (Airtable). Both price in dollars. At the ECB rate, $20 is about €17. Power Apps Premium costs €17.30 per user per month before VAT, paid yearly (Microsoft), or $20 in the US.

Look at the unit on the last two. Airtable bills everyone who can edit (read-only viewers are free), so fifty editors on Business come to $27,000 a year, about €23,000, which puts a no-code app back in the same table as Salesforce. The seat problem travels too.

The option I reach for first is the plainest: the shop stays on Shopify, the books stay in Xero, and a small program moves each day's orders across through Shopify's Admin API and Xero's Accounting API. It is the cheapest custom software there is, and switching it off leaves both tools exactly as they were.

For a new workflow, a small prototype can help define the brief. Record the fields, steps and exceptions as people use it, then use that experience to scope the application.

Three things move a no-code app to real software, and headcount is not one of them. The Team plan's 50,000-record ceiling forces the question for some teams; for others it is permissions that have turned awkward, or a seat bill that has crept past the price of a build. A base that works is worth leaving alone until one of those bites.

A focused first version of custom software takes one to six months. In a survey of software firms run in autumn 2025, small projects take one to three months and medium ones three to six. Client reviews of software projects put the usual timeline at about 13 months, a figure that includes large systems.

The way to stay near the short end is to make the first version small on purpose: one workflow, for one group of people, from the first screen to the last. In the returns example, the warehouse logs a return and the office sees it. Everything else waits. Approval rules, reports and anything customers touch belong to the second version, and what the first version's users do with it in their first month decides what gets built next.

I would rather launch something that handles most cases and passes the odd ones to a person than spend three more months on a version that handles every case on paper.

From your side, a build needs one person who knows the process and can answer a question within a day. The steps we follow, from written scope to launch, are on the software development page.

Get three things in writing either way: a way to take all your data out in a format another system can read, ownership of anything written for you, and enough documentation for a different developer to take over. For a subscription in the EU, the law now covers part of the first. For custom software, all three come from the contract.

The EU Data Act, applicable since 12 September 2025 (European Commission), covers cloud services, subscription software included. Under Regulation (EU) 2023/2854, the notice you give to leave can be no longer than two months, the switch itself must be done within 30 days, or seven months at most if the provider shows that 30 is technically unfeasible, and you then get at least 30 more days to retrieve your data (Article 25). Until 12 January 2027 a provider may charge no more than the switch directly costs it; from that date it may charge nothing (Article 29). Article 31 lifts those charge rules, and some technical duties, for services built for one customer and not offered at scale to others, so for custom software the exit terms and their cost belong in the contract. Whatever the law says, run the export on a trial account before signing, and open the file.

Code ownership is not automatic. Under UK law, copyright belongs to whoever wrote the code, or to their employer (Copyright, Designs and Patents Act 1988, s.11), and it passes to you only by an assignment in writing, signed by the owner (s.90(3)). US law lists nine kinds of commissioned work that a contract can declare “made for hire”, and software is not among them (17 U.S.C. § 101), so there too the contract needs an assignment. Bulgarian law gives commissioned work the same default: copyright stays with the author unless the contract provides otherwise (Copyright and Neighbouring Rights Act, Article 42). The clauses to ask for are set out in what a development contract should say about ownership.

Documentation gets skipped first. It never shows in a demo. The minimum I would want at handover is a page on how to build and release a new version, a list of every outside account (domain, hosting, payment provider, app stores) with what each costs a month, and every one of those accounts registered in your company's name. My test is simple: another developer should have the project running on their own computer within a working day, from the documentation alone, and ship a small change before the week is out.