Custom Software Development Cost (2025)

Published by Bles Software, a custom software and AI company based in Yehud-Monoson, Israel, building web apps, AI agents and API integrations for clients in Israel, the US, the UK and the EU.

Custom software cost is a wage bill, so start with published wages

Almost everything you will pay for custom software is somebody's time. Licences, hosting and tooling are real but small. So the honest way to read any quote is to convert it back into hours and then check those hours against a published rate, which anybody can do without taking a vendor's word for anything.

The United States Bureau of Labor Statistics publishes the distribution, not just an average, in its May 2025 Occupational Employment and Wage Statistics for software developers (code 15-1252), covering 1,687,890 jobs [1].

The handbook confirms the median separately at $135,980 a year for May 2025 [2]. Note what those figures show: the gap between the 25th and the 75th percentile is a factor of 1.6 on the same job title in the same country. A quote is not high or low in the abstract, it is high or low for a given seniority, and seniority is the variable vendors talk about least.

Wages are not what an employer pays

A wage figure understates an employed engineer by a predictable amount, and the Bureau publishes that too. For private industry workers in June 2026, total employer compensation costs averaged $46.89 per hour worked, of which wages and salaries were $32.82, or 70.0 per cent, and benefits $14.07, or 30.0 per cent [3]. That is all private industry rather than software specifically, so use the ratio rather than the dollar amounts: roughly three tenths on top of a wage before any overhead, recruitment, equipment or management.

This is why an in-house hour and a vendor hour are not comparable at face value. The vendor rate already contains benefits, idle time between projects, recruitment and the risk that your project ends.

What a non-US team costs, as published

Accelerance publishes annual rate bands, and its 2026 guide, released on 24 November 2025, is based on a survey of 60 software development partners worldwide [4].

Accelerance's own caveat is worth repeating because it argues against its own headline: it says hourly rates are a poor measure of true cost [4]. We agree, and the reason is the next section.

For Israel, where we are, the public picture comes through reporting of Central Bureau of Statistics data rather than a page we could open directly: Globes reported on 4 August 2026 an average monthly wage in the tech industry of NIS 35,760 in May 2026 against NIS 31,459 in May 2025, a rise of 13.7 per cent, with the national average monthly wage at NIS 15,223 in June 2026 [5]. That is a secondary report of an official figure and we flag it as such.

Why the hours, not the rate, decide the invoice

Multiply a rate by the wrong number of hours and the rate stops mattering. The published research on this is not flattering to anybody in our industry, and it is the most useful thing on this page.

Flyvbjerg and Budzier studied 1,471 IT projects and report an average cost overrun of 27 per cent, with one in six projects a "black swan" carrying a cost overrun averaging 200 per cent and a schedule overrun of almost 70 per cent [6]. That distribution is the point: the average is survivable, the tail is what kills a budget, and the tail is fat.

Read those two numbers against each other and a pattern appears that should change how you buy. In the black swan group the cost overrun averaged 200 per cent while the schedule overrun averaged almost 70 per cent [6], so cost ran away roughly three times as fast as the calendar. Projects do not usually run long and cheap. They run late and far more expensive, which is the signature of scope being added or discovered while the delivery date is being defended.

The widely repeated McKinsey and Oxford figures on large IT projects belong in this section and are deliberately not quoted here: their page could not be opened from our network on 9 October 2026, and we do not publish a number we have only seen summarised.

The lines that are not labour

Both stores have regional variations: Apple lists different terms for Brazil, the EU, Japan, the Netherlands, Russia and South Korea [8], and Google notes new regional fee structures from 30 June 2026 for the EEA, UK and US and 30 September 2026 for Australia and Japan, with different rates for new and existing installs [10]. If your revenue is consumer and mobile, that commission is a larger number than your build, and it is permanent.

On cloud, the only figure with a published methodology we found is an estimate rather than a measurement: Flexera's 2026 State of the Cloud report, published 18 March 2026, puts wasted cloud spend at 29 per cent, up for the first time in five years, from 753 cloud decision-makers' own estimates of their infrastructure and platform waste [11]. Self-reported waste is not a unit cost, and we quote it only as the scale of the slack most organisations believe they carry.

How to read the quote in front of you

Six questions, in this order. Each one is answerable before you sign.

  1. How many hours, at what seniority? Convert the total to hours and place it on the percentile list above [1]. A quote that implies 90th percentile engineers for a CRUD admin screen is mispriced, and so is one implying 10th percentile engineers for a payments migration.
  2. What is fixed and what is discovered? The overrun evidence says the calendar holds and the budget does not [6], which happens when scope is open. Name what is fixed.
  3. Who owns the integrations? Connecting to systems you do not control is the single most common source of the discovered work in point two.
  4. What is the commission on the revenue? If it ships to a mobile store, 15 or 30 per cent of the revenue is already spoken for [8][10].
  5. What does year two cost? Nothing in the cards [4] covers maintenance, and it does not become free.
  6. What is the smallest thing that proves it works? The cheapest defence against the fat tail [6] is shipping something real early enough to change your mind while changing your mind is still cheap.

We quote fixed price where scope can genuinely be fixed, and we say so when it cannot, because a fixed price over an open scope is just an argument scheduled for later.

Sources

More Costs and Timelines from Bles Software