System Alphas

The real cost of software: what the research says about building it right.

Large software projects run over budget, old code quietly drains every team, and AI is changing how software is built. We gathered findings from McKinsey, the University of Oxford, Stripe and GitHub, and added what each one means for your systems.

5 sources2012–2023 research6 min readBy System Alphas
1. Big-bang projects overrunLarge IT projects average 45% over budget and deliver 56% less value.
2. Tech debt taxes growthTech debt can equal 20–40% of the value of a company’s whole technology estate.
3. AI changes the economicsGenerative AI cuts time on new code and documentation by about half, except on complex work.
McKinsey · 2012McKinsey · 2020Stripe · 2018GitHub · 2022McKinsey · 2023
Finding 01McKinsey & Company · October 201245%average budget overrun on large IT projectsDelivering large-scale IT projects on time, on budget, and on value (with the University of Oxford)

What the research found

  • Research on more than 5,400 IT projects, conducted with the University of Oxford.
  • Large IT projects average 45% over budget and 7% over schedule.
  • They deliver 56% less value than predicted, and 17% go so badly they threaten the company’s existence.

What it means for you

Big-bang software projects are risky by design. Small, staged releases with clear value checkpoints keep budgets and outcomes on track.

How we apply it

We deliver in two-week sprints with working releases you can review.

See our process Read the source
McKinsey Digital, October 1, 2012.
Finding 02McKinsey & Company · October 202020–40%of the entire technology estate’s value is tech debtTech debt: Reclaiming tech equity

What the research found

  • Survey of 50 CIOs at financial-services and technology companies with revenue above $1 billion.
  • 10–20% of the technology budget for new products is diverted to resolving tech debt.
  • 60% of CIOs said their tech debt rose noticeably over the past three years.

What it means for you

Old systems and shortcuts quietly tax every new project. Paying debt down steadily frees budget for growth.

How we apply it

We modernise legacy systems step by step, without stopping the business.

See Custom Software Read the source
McKinsey Digital, October 6, 2020.
Finding 03Stripe · September 201842%of a developer’s week goes to maintenance, tech debt and bad codeThe Developer Coefficient (reported by ADTmag)

What the research found

  • Harris Poll survey of over 1,000 developers and over 1,000 C-level executives in five countries.
  • Developers spend about 17 hours of a 41-hour week on technical debt and bad code.
  • Bad code costs an estimated $85 billion a year in lost opportunity worldwide.

What it means for you

Quality built in from the start is cheaper than fixing it later. Clean architecture and testing protect your future roadmap.

How we apply it

Automated tests and code reviews run in every sprint.

See QA & Testing Read the source
ADTmag, September 10, 2018, reporting Stripe’s Developer Coefficient.
Finding 04 AI effectGitHub · September 202255%faster task completion with AI-assisted developmentResearch: quantifying GitHub Copilot’s impact on developer productivity and happiness

What the research found

  • Controlled experiment with 95 professional developers.
  • Average completion time fell from 2 hours 41 minutes to 1 hour 11 minutes.
  • The result was statistically significant (P=.0017).

What it means for you

AI-native engineering teams ship more with the same headcount, which lowers cost and time to value.

How we apply it

Our AI-powered development ecosystem speeds up delivery with senior review.

Why System Alphas Read the source
GitHub Blog, September 7, 2022 (updated May 2024).
Finding 05 AI effectMcKinsey & Company · June 2023~50%less time to write new code and documentation with generative AIUnleashing developer productivity with generative AI

What the research found

  • Lab study with more than 40 McKinsey developers in the US and Asia.
  • Documentation took about half the time, new code nearly half, and refactoring under two-thirds.
  • On highly complex tasks, time savings fell below 10%; some junior developers were 7–10% slower.

What it means for you

AI speeds up routine engineering a lot, but complex architecture still needs experienced people. The best results come from AI plus senior engineers.

How we apply it

Senior engineers lead the architecture; AI accelerates the routine work.

See Custom Software Read the source
McKinsey Digital, June 27, 2023.

Sources

  1. McKinsey & Company, Delivering large-scale IT projects on time, on budget, and on value (with the University of Oxford), October 2012. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/delivering-large-scale-it-projects-on-time-on-budget-and-on-value
  2. McKinsey & Company, Tech debt: Reclaiming tech equity, October 2020. https://www.mckinsey.com/capabilities/platforms-and-solutions/our-insights/tech-debt-reclaiming-tech-equity
  3. Stripe, The Developer Coefficient (reported by ADTmag), September 2018. https://adtmag.com/Articles/2018/09/10/developer-survey.aspx
  4. GitHub, Research: quantifying GitHub Copilot’s impact on developer productivity and happiness, September 2022. https://github.blog/news-insights/research/research-quantifying-github-copilots-impact-on-developer-productivity-and-happiness/
  5. McKinsey & Company, Unleashing developer productivity with generative AI, June 2023. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/unleashing-developer-productivity-with-generative-ai

Figures are quoted from the publishers listed above; follow the links for full methodology. System Alphas is not affiliated with these organizations. Sections titled “What it means for you” and “How we apply it” are our own analysis.

Build it right

Software that ships on time and stays easy to change.

Discovery gives you a fixed scope, architecture and estimate before any build.