Every number on this site, with the working.

The headline number

What we mean by "lost productivity". Not a vague morale figure. We mean countable staff time spent moving the same information between systems that don't talk to each other: the emails retyped into a job system, the figures exported from one tool and pasted into another, the payments chased by hand, the weekly report rebuilt from scratch every Monday. Hours you already pay for, spent on work that adds nothing a computer couldn't do. Our longer write-up on re-keying traces why no national dataset measures this — and why the only honest figure is one you can count.

How we get there — three honest steps

1. Price one hour a week.

   We use the same method HMRC uses to value admin time: hours, multiplied by an ONS median wage, plus employer National Insurance. Over 46 working weeks, one hour a week of manual work costs between £765 and £1,016 a year — depending on who does it.

   Who does the loop	Median hourly (Apr 2025)	Loaded (+15% NI)	1 hr/week, for a year
   Data-entry administrator	£14.45	£16.62	£765
   Book-keeper / payroll clerk	£16.33	£18.78	£864
   Office manager	£19.20	£22.08	£1,016

2. Multiply across the team.

   A ten-to-twenty person firm almost never has one person doing this. It's spread across whoever handles admin, book-keeping, operations and the reports. So add up the whole firm's manual hours in a normal week and price them at a book-keeper's loaded wage (£864 a year per weekly hour):

   Manual hours, firm-wide, per week	Cost per year
   20 hours	≈ £17,000
   30 hours	≈ £26,000
   45 hours	≈ £39,000
   60 hours	≈ £52,000

   To put the top row in human terms: 60 hours a week across a 20-person firm is three hours each — under 5% of the working week — moving information by hand. Or a smaller firm where a handful of people lose a lot more than that.

3. Read the ceiling honestly.

   £50,000 is the upper end, not a typical figure — roughly the twenty-staff, manual-heavy case above. A ten-person firm more often lands between £15,000 and £26,000. That is exactly why the claim says up to, and why we never dress it as an average.

The other numbers

The rest of the figures on the site.

These are context stats — facts about the market, not claims about our own results. Each is a single link to its primary source on the page where it appears; here they are in one place, with the labels kept on.

• "One in six big software builds runs to around 200% over budget."

  Flyvbjerg & Budzier, Why Your IT Project May Be Riskier Than You Think, Harvard Business Review, 2011 (n=1,471 projects). The authors' own remedy is to build in small modules — which is the point we make with it. Full text.

  Where — home page, "the three bad options"

• "When someone finally audits that spreadsheet, it nearly always turns out to hold errors."

  Raymond Panko, What We Don't Know About Spreadsheet Errors Today, EuSpRIG 2015 — a compilation of field audits finding errors in the large majority of operational spreadsheets audited. Scoped, honestly, to audited spreadsheets. Full text.

  Where — home page, "the three bad options"

• "Chasing overdue invoices alone runs the average affected firm 86 hours a year."

  London Economics for the Department for Business and Trade and the Office of the Small Business Commissioner, Late Payments Research, July 2025 (survey n=1,455, plus econometrics). Landing page.

  Where — home page, pricing

The header line "20 years building business software" is biography, not a statistic — it describes a track record of shipped systems (child-protection casework, an organisational social network, recruitment analytics, credit-note automation, and most recently a full operating platform for one firm), not a measured outcome. Measured outcomes trace to that last deployment, and we say which numbers those are.

Why bother with a page like this? Because the whole argument of this business is that most firms are running on numbers nobody has checked — and the fix starts with counting honestly. We'd be poor advertisements for that if we didn't do it ourselves. The model, in 90 seconds →