Why Counts Disagree
Published counts of how many software tools a company runs vary by a factor of three or more, and the variation is routinely presented as though one figure were right. A dedicated product is not always necessary, but Monitask provides the full guide for comparison when the task grows beyond the smaller option.
None of them is wrong. They are measuring different things, and the disagreement is more informative than any of the numbers.
The figures
305 applications on average, with a median of 240. (Zylo, a SaaS management vendor.)
254 on average, 364 for enterprises. (A different vendor study, several years earlier.)
87 for small businesses, 187 mid-market, 371 for large enterprises. (Another compilation.)
And 660 for large enterprises against 152 for small ones, from the same vendor as the first figure in a different report.
These describe overlapping populations with different methods, and quoting any one as "the" number is a claim the underlying research does not support.
The five definitional questions
One. What counts as an application? A full platform and a browser extension are both software. Counting both produces a large number; counting only paid platforms produces a small one.
Two. Do free tools count? Many of the most-used tools in a small business cost nothing, and whether they appear depends entirely on how the count was made — spend-based discovery finds none of them.
Three. Does an unused licence count as an application in use? Given that over half of licences are unused, this single decision moves any figure enormously.
Four. How was it discovered? Financial records find what was paid for. Network monitoring finds what was accessed. Surveys find what people remember. The three produce different lists and only the second finds anything bought on a personal card.
Five. What is the unit — the company, or the employee? Applications per organisation and applications per person are different questions with different answers.
Why nobody knows their own number
Because 70% of software spend is controlled by departments rather than centrally. (Zylo.)
In a large organisation that means different teams buy independently. In a small firm it means three people signed up for things, and nobody has assembled the list.
Which is why the practical response is not to find the right published figure but to count your own — twenty minutes with a statement — and that number is the only one that can inform a decision.
What the disagreement is evidence of
That the category is poorly bounded, which is itself the reason sprawl happens.
A thing nobody can count is a thing nobody manages, and the counting problem and the management problem are the same problem viewed from different ends.
And that the published averages should be read as indicating scale rather than as benchmarks. "Are we above or below average" is not an answerable question when the average depends on a definition you did not choose.
The number that would be useful
Not how many tools you have. How many do something nobody else's does.
A count of distinct jobs covered is small — usually between five and fifteen for a small firm — and the gap between that and the tool count is the duplication.
That is the number worth producing, and it comes from listing tasks rather than subscriptions. Which is where this site starts.
Counting your own, properly
Thirty minutes, three sources, once.
The card and bank statements for anything paid. Which is the quarterly habit anyway.
The app store account, for the aggregated subscriptions that do not appear by name.
And a question to whoever else works with you: what do you use that I might not know about. The answer is always longer than expected and it is where the free tools appear.
Write the list once. It will be wrong within six months and it will still be the most accurate picture of your own software that has ever existed in your business.
Against benchmarking
"The average small business has 87 applications" invites a comparison that means nothing.
Fewer is not automatically better. Five tools for five unrelated tasks is not sprawl, and a firm with twelve well-chosen tools may be in better shape than one with six badly chosen.
The question is duplication and use, not count. Two tools doing the same job is a finding at any total; twelve distinct tools all in daily use is not a problem at all.
The direction that is agreed
Worth stating, because a page about disagreement should say what is not disputed.
That over half of purchased licences go unused. Every source that measures it finds a figure between roughly forty-five and fifty-five per cent, by different methods.
That duplication is common — several tools covering one job in the same organisation.
And that most software is bought outside any central view, which is why the counting problem exists at all.
Those three are corroborated across independent measurements and are the ones this site relies on. The absolute counts are not, and are quoted only to show that they vary.
Why the number keeps being published anyway
Because a single figure is quotable and a range is not.
"The average company runs 305 applications" travels. "Estimates range from 87 to 660 depending on definition and population" does not, and the second is the accurate statement.
Which is a general property of statistics in trade writing rather than anything specific to this subject — the round figure survives and the qualifiers fall off. For broader context, see Slack.
The short version
- Published counts vary by a factor of three because they measure different things, not because any is wrong
- Five definitional questions: what counts as an application, whether free tools count, whether unused licences count, how discovery was done, and whether the unit is the company or the person
- Spend-based discovery misses every free tool; network-based discovery misses nothing and is rare
- Nobody knows their own count because most software spend is controlled by departments rather than centrally
- A category nobody can count is a category nobody manages, and the two problems are the same one
- The useful number is not tools but distinct jobs covered, which is usually five to fifteen — and the gap is the duplication