We Gave 40 Employees $150 a Month to Buy Their Own Software. The Receipts Were Wild.
Photo by Photo by Vagaro on Unsplash on Unsplash
Shadow IT has a bad reputation, and some of it is deserved. But most of the hand-wringing about rogue software assumes the alternative is a clean, centralized process. Ours was not clean. It was just slow.
So we ran an experiment. From August through October 2026, we gave 40 employees a personal SaaS allowance of $150 a month, a company card with a category restriction, and one rule: anything you buy shows up on the shared list. No approval queue, no ticket, no manager sign-off.
Then we audited everything.
The rules of the sandbox
Total budget across three months: $18,000. People could roll unused allowance into later months. They could not pool it into a single expensive seat unless the whole group agreed, and two teams did exactly that.
The clause that made the whole thing work: visibility, not permission. Every purchase appeared in a shared sheet within 24 hours, with a one-line note about what problem it solved. That note turned out to be worth more than any security questionnaire.
What people actually bought
By the end of the quarter, the group had signed 63 distinct paid subscriptions. Most were small. A lot were cheap. A handful were surprisingly expensive.
- Note-taking and docs: seven distinct apps across 40 people. Seven.
- AI assistants and meeting recorders: nine subscriptions, four of which were functionally identical.
- Design and prototyping: five tools, with heavy overlap in the two-person design group.
- Developer utilities: API clients, log viewers, code review helpers, and a surprisingly popular CLI tool nobody in IT had heard of.
- Support and scheduling: the usual suspects, plus one $9 tool that three people described as "the only reason I still do this job."
Actual spend: $14,260, about 79% of the budget. Nine subscriptions got cancelled by the buyer within the first month, unprompted. That is a churn rate no centralized procurement process we have run has ever matched, because nobody had to justify the decision to anyone.
The overlap problem is real
We grouped the purchases by job-to-be-done and found roughly 31% of total spend landed in categories where at least two people were paying for the same capability. Some of that is defensible. Two people solving the same problem differently is not automatically waste.
But some of it was pure duplication driven by nothing more than not knowing a colleague already had a seat. The shared sheet fixed that over time, slowly, the way social pressure fixes things.
The security gaps we could not ignore
This is where the experiment got uncomfortable. Twenty-two of the tools requested OAuth access to Google Workspace. Eleven had persistent read access to Drive. Two uploaded customer records to vendors with no SOC 2 report and no data processing agreement on file.
None of that was malicious. It was convenience. Employees picked the tool that let them connect their calendar in one click, because the alternative was a ticket that takes weeks.
We closed the gaps with three cheap moves:
- An allowlisted identity provider so new tools authenticate through SSO instead of a shared password.
- A one-page DPA template buyers could send the vendor themselves, no legal queue.
- A monthly five-minute review of the shared sheet, focused only on tools touching customer data.
That did not eliminate risk. It made risk visible, which is most of the battle.
Spend creep is mostly subscription creep
Nobody in this experiment bought a $500-a-month platform. They bought eleven $9-to-$29 tools. Small purchases do not trigger scrutiny, and that is exactly why they accumulate.
The creep was not caused by reckless spending. It was caused by forgetting to cancel. Six people kept paying for tools they described, in their own words, as "fine but not needed." Nobody had a reminder, so nobody made a decision.
A calendar reminder at the 30-day mark would have recovered an estimated $600 over the quarter. That is not a policy problem. That is a housekeeping problem.
Where the real gains showed up
The productivity wins were small, specific, and almost invisible from the outside. One engineer automated a weekly report that used to take an hour and now takes zero. A support rep bought a $12 text-expander that cut her handle time enough to clear the queue before lunch. A designer picked a plugin for a task our centralized stack did not cover at all.
None of those would have survived a formal business case. Each one survived because the person closest to the problem could just solve it.
The uncomfortable conclusion
Our internal procurement process had a median turnaround of 34 days for a software request last quarter. The median purchase in this experiment took about four minutes.
When the gap between those two numbers is that wide, shadow IT is not rebellion. It is a workaround for a system that has priced itself out of usefulness. People are not buying software because they want to break the rules. They are buying it because their job is happening right now and the ticket is not.
What we would keep
If we extended this, we would keep the allowance, keep the visibility sheet, and tighten only the identity layer. We would add a 30-day auto-nudge on every subscription and a mandatory DPA send for anything touching customer data.
The thing we would not bring back is the approval queue. The data does not support it. Decentralized buying, with a light guardrail and honest visibility, produced less waste and more genuine utility than our centralized process did in the same period.
Shadow SaaS is a symptom. Fix the speed of the front door and most of the shadowing goes away on its own.
