The Open-Tab Problem with AI Subscriptions

58% of small businesses now use at least one AI tool, up from 29% in 2023, according to the U.S. Chamber of Commerce. That doubling happened fast enough that most founders never stopped to ask whether each subscription earned its slot. They just kept adding.

The frequency check feels like a reasonable shortcut. If you open something every day, it must be working. The UK government's AI adoption research supports this instinct: among businesses already using AI, over 80% interact with it weekly. Regular use looks like integration.

Why opening a tool every morning proves nothing

The OECD's 2026 D4SME survey found 76% of SMEs use AI in a shallow way, relying on off-the-shelf tools for simple tasks. High weekly use and shallow integration describe the same firms. A founder who opens an AI writing tool every morning to generate a single social post uses it constantly and extracts minimal value relative to cost.

Frequency confirms a habit formed. It does not confirm the habit displaces paid time.

The U.S. SBA's 2025 research also notes that founders subscribe out of curiosity or peer pressure, then engage sporadically and never build dependable workflows. Breeze and Vertice data from 2026 puts 53% of SaaS licenses at idle, with waste rising more than 14% year over year, while spending on AI-native tools grew more than 75% in the same period. Spend and use moved in opposite directions.

The test frequency cannot run

The question worth asking is not "how often do I open this" but "what specific recurring task would take paid time if this tool disappeared tomorrow."

Paid time matters here because it sets a floor. If removing a $49/month tool would mean you spend an extra two hours a month on a task you'd otherwise delegate at $30/hour, the tool saves you $60 and costs $49. It earns its slot. If removing it changes nothing about how you spend money or time, the subscription is a habit, not an investment.

JPMorgan Chase Institute data shows newer cohorts of small businesses started AI spending at $20 to $30 per month, down from $50 per month for early adopters in 2019. That price drop is part of the problem. At $20/month, the financial pressure to evaluate fit disappears. The tool stays on the card because canceling feels like more work than the charge costs.

When daily use still fails the test

A reasonable objection: if a founder uses a tool every day, revealed behavior has already answered the task-alignment question. The tool is woven into something they do repeatedly, or they would have stopped.

The OECD data breaks this. Shallow use is the norm among frequent AI users, not the exception. A founder using an AI tool daily to do something a free-tier alternative handles equally well is using it constantly and losing money on it. Daily use proves the habit. It does not prove the habit is irreplaceable or that the cost is justified against an alternative.

Five questions that do close the question

Run these against each subscription:

• What specific task does this tool handle?
• How often does that task recur each month?
• What would you do instead if this tool vanished, and what would that cost in time or money?
• Does another tool in your stack do this already?
• When did you last use this for that specific task?

The third question is the one most audits skip. It forces you to price the alternative, not just log the usage. A tool used three times a month on a task you'd otherwise spend 90 minutes completing by hand is worth keeping. A tool used daily for something a free ChatGPT session handles is not.

Zylo's 2025 SaaS Management Index estimated license waste at $21 million per organization annually across its dataset. That figure comes from larger organizations, not founders. But the mechanism is identical at smaller scale: subscriptions accumulate faster than anyone evaluates them, and low monthly prices remove the friction that would otherwise force a decision.

The audit takes about 20 minutes per tool. The output is a list with three columns: task, alternative cost, last relevant use. Anything with no entry in the second column gets canceled first.