You Are Paying for Nine AI Tools and Using Only Three

Nobody decided to buy nine AI subscriptions. Three people trialled something, nothing got cancelled, and two of them are on a personal credit card. Here is how to run the inventory, decide what goes, and keep experimentation alive without the bill growing every quarter.

November 7, 2026 · 7 min read · By José Cabal

Nobody sat down and decided to buy nine AI tools. That is what makes this problem hard to see and easy to fix.

What happened is ordinary. Someone in marketing trialled a writing tool. Someone in sales trialled a meeting recorder. Operations found something for summarising documents. Each one was thirty dollars a month, each one was approved in a Slack message or not approved at all, and none of them were ever cancelled. Eighteen months later there is a line in the accounts that nobody can explain and a general sense that the company spends a lot on AI.

How nine subscriptions happen without a decision

Three forces produce this, and all three are working as intended.

The price is below the threshold that triggers scrutiny. A twenty-dollar-a-month subscription does not need a business case, so it does not get one. Nine of them is over two thousand dollars a year, which would have needed a conversation if it had arrived as one invoice.

Trials convert to paid silently. The whole model depends on it. You enter a card for a fourteen-day trial, the person who set it up moves on to something else, and the charge appears every month afterward with no prompt asking whether you still use it.

And cancelling has a small emotional cost. It feels like admitting the experiment failed, and there is always a chance someone still uses it. So it stays.

None of this is anybody's fault. It is a predictable outcome of cheap software and no owner, and the same pattern shows up across the whole marketing tool stack, including outside AI.

Run the inventory, including the personal cards

Start with what you can see. Pull twelve months of statements for every company card and every account with a payment method attached, and list every recurring charge that is a software subscription. Do not filter for AI at this stage. You want the full list, because the question of what overlaps with what needs all of it.

Then find the invisible ones. Three sources, in this order.

Ask every team directly, and ask in a way that gets an honest answer: what tools do you use in a normal week, including anything you pay for yourself or use on a free tier. Say plainly that nobody is in trouble. If people think this is an audit with consequences, the shadow tools stay hidden and you have wasted the exercise.

Check expense claims for the last six months. Small software charges reimbursed through expenses are the most common form of invisible subscription at a company this size.

Check which outside services are connected to your main systems. Your email platform, your CRM, and your file storage all keep a list of connected applications, and that list often contains tools nobody remembers authorising.

For each tool record five things: what it costs, who pays, who actually uses it, what it is used for, and when it renews. That last field matters more than any other and is the one most people leave out.

The consolidation test

Skip the scoring matrix. One question sorts the list faster and better: if this tool disappeared tomorrow morning, what would break?

Three answers, and each has an obvious action.

A person could not do their job. Keep it. This is a real tool with a real user and the conversation is over.

Something would be slower or more annoying, and there is a workaround. This is the interesting middle. Ask what the workaround costs in hours per month and compare it to the subscription. Plenty of tools save two hours a month and cost the equivalent of four hours of someone's time, which means they lose that comparison.

Nothing would break. Cancel today. On a nine-tool list there are usually two or three of these, and cancelling them is free money.

Ask the question of the person who uses the tool, not the person who bought it. Those are frequently different people and only one of them knows the answer.

Watch for the tool with one enthusiastic user and no other adoption. Sometimes that person is genuinely getting value and you keep it. Sometimes it was a trial that became a habit for one person while the intended use case never materialised. The way to tell is to ask what would break, and to listen to whether the answer describes work or describes preference.

Where one assistant replaces four point tools

A lot of AI tool sprawl is a timing artefact. Companies bought specialised tools in a period when the general assistants could not do those jobs, and several of those jobs have since been absorbed.

Categories that have largely collapsed into a general assistant: drafting and rewriting text, summarising documents, generating ideas and outlines, basic translation, reformatting content between formats, and simple data analysis on a spreadsheet you paste in. If you pay separately for any of these, test whether your main assistant does it acceptably. Often it does, and four subscriptions become one.

Categories where a purpose-built tool still earns its place: anything that connects directly into your systems and writes data back, such as a meeting recorder that puts summaries into your CRM automatically. Anything with a specialised interface that matters, such as image or video editing. Anything where the tool is the system of record rather than a helper. And anything with compliance or audit requirements attached.

The test is whether the tool does work inside your systems, or whether it produces text you then copy somewhere. If you are copying and pasting, a general assistant probably covers it.

One caution. Consolidating onto a single assistant means a single point of dependency on one vendor's pricing and availability. At a small company that is an acceptable trade against paying for nine overlapping tools, and it is worth naming rather than discovering later.

Shadow tools are a security problem, not a budget problem

The subscriptions on personal cards cost you little. The exposure is elsewhere.

When an employee uses a personal AI account for work, your company information sits in an account you do not control and cannot access. If that person leaves, the account leaves with them, along with whatever they put in it: client lists, pricing, contract text, strategy documents. You have no way to retrieve it and no way to know what is there.

There is a second layer. Consumer tiers of these tools handle your content differently from business tiers, and the business and enterprise tiers generally commit in writing that your content is not used to train the underlying model. A free personal account often carries no such commitment. So client material entering a personal free account may be in a category your client contracts do not permit.

The fix is not a ban. Bans push usage further underground and you lose visibility entirely. Provide a company account for one or two approved tools, make it easy to get, and ask people to use it. Most shadow tool use happens because the official route was slow or did not exist. The wider question of what data can go where is covered in the AI risk and policy guide.

An approval process that does not kill experimentation

You want people trying new tools. A process that stops that costs more than the subscriptions do. Two rules are enough.

Trials are free to start, with one condition: tell someone. A shared document or a channel where anyone posts what they are trying, what they hope it does, and when the trial ends. No permission needed. The only obligation is visibility.

Converting a trial to a paid subscription requires one named person to approve. Not a committee. One person, usually whoever owns the software budget, who asks two questions: does this overlap with something we already pay for, and what data goes into it. Approval takes five minutes.

The split is deliberate. The friction sits at the point where money starts, and there is none at the point where learning happens. Most trials will not convert, and that is the process working.

Renewal dates are the control point

One habit makes all of this sustainable.

Put every subscription's renewal date in a shared calendar, with a reminder two weeks before, assigned to the person who uses the tool. The reminder asks one question: do we still need this? Monthly subscriptions get a quarterly check instead, since a monthly reminder becomes noise.

This works because it moves the decision to the moment when cancelling is easy. Cancelling mid-term feels like waste and rarely gets done. A reminder two weeks before an annual renewal arrives at the exact moment when the choice is real and the money is still yours.

Annual plans deserve care for the same reason. The discount is usually worth taking for a tool you have used for a year. For anything new, pay monthly until you know, because a year of an unused annual plan costs more than the discount ever saved.

The takeaway

Pull twelve months of statements, ask every team what they actually use including personal accounts, and record the renewal date for each tool. Then ask one question per tool: if this vanished tomorrow, what would break? Cancel everything where the answer is nothing.

Test whether your main assistant covers the drafting, summarising, and translating tools you bought separately eighteen months ago. Several of those categories have been absorbed and the subscriptions are still running.

Then keep trials free and put the approval at the point where money starts. My verdict: the money saved is real but modest. The larger gain is knowing where your company's information is, which is worth more than the two thousand dollars.

About the author

José Cabal, AI Marketing Consultant

AI Marketing Consultant · HubSpot Certified Trainer · Vancouver, BC

José Cabal has 20+ years of B2B marketing experience across Canada, the US, and Latin America, working with companies in tech, real estate, professional services, health, education, and manufacturing. He is a HubSpot Certified Trainer, holds an MBA from University Canada West, and founded Expresa3 in 2010 — one of the first inbound marketing agencies in Latin America. He has trained hundreds of marketers in English and Spanish, served on the HubSpot LATAM Partner Advisory Council, and spoken at conferences and universities across North America. He currently consults for SMBs building AI-powered marketing systems alongside his role as Head of Marketing and Business Systems for a growing company.

More about José Cabal · LinkedIn

Frequently asked questions

How do I find out which AI tools my company is paying for?

Pull twelve months of statements for every company card and list every recurring software charge, then check expense claims for the last six months and the connected-app lists inside your email platform, CRM, and file storage. Ask each team directly what they use in a normal week, including personal accounts, and say clearly that nobody is in trouble. Fear of consequences is what keeps shadow tools hidden.

What is the fastest way to decide which AI subscriptions to cancel?

Ask one question per tool: if this disappeared tomorrow morning, what would break? If a person could not do their job, keep it. If nothing would break, cancel it today, and on a nine-tool list there are usually two or three of those, often trials nobody remembered starting that renewed quietly for months on a card nobody checks.

Can one AI assistant replace several specialised tools?

Often yes, for drafting and rewriting text, summarising documents, generating outlines, basic translation, reformatting content, and simple spreadsheet analysis. Those categories have largely been absorbed. Purpose-built tools still earn their place when they connect into your systems and write data back, or when the interface itself matters.

Should I ban employees from using personal AI accounts for work?

No, because bans push usage underground and you lose the visibility you were trying to gain. Provide a company account for one or two approved tools and make it easy to get. Most shadow tool use happens because the official route was slow or did not exist.

Why does it matter if staff use their own AI accounts?

Company information ends up in an account you do not control and cannot access, and if that person leaves, the account and its contents leave with them. Consumer tiers also handle your content differently from business tiers, and free personal accounts often carry no commitment that your content stays out of model training.

How much does AI tool sprawl actually cost a small company?

The subscription total at a company running nine tools is typically a few thousand dollars a year, which is real but modest. The larger cost is not knowing where company and client information is stored. Treat the money as the reason to start and the visibility as the reason it was worth doing.

What approval process should we use for new AI tools?

Two rules. Trials are free to start provided the person posts what they are trying and when the trial ends in a shared place. Converting a trial to a paid subscription needs one named person to approve, asking whether it overlaps with something you already pay for and what data goes into it.

Should we buy annual or monthly AI subscriptions?

Monthly for anything new, until you know it gets used. Annual is worth the discount for a tool your team has relied on for a year, once actual daily use has proven it earns its place. A year of an unused annual plan costs more than the discount ever saved, and it locks you into a tool nobody remembers choosing.

How do I stop subscriptions from piling up again?

Put every renewal date in a shared calendar with a reminder two weeks before, assigned to the person who uses the tool, asking whether it is still needed. Monthly subscriptions get a quarterly check instead so the reminders do not become noise. The date matters because cancelling mid-term feels like waste and rarely happens.

What if only one person uses a tool but they love it?

Ask what would break if it went away, and listen to whether the answer describes work or describes preference. Sometimes that person is getting genuine value and you keep it, because a single strong user can justify the cost of a specialised tool. Sometimes a trial became a personal habit while the intended use case never arrived, and the honest answer is that nothing measurable would change.

Is consolidating onto one AI assistant risky?

It creates a single dependency on one vendor's pricing and availability, which is worth naming before you do it. At a company of thirty people that trade is usually acceptable against paying for nine overlapping tools. Keep your important outputs stored in your own systems rather than only inside the tool.