Disclosure, up front
We are frequently one of the five. When a client counts up what they are paying for, something we set up is often on the list, and more than once the honest recommendation has been to remove it because a tool they already had did the same job. This article contains the audit we run before quoting anything, which is also the audit most likely to end with us not selling you anything.
The typical small business using AI is now running about five AI tools — something for content, something for customer service, something for scheduling, something for analytics, something for workflow automation. Nobody sat down and designed that stack. It accumulated.
Five is not a scandal. Plenty of businesses genuinely need five things. The problem is what tends to be true about how those five arrived: bought at different times by different people to solve problems that overlapped more than anyone realised, each carrying an AI pricing premium, and each renewing quietly on a date nobody has written down.
The useful question is not "how many tools do we have". It is "how many times are we paying for the same capability", and that number is knowable in about ninety minutes.
Where the sprawl numbers come from
- Adoption jumped hard. Among companies with 10 to 100 employees, AI adoption went from about 47% to 68% in a single year, and more than half of small and mid-sized businesses now run at least one AI-powered automation.
- Most of it is shallow. Around 51% of small business owners describe themselves as testing AI without full commitment, and only about 8% reach what researchers class as advanced adoption.
- AI subscriptions waste faster than anything else. 2026 reporting puts AI tools as the fastest-growing software waste category, at roughly a 41% waste rate — the highest of any category tracked.
- The 2026 shift is duplication, not volume. The pattern is not more apps; it is more duplicates of the same workflow, each charging an AI premium.
Adoption figures are from 2026 small-business AI research. The waste-rate and duplication figures are third-party SaaS-spend reporting with varying methodology — directionally useful, not precise, and mostly gathered from companies larger than yours.
01How you got to five without deciding to
The sequence is remarkably consistent, and nothing in it is a mistake at the time it happens.
- Someone tries a writing assistant at $20 a month. It is good. It stays.
- The website vendor adds an AI chat widget. It is bundled, or nearly, so nobody scrutinises it.
- The scheduling platform ships AI features and moves you to a higher tier to get them.
- Somebody signs up for a transcription or meeting-notes tool on a personal card and expenses it.
- An automation platform gets added to connect two of the above.
Five tools, five decisions, each defensible. The thing that makes it expensive is that steps two and three quietly included capabilities you were already paying for in step one — and because they arrived as part of something else, nobody compared them.
This is not a small-business failing. In larger organisations the same dynamic runs at a scale that is genuinely startling: reporting on enterprise SaaS estates describes companies running hundreds of applications with roughly half of provisioned licences never opened. Small businesses do not have that problem in dollar terms. They have it in proportion, and with far less slack to absorb it.
02The overlap is the cost, not the count
Here is the exercise. Do not list your tools. List your capabilities, then write next to each one every tool you pay for that can do it.
| Capability | Tools that can do it | What usually turns out to be true |
|---|---|---|
| Draft written content | Standalone assistant, CRM's AI add-on, website builder, email platform | Three or four. One is used; the rest are paid for and idle. |
| Answer customer questions | Chat widget, helpdesk AI tier, phone answering service | Two, with different answers to the same question — a support problem, not just a cost one. |
| Transcribe and summarise calls | Meeting tool, phone system, CRM add-on | Two or three. The phone system's version is usually already included. |
| Move data between systems | Automation platform, native integrations, CRM workflows | Frequently paying for a connector to do what a built-in integration already does. |
| Report on what happened | Analytics tool, platform dashboards, spreadsheet | One is trusted. The others are open in a tab and quietly ignored. |
Illustrative, drawn from the pattern across client audits rather than from any single business or dataset. Your rows will differ. The point is the shape: capabilities down the side, tools across, and the duplicates become visible immediately in a way that a list of subscriptions never makes them.
Two rows on that grid tend to produce a reaction. The first is content drafting, where businesses routinely discover they are paying three times over. The second is customer answers, where the discovery is worse than financial: two systems giving customers different answers to the same question, because they were configured a year apart by different people and neither was updated.
03The AI premium, charged repeatedly
The specific thing that changed in 2025 and 2026 is that AI capability stopped being a product and became a surcharge. Your scheduling platform did not launch an AI product; it added AI features and moved you up a tier. Your CRM did the same. Your email platform did the same.
Individually each increase is small enough to approve without much thought. Collectively you are paying an AI premium four or five times for capabilities that substantially overlap, and in most cases using one of them.
This is also why sprawl builds faster in AI than it ever did in other software categories. Overlap that took years to develop elsewhere takes months here, because the overlap arrives through upgrades to things you already own rather than through new purchases anyone had to justify.
The question that finds the premium
For each subscription, ask: what did this cost twelve months ago, and what does it cost now? Where the price moved, ask what you got for the increase and whether anyone uses it. A meaningful share of small business AI spend in 2026 is tier upgrades taken to unlock features that duplicate a tool already in the stack.
04The ninety-minute audit
No software required. A spreadsheet and access to your card statement.
- List every recurring charge, from the statement rather than from memory. Twelve months of card and bank statements, including personal cards being expensed. This step alone routinely surfaces one or two subscriptions nobody remembered.
- Write the capability, not the product name. "Drafts email copy", not "Acme Pro". Products get renamed and rebranded; capabilities are comparable.
- Mark each one used weekly, used monthly, or not used. Be honest. "We should use it more" means not used.
- Name an owner for each. A person, not a department. Unowned subscriptions are the ones that renew forever.
- Write the renewal date. Most of the recoverable money is decided on dates nobody has in a calendar.
Reporting on first-time SaaS audits suggests teams that apply named ownership, proactive renewal review and subscription-level tracking recover meaningful annual sums — one 2026 figure puts it around $6,800 a year on average, though that is drawn from businesses larger than most reading this and should be treated as an indication of direction rather than a number to budget against. For a five-tool small business the recoverable amount is usually one or two subscriptions, which is real money and not life-changing money.
The larger return is not the cancellation. It is that after this exercise somebody knows what the stack is, which is the precondition for every other sensible decision about it.
05What to cancel, and the trap of cancelling too fast
The obvious move after an audit is to cut everything marked unused. That is right about three quarters of the time and produces a specific, avoidable mess the rest.
Before cancelling anything, check three things:
- Is anything automated pointing at it? A tool nobody opens may still be receiving data, or feeding a report someone does read. Cancelling it breaks something silently, and the breakage shows up weeks later as data that quietly stopped arriving.
- Where does the data go? Export before you cancel, not after. Access frequently ends the day the subscription does, and "we'll grab it later" is how a year of call transcripts stops existing.
- Is it the cheap one you are cutting? The instinct is to cancel the standalone $20 tool and keep the capability inside the $200 platform. Sometimes correct. Often the $20 tool is the one people actually use, and the platform's version is the one nobody could get on with.
Our rule: cancel one thing, wait a full billing cycle, then cancel the next. Cutting four subscriptions in one afternoon means that when something breaks, you will not know which cut did it.
06When five tools is the right answer
Consolidation is a fashion and it is not always right. Keep the sprawl if:
- Each tool is genuinely best at its own job and the overlap is small. Five specialised tools that each do one thing well, all used weekly, is a good stack. The count was never the problem.
- Consolidating means a platform migration. Moving to an all-in-one to save $60 a month, at the cost of weeks of disruption and retraining, is a bad trade that gets made regularly.
- The duplicate is a deliberate fallback. Sometimes two systems that can do the same thing is resilience rather than waste, particularly for anything customer-facing. Just make it a decision rather than an accident.
07The honest summary
Five AI tools is now normal, and normal is not the same as intentional. The cost is rarely the number of subscriptions; it is the same capability bought three times through tier upgrades nobody compared, and the customer-facing systems giving different answers because they were set up a year apart.
Ninety minutes with a card statement and a capability grid will tell you which of those you have. The money you get back is worth having. The knowledge of what you are actually running is worth considerably more, and almost nobody has it.
08Common questions
How many AI tools does a typical small business use?
About five — commonly something for content, customer service, scheduling, analytics and workflow automation. Adoption among companies with 10 to 100 employees rose from roughly 47% to 68% in a single year, and more than half of small and mid-sized businesses now run at least one AI-powered automation. Most of that adoption remains shallow: around 51% of owners describe themselves as testing rather than committed, and only about 8% reach advanced adoption.
Is running five AI tools a problem?
Not by itself. Five specialised tools that each do one job well and are all used weekly is a good stack, and the count was never the issue. The problem is what is usually true about how they arrived: bought at different times by different people, with capabilities that overlap more than anyone realised, each carrying its own AI pricing premium. The question worth answering is how many times you are paying for the same capability, not how many subscriptions you hold.
How do I audit our AI subscriptions?
Pull twelve months of card and bank statements rather than working from memory, including personal cards being expensed. For each charge write the capability rather than the product name, since products get rebranded but capabilities are comparable. Mark each as used weekly, monthly or not at all, name an individual owner, and record the renewal date. It takes about ninety minutes and needs nothing but a spreadsheet.
What is the AI pricing premium?
AI capability largely arrived as a surcharge on software you already owned rather than as new products. Your scheduling platform, CRM and email tool each added AI features and moved you to a higher tier to reach them. Individually each increase is small enough to approve without much thought; collectively you may be paying an AI premium four or five times for overlapping capability while using one of them. It is also why AI sprawl builds in months where other software categories took years.
What should I check before cancelling a subscription?
Three things. Whether any automation is still pointing at it, because a tool nobody opens may still be feeding a report someone reads and cancelling breaks that silently. Where the data goes, because access usually ends the day the subscription does and exporting later is often not an option. And whether you are cutting the cheap tool while keeping a more expensive platform's version that nobody could get on with. Cancel one thing per billing cycle so you can tell what broke.
How much money does a subscription audit recover?
2026 reporting on first-time SaaS audits puts average recovery around $6,800 a year for teams that adopt named ownership, proactive renewal review and subscription-level tracking, though that figure comes from businesses larger than most small companies and should be read as directional. For a five-tool small business the realistic result is usually one or two cancelled subscriptions. The larger return is that someone finally knows what the stack is.
What is the worst overlap to have?
Two systems answering customer questions. Businesses often find a chat widget and a helpdesk AI, or a chat widget and a phone answering service, configured a year apart by different people and never reconciled — so customers get different answers to the same question depending on which one they reach. That is a support and trust problem rather than merely a cost problem, and it is the row on the audit grid that tends to produce the strongest reaction.
When should we not consolidate our AI tools?
When each tool is genuinely best at its own job with little overlap and all are used weekly. When consolidating means a platform migration whose disruption and retraining cost exceeds the saving, which is a bad trade people make regularly to save small monthly amounts. And when a duplicate is a deliberate fallback for something customer-facing, where having two systems capable of the same job is resilience. Just make that a decision rather than an accident.
Send us your capability grid
Write down what you can do with AI today and every tool you pay for that can do it. We will tell you where you are paying more than once, what is safe to cut, and what would break if you cut it. If your stack is already clean, that is what we will tell you — and if the tool we should cut is one of ours, we will say that too.
Ask for a stack auditSources, read 7 September 2026: 2026 small-business AI adoption research for the five-tool median, the 47% to 68% adoption shift, the 51% "explorer" figure and the 8% advanced-adoption figure; 2026 SaaS spend and waste reporting for the 41% AI waste rate, the duplication finding and the $6,800 audit-recovery figure. The spend and waste figures are third-party, drawn from varying methodologies and from companies larger than most small businesses, and are presented as directional rather than precise. The capability grid and the ninety-minute audit are ours. Related: Half of Companies Bought AI and Changed Nothing Else and Your Software Is Growing Its Own AI Agents.
Hero image from Unsplash, used under the Unsplash License.