AI Tool Sprawl: Paying for Overlapping AI Subscriptions Without Measuring Value

The tolerated pain

AI tools arrived in most businesses from the bottom up. The marketing coordinator signed up for a writing assistant. A developer put a coding assistant on a company card. The managing director bought a premium chatbot plan, then the business added an AI add-on to its office suite, and the CRM vendor switched on an AI tier at renewal. Each decision was small, sensible and individually approved, or at least individually expensed.

A year later, the business is paying for five or six AI products, several of which do broadly similar things. Some licences are used daily. Some were used enthusiastically for a fortnight. Nobody can say which tools are producing value, because nobody defined what value was supposed to look like.

This article is about the spending and value side of that problem. The data-protection risk of staff using unapproved AI tools is a separate issue, covered in the financial risk of uncontrolled AI use with company data.

Why AI sprawl is tolerated

• Prices are per user and look small. A monthly fee per person rarely triggers procurement scrutiny, even when it is multiplied across many people and several tools.
• Fear of missing out. Leaders worry that restricting AI tools will leave the business behind, so they approve everything.
• Spend is scattered. Subscriptions sit on personal cards, expense claims, vendor add-ons and software bundles, so no single report shows the total.
• Usage looks like value. Staff say they "use it all the time", which is not the same as saving time, improving quality or winning work.
• Features overlap quietly. General-purpose assistants increasingly do what specialist tools were bought for, but the specialist subscriptions keep renewing.

Operational symptoms

• different teams use different AI tools for the same task, with different results and no shared prompts or standards;
• outputs are copied between tools by hand, because none of them connects to business systems;
• finance discovers AI subscriptions only when reviewing card statements;
• licences remain assigned to people who have changed roles or left;
• nobody can answer "what has this tool changed for us?" with anything more specific than an anecdote.

How the pain becomes money

• Technology waste: paying for licences that are inactive or that duplicate capability already owned.
• Currency exposure: many AI tools are billed in US dollars, so the rand cost moves with the exchange rate even when the plan does not change.
• Fragmented work: staff spend time moving content between tools and reconciling inconsistent outputs.
• Opportunity cost: budget spread thinly across many tools could fund one well-integrated use case that changes a real process.
• Weak negotiating position: without usage data, the business cannot right-size seats at renewal.

Usage data is available, but rarely read

Some vendors provide the evidence needed to start. Microsoft's Copilot usage report in the Microsoft 365 admin centre shows enabled users (people holding a Copilot licence), active users (licensed people who used a user-initiated Copilot feature) and the active users rate, which divides one by the other. It can be viewed over the last 7, 28, 90 or 180 days, lists each licensed user's last activity date by app and can be exported to a spreadsheet. Microsoft counts a user as active only when they take an intentional action, such as submitting a prompt, not merely opening the Copilot pane.

That is a useful starting point, and comparable admin reporting exists in some other platforms. But it measures adoption, not value. A person can submit many prompts and save no meaningful time. Value has to be defined by the business, in terms of the work the tool is supposed to change.

For a structured way to think about this, the US National Institute of Standards and Technology's AI Risk Management Framework, released in January 2023 for voluntary use, organises AI management around four functions: Govern, Map, Measure and Manage. A small business does not need to adopt the framework formally to borrow its logic: decide who is accountable, map where AI is used, measure whether it works, and manage the portfolio accordingly.

An illustrative cost model

This example is illustrative; all figures are assumptions. Picture a Johannesburg marketing and events agency with 25 staff.

In this example, the agency pays for 58 seats and sees activity on 29. If inactive seats were removed and the writing tool cancelled because the general assistant covers the same need, monthly spend would fall from R27,950 to about R13,550 on these assumptions, roughly half. That is before asking the more important question: which of the remaining tools has measurably improved turnaround, quality or capacity? Replace every number with figures from your own invoices and admin reports.

When it becomes commercially urgent

1. AI spend appears on more than two or three cost lines or cards.
2. A major AI subscription or bundle is approaching renewal.
3. The active rate on any paid AI tool is well below the seat count.
4. Leaders are asked to approve a larger AI investment without evidence from current tools.

What better looks like

The business keeps one list of AI tools, with owner, cost, seats, billing currency and renewal date. Each tool has a stated purpose and one or two measures of value tied to real work, such as proposal turnaround time or hours spent on meeting minutes. Seats are reviewed quarterly against usage. New requests are checked against tools already owned. New AI tools start as time-limited pilots with a named owner, a small number of seats and an agreed review date. At the review, the pilot either shows evidence against its measure and is expanded, or it is cancelled before it quietly becomes a permanent subscription. AI costs are attributed to the outcomes they support, a theme explored in why AI usage costs are not allocated to business outcomes.

Practical next actions

1. Find every AI charge across cards, expense claims, vendor invoices and software bundles for the last 12 months.
2. Pull usage reports for each tool that offers them, using at least a 90-day window.
3. Group tools by job: writing, meetings, images, coding, analysis. Overlaps become obvious.
4. Define one value measure per tool and ask its owner to report against it.
5. Remove inactive seats and cancel duplicates before the next renewal.
6. Redirect savings into one integrated use case that connects AI to a real process.

The same discipline applies to software generally, as described in the financial waste of unused software licences. Where AI should be doing structured work inside business processes, Mellorca's AI workflow service focuses on connecting it to systems and outcomes rather than adding another subscription.

Bottom line

AI tools can be valuable, but a collection of subscriptions is not an AI strategy. When nobody measures value, the business pays for enthusiasm, overlap and inactivity. A short inventory, a usage review and one clear measure per tool usually reveal what to keep, what to cut and where the real gains are.

Sources and further reading

• Microsoft Copilot Usage Report - Microsoft 365 admin (last verified 10 October 2026)
• AI Risk Management Framework (last verified 10 October 2026)

Related Mellorca reading

• How Many Software Subscriptions Do You Pay For? The Cost of Not Knowing
• Paying for Several Applications That Do the Same Job
• Paying Suppliers Without Measuring Performance