Shadow AI is a spending problem, not just a security one
A report from Marketing Signals positions shadow AI as a financial governance failure at least as significant as its security risks. The firm's Subscription Creep Report, published this month, finds that individually expensed AI tools, including copywriting assistants, image generators and transcription services, have grown 108% year-on-year in spend across UK businesses, largely without appearing on any formal software budget.
The scale of unapproved tool use is striking. Marketing Signals cites research placing the share of employees using AI tools their employer has not sanctioned at between 50% and 71%. Separate figures in the report suggest 98% of executives admit to having bypassed IT procurement to buy technology directly. ChatGPT is identified as the single most expensive individual software purchase in the underlying data, ahead of any traditional SaaS product.
The cost no one is tracking
The most concrete figure in the report concerns marketing departments specifically. Marketing Signals estimates that forgotten AI subscriptions are costing UK marketing functions approximately £63.7 million a year in aggregate, with a wide methodological range of £34.0 million to £112.1 million. The firm has also published a free calculator allowing businesses to estimate their own exposure using the same national model.
Harry Nisbet, General Manager at Marketing Signals, said: "A $20-a-month tool feels invisible on a personal expense claim, and that's exactly why it never gets reviewed or cancelled. Multiply that by however many people are doing the same thing across a business, and it adds up faster than most finance teams realise."
Nisbet noted that an internal audit at Marketing Signals itself identified two redundant tools and replaced three others with lower-cost alternatives, cutting the firm's outlay on those five products from £1,875 to £713 a month, a saving of roughly £13,200 annually.
The wider SaaS waste problem is also deteriorating. Marketing Signals cites data from procurement analytics vendor Vertice showing that combined SaaS waste has climbed from 62% to 65% of all licences over the past year, even as AI-native software spend accelerates.
Market context and governance read-across
The shadow AI spending story sits within a broader enterprise tension. Procurement and finance teams that built licence-management workflows around negotiated enterprise agreements are poorly equipped to track individually expensed micro-subscriptions. Most software asset management tools are designed to audit installed or cloud-provisioned software, not personal card charges that never enter a purchase-order workflow.
A number of SaaS management platforms, including Torii, Zylo and Productiv, have moved to address this gap by ingesting expense-report data alongside SSO and network signals. Whether these platforms can keep pace with the pace of AI tool adoption is an open question. The density of new AI point solutions, many priced below the threshold that triggers a formal procurement review, makes the category structurally harder to govern than traditional software.
On the regulatory side, the EU AI Act's prohibited and high-risk provisions are largely tool-agnostic: an employee using an unapproved AI tool to process customer data may inadvertently create a compliance exposure their employer cannot audit because the tool was never disclosed. The UK's ICO has signalled similar interest in how organisations document AI use in the context of data-protection-by-design obligations. Finance and IT leaders will need governance frameworks that extend to expense-level tooling, not just enterprise-licensed systems.
Marketing Signals says the report and calculator are available on its website. Harry Nisbet is available for further commentary.