Why Your AI Marketing Budget Is Becoming Difficult To Explain
AI spending in marketing is rising rapidly, yet few companies know exactly what they are paying for. Four principles can help businesses identify hidden costs, reduce duplication and keep their budgets under control. A few years ago, the calculation was simple. A marketing department bought a software licence, agreed an annual price and assigned the expense to a budget. Generative AI has made that picture less clear. Companies now pay for user accounts, premium models, tokens, API calls, image credits, transcription, translation, storage and AI functions built into software they already use. The costs rarely appear in one place. Microsoft Copilot may sit in the IT budget, Adobe’s generative functions under design, a translation platform under communications and an AI-powered chatbot under digital services. An agency may use its own AI tools within an approved retainer, while individual employees work with additional applications bought through departmental budgets. Every contract may have been approved correctly. The problem starts when management asks how much the company spends on AI in marketing and what that money produces. The pressure is growing because AI has moved beyond occasional experiments. Marketing teams use it for research, drafting, localisation, image production, campaign variants, reporting and client service. Once companies connect these systems to automated workflows, usage can rise quickly. Tokens, the units used to process prompts and responses, have become a production cost that businesses need to manage alongside licences and labour. One of our clients, a large Swiss software company, told us that is AI expediture had at least doubled since the beginning of 2026.
1. Calculate the cost of the finished work
The price of an AI subscription says little about the cost of the work it supports. A tool may produce a first draft within seconds, but the company still needs to check facts, remove weak claims, adjust the tone, review the legal implications and approve the final version. Consider a Swiss professional-services firm preparing a market report for clients in German, French and English. The communications team uses an AI assistant for preliminary research and a first draft. A translation platform creates the language versions, design software produces visual alternatives and an agency edits the text and prepares the publication under an agreed monthly retainer. The direct software expense may remain modest. The larger cost lies in the work around it. A senior employee verifies the content, a subject specialist corrects technical passages, compliance reviews sensitive claims and the agency rewrites sections that sound generic or imprecise. The report may still reach the market faster than before, but the company needs to count the entire process from the first prompt to publication. This calculation can produce an uncomfortable result. AI may reduce the time spent drafting while increasing the time spent reviewing. A less expensive model may require more editing than a premium one. An automated translation may save an external fee, yet still demand a native-language review when wording affects reputation or legal accuracy. Companies should therefore compare complete workflows. How long did the task take before AI? Which external costs disappeared? How many senior hours did the new process require? Did the final work improve, or did the team simply produce more material?
2. Find the functions you already own
Most marketing departments do not lack AI tools. They own several systems that perform the same task. Software providers have added generative functions to office suites, design platforms, CRM systems, analytics tools and social media applications. One company may therefore pay for document summaries in Microsoft 365, writing support in its CRM, image generation in Adobe, translation inside another platform and campaign variants within its marketing automation software. Employees may still buy specialist tools because they do not know what the approved systems can do, find them difficult to use or prefer the output of another provider. The result is a technology stack built through a series of reasonable decisions that no longer makes sense as a whole. A useful audit starts with functions, not brand names. Which tools can draft copy? Which can translate it? Which can summarise documents, transcribe interviews, generate images or prepare campaign variations? Which systems may process confidential information, and which must remain limited to public material? The answers often show that several departments pay for the same capability. A design team may use one image generator, while social media managers use another and the agency works with a third. The company may have three writing assistants even though the enterprise office package already includes one. Cancelling duplicate subscriptions can reduce costs, but the larger gain comes from clearer rules. Employees know which platform to use, procurement can negotiate fewer contracts and IT can protect sensitive information more effectively. The company also gains a better basis for deciding whether a specialised tool deserves its place.
3. Measure use and value separately
AI providers can report logins, token consumption, generated images and API calls. These figures show activity, not value. A marketing team may generate hundreds of social media posts that never receive approval. Another department may use an expensive research model only a few times each month and save several days of specialist work. The first tool appears busy, while the second creates the greater financial benefit. Management therefore needs two sets of numbers. The first shows what the company consumes: licences, tokens, API calls and storage. The second shows what changed in the work: production time, agency expenditure, translation costs, campaign volume or the number of repetitive tasks removed from employees’ schedules. Nobody needs to calculate a return on every prompt. The company does need to know why it pays for a platform. A translation system should shorten localisation or reduce outside costs. A writing assistant should accelerate preparation without creating an equal amount of correction. An image tool should increase the number of usable concepts rather than merely generate more options. The same rule applies to agency work. An agency does not need to list every AI tool on its invoice when the client has approved the scope and fee. The client should nevertheless understand how the production model has changed. Does the agency deliver more variants within the same retainer? Has turnaround become faster? Does the fee now place greater weight on strategy, editing and quality control because routine production takes less time? AI does not automatically make professional work less valuable. Research, judgement, creative direction, compliance and accountability still require experienced people. The commercial conversation should focus on what the client receives rather than how many minutes the software took to generate a draft.
4. Put one person in charge of the full picture
AI expenditure usually crosses several departments. IT buys enterprise licences, procurement negotiates contracts, marketing chooses applications, finance monitors invoices and agencies use their own systems. Each team sees part of the cost. One person needs to see the whole picture. This does not have to be a new executive role or a central committee that approves every experiment. A marketing operations lead, procurement manager or digital transformation specialist can maintain the list of tools, contracts, owners and approved uses. That list should answer practical questions. What does each tool do? Which department pays for it? How many people use it? Does the price remain fixed, or does it rise with consumption? Which data may employees enter? Who checks the output? When does the contract renew? Usage reports can then expose unusual developments early. A sharp rise in token consumption may reflect a successful new workflow, or it may show that an automated system produces unnecessarily long outputs. A tool with few users may be redundant, or it may serve a small team working on high-value tasks. The figures need context before management decides whether to expand or cancel the service. Clear ownership also helps companies deal with shadow AI. Employees often turn to unofficial tools because the approved alternatives are slow, poorly introduced or unsuited to the task. A blanket ban may push the activity further out of sight. A small set of useful, clearly explained platforms usually works better.
AI costs need an operating logic
The purpose of cost control is not to make every prompt as short as possible or force employees to use the least expensive model. A stronger system may justify a higher price when it produces better results and reduces review time. An expensive specialist tool may earn its place if it replaces a larger external cost. The budget becomes defensible when management can explain it without reading out a list of software names. Which workflows use AI? What does each tool replace or improve? Where do costs rise with consumption? Which tasks still need substantial human review? Who decides whether the result is good enough to publish? Companies that can answer these questions do not necessarily spend less. They spend with greater control. Those that cannot may continue to approve every individual invoice correctly while losing sight of what their AI marketing budget has become.
