The Hidden Costs of AI – Looking Beyond the Subscription Fee

AI platforms are often marketed like software. A monthly subscription, a per-user fee, sometimes even a free version to get started.

That pricing makes adoption appear relatively inexpensive. For a small trial, it often is.

But the picture has changed as AI is rapidly moving beyond experimentation and becoming part of everyday business. Once staff begin using it regularly, workflows are redesigned with it and business systems become connected to it.  As a result sometimes unknowingly the subscription fee often becomes only a small part of the total investment.

This is particularly important for small and medium-sized businesses. Larger organisations may be able to absorb implementation costs or dedicate specialist resources to managing new technology. SMEs typically have tighter budgets, leaner teams and less room for error.

Understanding the full cost of AI usage before making an investment can help businesses avoid surprises and make better decisions about where AI will genuinely create value.

Research into AI implementation, including work published by Harvard Business School, consistently identifies costs that sit outside the advertised licence fee and are often underestimated during the buying process. These costs can have a greater impact on the return on investment than the subscription itself. In a 2026 NBER survey of nearly 6,000 senior executives across the United States, United Kingdom, Germany and Australia, around nine in ten reported no productivity impact from AI over the previous three years despite increasing adoption.

The first is training. An AI platform is only as valuable as a team’s ability to use it effectively. Whether training occurs through formal courses, internal workshops or simple trial and error, it takes time. That time has a cost. For many SMEs, every hour spent learning a new platform is an hour not spent serving clients, generating revenue or completing operational work. A common mistake is assuming that because AI tools are easy to access, they are immediately easy to use. In reality, achieving meaningful productivity gains often requires staff to learn new skills, develop effective prompting techniques and understand how AI fits into existing workflows and at the same time do their normal job.

The second cost is integration. Very few AI platforms operate in isolation. Most businesses want AI connected to the systems they already use, whether that is accounting software, customer relationship management platforms, document management systems or project management tools. So its not just a tool, it’s a transformation of your business operations. The effort required to transform these systems is often overlooked when evaluating costs of an AI platform. Integration may involve technology consultants, software developers, internal staff time, process redesign or all four. The larger and more complex the workflow, the greater the likelihood that implementation costs will exceed the original estimate. Also the challenge remains if you invest now and update your products and processes will you need to redo the work (and the investment cost again) because its widely predicted that the pace of AI development will continue at its current rapid rate.

The third cost is downtime. Most new technology slows a business down before it speeds it up. During a transformation project as teams learn new processes and adjust the way they work, productivity often dips temporarily. This adjustment period is a real business cost even though it does not appear on an invoice. For businesses operating with small teams or tight margins, a few weeks of reduced productivity can have a bigger financial impact than the software subscription itself.

Another cost that is easy to miss is usage-based pricing.

Many business owners assume that once they have paid the monthly subscription fee, they can use the AI platform as much as they like. In some products that is true. Microsoft Copilot, ChatGPT Team and similar subscription services generally include a level of usage within the monthly licence. Other AI products, however, charge based on how much the system is used.

These usage charges are often measured in “tokens”. A token is simply a unit that AI providers use to measure how much information the system processes. Every time a user enters a prompt, uploads a document, asks the AI to analyse data or receives a response, tokens are consumed.

The technical definition is less important than the business impact.

The more information the AI needs to read and the more content it generates, the greater the token usage. A business using AI to draft a few emails each day may incur very little additional cost. A business using AI to analyse contracts, prepare reports, process customer data and automate workflows across multiple staff may consume substantially more.

Consider an accounting practice that uses AI to review client documents, extract information, prepare summaries and draft the first version of advice. Each individual task may cost only a few cents. However, when that process is repeated hundreds or thousands of times every month, the cumulative usage cost can become significant.

Before adopting any AI solution, businesses should understand whether usage is included within the subscription fee or charged separately. If usage-based pricing applies, it should be included in the financial analysis just like any other operating expense. OpenAI and Microsoft both publish pricing models where usage charges apply in addition to base platform costs for certain deployments and custom AI solutions.

The good news is that AI usage costs have fallen substantially in recent years but how long will this last? Businesses are now using AI every day across their firm’s operations. As a result, overall expenditure can still increase if usage grows faster than the cost per transaction falls. There has been recent commentary that is predicting that the prices of tokens will increase in the near future which will create budget issues particularly for SMEs who may have built systems by them completely reliant on AI.  We will have to watch this carefully to understand the medium term financial cost of AI throughout business operations.

None of this means AI adoption is not worthwhile. It simply means that the true cost extends well beyond the number displayed on the pricing page.

Because AI platforms are typically purchased through a software subscription, many businesses evaluate them the same way they evaluate software. The question often becomes whether the monthly fee is affordable.

That is not the most important question.

A better question is whether the investment will create measurable value.

If AI is going to be embedded into a core workflow, used across multiple staff and relied upon over several years, it should be assessed with the same discipline applied to any other business investment. The focus should be on outcomes. How much time will it save? What is that time worth? Will it improve service delivery? Will it reduce errors? Will it generate new revenue? How long will it take for the total investment to pay for itself? Hence it should be seen as a transformation and change to your business not simply a tool to do the same task you have done before.

Low-cost AI that saves little time is not necessarily a good investment simply because it is inexpensive. Equally, a more expensive solution that transforms a critical business process may generate a strong return.

Tax treatment and cash flow considerations

The way an AI investment is structured can influence both tax outcomes and cash flow.

A monthly software subscription is generally treated as an operating expense and is typically deductible in the year it is incurred. A larger upfront investment, such as custom AI development, major licence commitments or associated hardware, may instead be treated as a capital investment with deductions spread over time.

The distinction matters because the timing of the cash outflow and the tax benefit may not be the same.

Businesses considering a significant AI investment should evaluate both the accounting treatment and the cash flow impact before making a commitment. Given how quickly AI products, pricing models and incentive programs are evolving, it is wise to seek advice before assuming a new investment will be treated the same way as previous technology purchases.

A simple framework before Investing in AI

Before investing in any AI solution, ask four questions:

1. What specific business problem or recurring task will it improve?
2. What is the total cost, including training, integration, usage charges, monitoring and temporary productivity loss?
3. How long will it take for those costs to be recovered through time savings, efficiency gains or additional revenue?
4. What are the short and medium term costs of token usage to the business? Who controls this budget spend?
5. If the solution fails to deliver the expected benefits, how easily can the investment be unwound?

The answers will differ from one business to another.

An AI platform or agent that makes perfect sense for a business with fifty employees may not stack up for a business with five. The technology may be identical, but the commercial outcome can be very different.

The objective is not to adopt AI because it is available. The objective is to invest in AI where it creates measurable business value and creates transformational change.

If you are considering an AI investment and want to understand the likely costs, cash flow implications and potential return before you commit, speak with your Bentleys advisor.

The businesses achieving the greatest value from AI are not necessarily the businesses spending the most. They are the businesses asking the right questions as they invest.

Talk to us today

If you are weighing up an AI investment and want a clear view of the full cost, the cash flow impact, and the likely tax treatment before you commit, speak to your Bentleys advisor. Talk to us today.