Beyond “Tokenmaxxing”: How CEOs Can Measure the Real ROI of Enterprise AI

There was a time when simply using AI felt like progress. In many organisations, that mindset is still around. Companies have rolled out AI copilots, added generative AI features to their products, and encouraged employees to experiment with tools such as ChatGPT. Alongside that adoption has come a new set of numbers that are easy to report: prompts sent, tokens consumed, users activated and AI licences deployed. 

These numbers tell us something about adoption, but very little about whether the investment is actually working. The term “tokenmaxxing” has emerged in technology circles to describe this obsession with AI activity, and while the term is tongue-in-cheek, the underlying issue is very real. For CEOs, founders and technology leaders, the question is no longer simply whether the organisation is using AI. It is whether those investments are making the business measurably better. 

We are seeing this shift firsthand at Cubet. Companies that were initially asking how they could start using AI are now asking a much more difficult question: how do we know whether our AI investments are delivering a return? As AI moves from experimentation into digital products, engineering teams and business operations, measuring that return is becoming just as important as implementing the technology itself. 

Start with the business outcome 

One of the easiest mistakes to make with an AI initiative is to start with the technology. A team discovers a powerful model and then looks for somewhere to use it, rather than starting with a business problem and determining whether AI is the right way to address it. 

For a digital product, the objective might be to improve conversion or retention, reduce customer support costs, accelerate software development, automate a repetitive process, or introduce a capability that was previously too expensive or complex to deliver. Models will continue to change rapidly, but the business problem you're trying to solve is likely to remain much the same. 

That is why the first question should be straightforward: what should be measurably better because we introduced AI? If that cannot be answered clearly, it becomes very difficult to establish a meaningful business case or measure AI ROI later. 

Measure outcomes, not activity 

Prompt volume and token consumption can indicate that an AI capability is being used, but they don't tell you whether it is creating business value. The starting point for measuring enterprise AI ROI should instead be a baseline. 

How long does the process take today? What does it cost? How many people are involved? What is the current conversion rate? How many customers can the support team handle? Once those numbers are understood, the impact of an AI implementation can be measured against them. 

For most digital businesses, the impact will eventually show up in a few familiar places: revenue, productivity and cost. 

AI can influence revenue through better personalisation, customer engagement, conversion, retention, upselling and new product capabilities. An AI feature used 100,000 times isn't necessarily valuable if those interactions don't contribute to a meaningful commercial outcome. 

Productivity is another important area, particularly for technology companies. AI can help engineering, product, sales, support and operations teams spend less time on repetitive work. However, simply reporting that developers are using AI every day isn't a productivity measure. Delivery cycle time, workflow completion time, resolution time and output per employee provide a much better indication of whether AI is actually changing how the organisation operates. 

Cost is often the easiest business case to quantify. If an AI-powered support workflow allows an existing team to handle substantially more customers without a proportional increase in headcount, the benefit is tangible. The same applies to data processing, quality assurance, documentation and other repetitive operational work. 

The real opportunity, however, is when an AI implementation starts improving more than one of these areas at the same time, for example, when it reduces the cost of serving customers while also increasing the capacity of the existing team. 

Expect an initial dip in returns 

Enterprise AI rarely delivers its full return immediately. There is usually a period of investment involving data preparation, system integration, workflow redesign, model evaluation, security and governance, as well as the time required for teams to adapt to new ways of working. 

This is the AI J-curve: an initial period where costs can rise, and productivity may temporarily fall before the benefits begin to compound. As the implementation matures, organisations learn where AI performs reliably, where human oversight remains necessary and which processes should actually be redesigned around the technology. 

Some initial friction is inevitable, but that shouldn't give a project a free pass indefinitely. There still needs to be evidence that the implementation is moving in the right direction. The right approach is to establish measurable milestones from the beginning. Early measures might include accuracy, workflow completion, adoption quality and the amount of human intervention required. As the system matures, those measures should connect to the metrics that matter to the business, such as cost per transaction, productivity, revenue and margin. 

Not every AI initiative will justify continued investment, and knowing when to redesign or stop is an important part of managing AI investment. 

The real value comes from integration 

There is a significant difference between adding AI to a product and integrating AI into the product itself. A standalone chatbot can answer questions, but an AI capability connected to customer context, proprietary data and application workflows can do much more. It can make recommendations, automate decisions, trigger workflows and take action within the product. 

This is where enterprise AI becomes particularly relevant to digital product companies. The most valuable implementations are often connected to the systems that already run the business, including product databases, customer information, internal knowledge, analytics, CRM, support systems and operational workflows. 

Technologies such as retrieval-augmented generation (RAG), AI agents and workflow automation can play an important role here, but they should serve the business requirement rather than become the objective themselves. There is little value in building a sophisticated AI architecture simply because the technology allows us to. What matters is whether the architecture supports the product, the users and the business case behind it. 

For a digital product, that distinction can be significant. An AI feature may improve the user experience, but AI embedded into a core workflow can change the economics of the product itself. 

Measure the economics at scale 

A proof of concept can look very promising in a controlled environment and still run into problems when it becomes part of a real product. Once usage grows, model costs, infrastructure requirements, latency, data processing and human intervention can all become significant factors. 

This is particularly important for digital businesses, where an AI capability designed for a few thousand users may behave very differently when it reaches hundreds of thousands or millions of users. AI ROI therefore needs to be considered not only at the point of implementation, but at the scale the business expects to operate. 

Cost per interaction or transaction, human intervention, accuracy, infrastructure requirements and performance at higher volumes are all worth tracking. The question isn't simply whether the technology works; it is whether it continues to make economic sense as the business grows. 

The CEO test 

Before approving an AI initiative, leadership should be able to explain: 

• what business problem it is solving,  

• what the baseline looks like today and  

• which metric should improve as a result. 

It is equally important to understand what the solution will cost as usage grows and what evidence will determine whether the initiative should be scaled, redesigned or stopped. 

There is also a broader question worth asking: 

Does this create an advantage that matters to our customers and our business? 

From AI adoption to AI advantage 

The enterprise AI conversation is changing. Most companies have already started experimenting with the technology. The more difficult task now is deciding which initiatives deserve further investment, which need to be redesigned and which should be stopped. 

For CEOs and founders of digital businesses, that is not simply a technology decision. It is a product, business and operating-model decision. The companies that create lasting value from AI will not necessarily be the ones consuming the most tokens or deploying the most models. They will be the ones that connect AI to outcomes the business genuinely cares about: revenue, productivity, cost, customer experience and competitive advantage. 

Cubet helps technology companies identify where AI can create meaningful business value and integrate those capabilities into their digital products, applications and workflows. Our focus is not simply on introducing AI, but on making sure it fits the product, the architecture and the business case behind it. 

Because the objective of enterprise AI isn't to use more AI. It is to build a better business with it.