Stop counting tokens: how to generate a real return on investment in AI

Where is the big money in AI (Image: Dreamstime)

By Roi Dimanick

Most organizations have already implemented AI, but most of them do not know if it is actually generating money for them. The discussion around AI is shifting: if until recently we asked "what value is the technology capable of generating", now the central question has sharpened: how can we generate ROI from it, and how do we measure it? A 2025 McKinsey survey reveals the root of the problem: 88% of organizations use AI in at least one business function, but only 39% report a material impact on the bottom line. Less than a fifth have defined dedicated KPIs for AI solutions and tracked them.

Many organizations measure model usage—number of licenses, activation frequency, and monthly token consumption—instead of measuring business results: whether problem handling time shortened, output increased, error rates decreased, and whether freed-up time is utilized for additional value. It is vital to clarify: an improvement in productivity is not necessarily ROI. Only when the organization changes its way of working and utilizes freed-up resources for business value can one talk about return on investment.

Where does AI implementation generate the most business value?

From our experience, three areas consistently prove ROI: service and support, back-office automation, and organizational knowledge management. Their common denominator is a large volume of repetitive actions where it is easy to measure time, cost, quality, and output. This differs from situations where AI remains a "smart chat" not connected to a defined process.

It is important to distinguish between an employee finishing a task faster, a department increasing output, and an organization reducing costs or increasing revenues. The transition between these levels does not happen by itself.

Service and support: output vs. costs

In customer service, AI is capable of understanding inquiries and formulating answers. A study in the Quarterly Journal of Economics found that an AI assistant increased the number of inquiries handled per hour by 15% on average, with improvements reaching 30% for less experienced employees.

However, a 15% improvement in output does not automatically equal a 15% cost saving. If the freed-up capacity is not utilized to handle more customers or stop new recruitment, the organization has a more efficient tool, but not necessarily ROI. One must measure resolution rates, waiting times, and cost per inquiry.

Back-office: where the big money is

Back-office projects often yield the fastest ROI. Processing forms, checking invoices, bank reconciliations, and updating systems are processes with clear manual costs and measurable error rates. In one project, we integrated OCR, AI, and an ERP system, resulting in 30%–40% savings in ongoing work. Business value is created when AI is combined with automation and control mechanisms, keeping a Human in the Loop for final decisions.

Organizational knowledge management

Searching for procedures, contracts, and regulatory information can shorten work time by 20%–30%. A good system must present sources, respect authorization systems, and rely on the most updated document versions. Organizations cannot settle for just adding a tool; they must redesign work around it.

---

When to wait with AI investment

I would be wary of implementing an organizational chat before the organization has dealt with its data. If information is distributed across servers, outdated, or unclassified, AI might amplify existing messes and endanger sensitive information. Wherever there is financial or regulatory significance, I would leave a Human in the Loop.

Where to start?

Most AI project failures stem from management choosing the wrong problem or failing to change work processes. Every CEO must answer 4 questions before investing in AI:

1. What is the business problem?

2. How will we measure success?

3. What will we do with the time we save?

4. Who is responsible for ensuring the change happens?

If you have a limited budget, look for the project with the best ratio of business value to complexity, evaluating potential financial value, activity volume, data quality, and implementation complexity. The true AI revolution will not be measured by tokens consumed, but by the business processes that have changed.