Why Is This Happening?

There are several reasons:

• One is that AI providers offered extremely cheap initial versions to entice users. Meanwhile, as agentic technology developed, AIs themselves started using more and more tokens (the unit of consumption). Even as token costs decreased, token usage increased to the point where the net costs ran higher.
• Meanwhile, companies aggressively promoted, encouraged, and often required AI use, in some cases tying AI use to promotions. These behaviors caused employee AI use to skyrocket, and not necessarily with a tie to Return on Investment (ROI). Negative practices, such as tokenmaxxing - where employees vie to use the most tokens, emerged.

Now, the bills are coming due, and awareness of the AI bill is increasing. For a CEO or CIO, here is a way to think about the issue and what to do.

First, The Good

The good news is that the bill does mean that your company is using AI, and probably aggressively. This is the first hurdle of AI transformation, ensuring that teams embrace the use of AI in their respective roles. This accomplishment is not to be taken lightly, given that resistance to AI use is very real, and the first step towards using AI effectively is to use it in some form and understand what it can and cannot do for you. This is itself a process of exploration.

Now, The Bad

However, just encouraging employees to use AI is far from enough. Just because an employee has embraced AI use does not mean they are doing it in a way that connects to ROI. Ask yourself these questions:

• Do they know how to best complement AI and their own skills, i.e., maximize their AI force multiplier?
• Have they been given access to the most effective tools for augmenting their workflows with AI? What tasks are they using AI for? Do these tasks link to ROI for the business?
• If their productivity has improved with AI, what are they doing with the rest of their time?

And Then, The Ugly

The rapid pace of AI growth and the noise around it make it hard to provide guidance for organizational use. Beyond inefficiencies, practices have emerged that cause additional challenges:

• The first example is tokenmaxxing. If you provide the impression that AI use is mandatory, but do not qualify that statement with a why and a what, you risk employees striving to follow the new rules by using as much AI as possible. This is tokenmaxxing, and it can drive up your bills without driving up your revenue. It gets even worse if you tie the use of AI to promotions or risks of layoffs.
• A second example is Shadow AIs. As AI tools proliferate and access becomes trivial, many AIs may be in use in your company at the same time, and not just those approved and officially issued by your IT organization. In addition to driving up bills, this AI version of “Shadow IT” can cause security and other business risks.

What To Do?

To address these issues, ask yourself the following questions:

• Do you have metrics for AI use by your teams? Tokens emerged as an early metric because they are easy. They are the direct measurement of AI consumption and are directly tracked to billing. However, tokens do not measure how the AIs were used in relation to your business objectives. Effective metrics need to be tied to ROI, and ideally reward employees for business outcomes as well as effective AI use that optimizes the AI force multiplier.
• What has been messaged about AI to your teams? If the message is that AI use is required and a lack of AI use will result in some kind of negative action, all forms of AI use will soar. Do your employees understand the metrics? Have they received training not just on AI use but effective AI use as it ties to their domain and role? Is their AI use and consequent ROI visible for them to measure and optimize?

Metrics and Non-Negotiables

In a recent article, I illustrated key metrics that can be used for AI-based coding. These can serve as a template for other domains. This article also noted that understanding your non-negotiables is key to success. Ultimately, AI use is not a goal; it is a means to an end. The end, which is usually better business outcomes, should be a non-negotiable.

Budgets and Forecasting

AI is changing in every way, and at tremendous speeds. Costs per unit (tokens) are decreasing, while new usages often generate higher token costs per API call, business transaction, and other business meaningful unit metrics. The art of getting, tracking, and forecasting AI is new to all companies and will likely take time for your organization to master. A good starting point is to recognize this and determine who in the organization owns this responsibility, which is likely to be cross-functional and multi-disciplinary. It would be wise to consider AI costs as a new budget line item on the same scale as other major business costs, such as facilities and travel, and manage it accordingly.

Along these lines, procurement of AI should be a corporate strategic item at the same level as procurement for the other top-level budget line items. Decide who is responsible for understanding what is available, negotiating with suppliers, and tracking the cost landscape as it churns with new providers and new offerings. Consider tiering AI services, with the most expensive being available for business-critical usages (much like one would do with last-minute travel), and business unit and team level allotments for general use.

Takeaways

• Using AI is a necessary and powerful first step. If you have that, do rejoice. Getting to this state is harder than it looks.
• However, it is not enough. If you do not upskill your teams on being critical-minded in their use, the likely first outcome will be just a big bill.
• Your business metrics have likely not changed. These are your non-negotiables. Keep them front and center at all times. AI is exciting. Using it to improve ROI is the only real goal.