It’s the 17th of the month and one of our engineers wants more credits for Claude. They’ve already spent $4,000 and we can’t believe it. Barely halfway through July and they’ve not only overshot their monthly budget but need an even larger allocation. Soon after, alerts start going out that we’ve reached 75% of our allocated monthly spend limit.
This request, among many others, made us think twice about AI spend. The interest is there and often the innovation is, too. But what about the value? If anyone asked us to approve an IT product that costs a few thousand dollars per seat, per month, would we say yes? Especially when the cost fluctuates and the return on investment isn’t guaranteed?
At Paessler, like many companies, we’re building the AI boat as we sail it. The trouble is that spend leaks are common in the wild waters of automation. We realized the need to correct course: value-vs-spend checks serve as a necessary compass and a cross-functional AI team keeps the ship structurally sound, preventing things like shadow AI or unchecked agents from sinking the budget.
We’ve learned the hard way that protecting AI value means being willing to sometimes say no, putting the processes and the people in place to make that call before the bill arrives. Heading into the new year, let’s explore why “tokenmaxxing” is out and “valuemaxxing” is in.
AI got a longer leash than any other software
Earlier this year, our first AI-focused dev day showed us just how quickly expenses can get out of hand. We offered a $5,000 prize for the best feature we could build into our network monitoring platform. The team loved it, resulting in a record 29 submissions, with some deliberately choosing more expensive approaches to build things that would otherwise have been impossible, ranging from context-aware assistants to root cause analysis helpers. We encouraged that experimentation and gave a very long leash on cost, precisely because we’d never done it before and wanted to see the outcome.
Despite some impressive applications, larger-than-expected invoices followed. More prompts meant more tokens and bigger bills, and agents doing multi-step work burned through credits even faster. Additionally, newer models became available, and the use of high effort levels began to send our bills to the stratosphere.
This is a familiar story. McKinsey reports that more than 9 out of 10 companies (93%) are exceeding their AI budgets. We’ve seen this anecdotally with another leader in our network, a CIO, who shared that their annual AI budget evaporated within two months. The one-two punch of AI being consumption-based and curiosity-driven is a dangerous combination for enterprise spend, especially as major providers are starting to meter heavy, agent-style usage separately. By 2028, Gartner predicts AI coding costs will overtake the average developer’s salary. In other words, the tools could cost more than the developer using them.
A spending cap is a decision, not a ban
Of course, we still want everyone to chase their wildest creative dreams with this technology, but it needs to happen in line with the cost reality. This means creating hard-and-fast rules like group limits. This needs to happen first because you can’t retrofit discipline onto a tool people already use freely. So, set the ceiling at rollout and make sure token limits align with spending limits.
We suggest finding the budget sweet spot by working backward. Figure out how much you’re comfortable spending per person or team and calculate from there. For some reason, many overlook this due diligence step in the race to automate. Other consumption-based tools never get as much leeway, so why AI? Treat it like any other resource by setting a maximum spend limit and raising it when the value shows up.
Done right, caps don’t get in the way of innovation. Rather, they’re a sense check between employee and employer, surfacing genuine use cases and ensuring alignment between spend and outcome.
For example, say an executive hits their monthly cap while performing heavy analysis before a board meeting. We don’t want to impede that critical work, especially when it’s time-sensitive, but we also want to keep an eye on who’s doing (and spending) what. A warning can evolve into a quick check and an increase. On the other hand, users may be experimenting with AI that isn’t yet tied to a critical business need. Again, a cap enables discussions and decisions, and in this case, a manager decides to pause spending. In both instances, the cap does its job and turns invisible outlay into a deliberate decision.
Saying no takes the right people in the room
It’s worth stressing that employees aren’t doing anything wrong. They’re understandably excited and legitimately experimenting, usually with positive results. But too much of a good thing can hit the bottom line. This is why the right culture is just as important as spend controls. If employees know to be cost-aware, they’ll be cost-aware, innovating within set limits and expectations.
The same goes for shadow AI. Ideally, employees shouldn’t onboard their own tools or duplicate processes across platforms, as it both wastes resources and creates an attack surface. But in reality, two-thirds of office professionals at large companies say they’ve used AI tools at work they believed weren’t permitted. The waste is one thing, paying for capability you don’t need or already own, but unmanaged exposure is another. Consolidation does double duty: cutting the spend and reducing the surface.
At Paessler, we try to catch such problems early by uniting around an AI Center of Excellence (CoE). This cross-functional group brings together decision makers from IT, information security, human resources, finance, legal and operational excellence, meeting weekly to safely and strategically scale AI adoption.
Before approving a new tool, we ask whether something we already pay for can do the job through an integration. New AI features are increasingly rolling out in software we use, so part of the work is simply keeping track of what’s already in our ecosystem. Once we know a feature exists, our pen tester also tries to break new AI apps by accessing data he shouldn’t be able to see.
Communicating more cost-effective ways of using AI can go a long way to prevent costly overruns and help users feel more in control. For example, Claude lets users select the amount of “effort” needed to answer a question or perform a task. More thorough responses require higher token usage and therefore higher cost. By educating employees, we can help them become more aware of model selection and adjust it based on their actual needs, not just a default setting.
The CoE is now our single voice on publishing guidance, approved use cases, education and training, so employees know where we’re heading and where to bring their questions. This also matters for compliance. Having an AI tool repository backed by one set of usage standards means we always know what’s running and why.
And sometimes, after all those checks, the answer is no. For example, one app jumped in costs unexpectedly and, when we realized our main AI platform could handle it anyway, we consolidated. We know what each tool is doing and stay clear-eyed about what it achieves relative to what it costs. And the process is ongoing: evaluating performance, aligning spend with business goals and when needed, declining it.
This last part is crucial. As Jay wrote in September, many of us have gotten swept up in the hype and forgotten that AI is a tool that requires a business case. Again, this isn’t about shutting down new ways of doing things, but giving visibility into where efforts are going, backed by a process that keeps costs from snowballing.
Tokens for the sake of tokens just isn’t sustainable. In the year ahead, “no” should be the most valuable word in your AI vocabulary, not because AI isn’t valuable, but because maximizing its value through meaningful constraint is.