主なポイント
- Enterprise AI spending keeps climbing even though most pilots show no return — MIT’s Project NANDA found 95% of organizations see no measurable ROI.
- Not every problem needs AI. Deterministic, if-then-else automation is often faster and cheaper than probabilistic agents.
- The Boomi Enterprise Platform, including Boomi Agentstudio, gives enterprises the visibility and governance to control AI costs and stop budget overruns.
Two things are happening with AI in parallel. Enterprises are embracing agentic processes with a desperate urgency. And they’re burning through budgets at a relentless pace.
The AI rush isn’t slowing down, and neither is runaway spending. One global survey of senior leaders found that businesses are doubling their investments year over year, even if current initiatives fail to deliver returns. Yet we consistently hear from business leaders that it’s far too easy to spend too much money, too quickly.
The problem is growing. Uber exhausted its entire 2026 AI allocation in just four months. Also, 79% of S&P 500 companies mentioned AI in their most recent earnings calls, but only 8% disclosed any AI revenue. Snowflake CEO Sridhar Ramaswamy was recently quoted in a The Information article as saying, “Are we worried about how much we are spending on AI inference across our internal teams? Absolutely.” The drumbeat of discouraging outcomes offers further confirmation of the 2025 MIT NANDA report, which found that 95% of organizations had achieved no ROI, sending a chill through C-suites everywhere.
The stakes go beyond busted budgets. The risks to brand reputation and noncompliance with data sovereignty regulations expand without AI guardrails. In some cases, such as within regulated industries, the toll can even entail legal consequences.
The real price isn’t counted in dollars or tokens, though. It’s measured in trust.
The Deterministic-Probabilistic Decision
The reasons why businesses lose control of their AI budgets can vary. For example, as in Uber’s case, overspending can be traced to a surge in the use of AI coding tools. For others, it’s the day-to-day use of tokens that comes with running agents. The bottom line about the bottom line? AI isn’t cheap.
At the same time, Economics 101 still applies. The question isn’t about AI’s capability. It’s whether the payoff justifies the investment. Making AI useful and practical requires a course correction. It requires you to reconsider everything you’ve ever known about IT total cost of ownership. You need to factor in more than the cost of AI itself. Capturing value requires a transparent, well-governed framework that lets you know exactly what you’re spending on AI, and what you’re getting for it.
Something else is equally important. You must decide when you even need AI models and agents to solve specific problems.
Treating AI as one-size-fits-all is a mistake. Ed Macosky, Boomi’s chief product and technology officer, likes to say that while AI is an excellent hammer, not every problem is a nail. Our CEO, Steve Lucas, shared a perfect example in his recent Boomi World keynote – payroll. If it’s Friday, then you pay your workforce. That’s a straightforward task that doesn’t require a probabilistic engine or costly tokens.
The new imperative is knowing when to run a deterministic, if-then-else workflow that’s the foundation of traditional automation, and when to spend those precious tokens on probabilistic processes designed for optimization.
Most people think they need AI because – let’s be honest – it’s really cool. But what they’re actually looking for is a deterministic outcome, only faster. That’s standard automation, which operates at a fraction of the cost of AI. So, if it’s not broken, don’t fix it. However, when those if-then-else statements become too brittle or unwieldy to scale, that’s when you should turn the keys over to agentic processes.
Think about it this way. We now have three types of computational logic: generative AI, discriminative AI, and if-then-else rules that developers have been writing for decades. The first two are forms of AI. The third is traditional programming. The difference is predictability. Ask a deterministic system the same question twice, and you’ll get the same answer. Ask generative AI the same question twice, and you might get different answers. That’s not a bug. It’s the nature of probabilistic thinking. One type of logic follows rules, and the other reasons through possibilities.
The trick is figuring out which is more appropriate and cost-effective.
Boomi Helps You Spend Intelligently
The Boomi Enterprise Platform connects things. Connecting things is what we’ve done for over 20 years. Applications, APIs, and data sources. Anything to everything, everywhere — from mainframes to MCP. That allows businesses to put data in motion to the right places and right people with quality and accuracy. Now, expertise in integrating systems and automating workflows also serves as the infrastructure for orchestrating the activities of AI models and agents within operations. We provide trusted, AI-ready data to the agents that need it.
As a result, our unified platform helps enterprises leverage AI when it makes sense and save money when it doesn’t. Boomi acts as the referee, offering both structural and financial benefits. The platform manages workflows that are either probabilistic, deterministic, or a combination of both.
For AI, Boomi provides a unified layer of orchestration and governance, keeping agents on task and giving full visibility into what they’re doing. It sits between your expensive models and your systems of record to enforce efficiency, which then mitigates unexpected budget meltdowns.
We’ve built the platform with functionality and cost containment in mind.
- Embedding AI at specific decision points within your workflows, and not across the board, prevents unnecessary token usage from accumulating. Boomi enables the creation of agentic workflows, so that AI runs only at the exact moment you need that “brain” for contextual reasoning, not constantly in the background, needlessly eating up tokens.
- Bounding agents within deterministic workflows stops them from gathering more context than needed before acting, looping, reevaluating, drifting, and over-generating outputs. Controlling these behaviors, all of which are unnecessary token burners, is not only possible but also straightforward on our platform.
- Running agents across your business that have been deployed from disparate systems – such as Agentforce, Glean, and Copilot — can become an accounting nightmare. The solution is to unify fragmented agent management in one place so you can monitor all your agents, regardless of where they were built. You can observe activity, validate access, prevent rogue agents from running amok, and track token consumption to stay ahead of escalating costs. That centralized governance is available in Boomi Agentstudio.
- Applying specific guardrails to agents narrows the scope of their work. You need the ability that Boomi offers to create restrictions that prevent agents from overreaching, keep them strictly focused on their tasks, and tightly control token consumption.
You can’t manage what you can’t measure. That doesn’t change with AI. If every agent, tool, and workflow isn’t auditable, observable, and controllable, you won’t have fiscal discipline. AI will become a budget black hole.
Balancing Cost With Value
It really shouldn’t be a surprise that most organizations are not yet reaping the benefits of AI. Productivity doesn’t just magically increase when you spot-weld models and agents into your processes. They require supporting infrastructure to provide the contextual data that will create magic.
AI has our attention. But ROI still pays the bills. Both the wondrous possibilities and the harsh downsides of AI are still revealing themselves. As enterprises struggle to align escalating expense concerns with business value, one thing is obvious. Cost oversight has rocketed to the top of AI priorities for every business.
It starts with asking this question when facing any challenge: “To use AI or not to use AI?”
Whatever you choose, Boomi connectivity provides the framework you need to solve the problem. And when the answer is that you do need a probabilistic system, Boomi will prevent it from becoming a crushing financial liability.
Spend on AI. But more importantly, spend wisely.