Earlier this summer, Uber made headlines when executives said employees burned through its multi-million-dollar 2026 AI budget by the end of April.
While Uber’s situation might be extreme, it also illustrates tensions facing all types of businesses.
AI tools can build more efficient businesses, but it’s hard to know what they can do unless you try them out. That experimentation costs time and money, so how much is too much to spend to implement these new tools?
Gerald Kane, head of the UGA Department of Management Information Systems, has studied how businesses incorporate new technologies into their cultures and long-term strategies. While AI is different, managers can take lessons from past innovations to make adoption and adaptation less stressful.
1. Is what happened at Uber the norm? What are you hearing from business leaders?
I wouldn’t say everybody’s dealing with this, but I would say Uber could be the canary in the coal mine.
AI providers have changed their enterprise pricing models over the last six months. The pricing had been per seat, and now it’s a per-usage model. People are realizing it’s far more expensive, particularly as people experiment with agents. That’s where the expense really adds up.
The price of usage is driven by supply and demand. As more people experiment, demand for these tools rises, and there isn’t enough computing capacity to meet it, so AI companies must charge for it.
I think everybody’s dealing with the question of “Are we getting the return that we expected?” And “How much do we want to pay for these tools?”
If they haven’t faced the topic in the last six months, it’s a decision every executive will face in the next six months.
2. In the past, it felt like employers would roll out new tools paired with specific uses. With AI, it feels like employees are being asked to find the best ways to use the programs. Are there benefits to this rollout method?
Yes, because the tool is fundamentally different. It is trained on data rather than built from code, so it’s immensely flexible and powerful.
When you build something from code, you tell it exactly what you want it to do, and it does that, and doesn’t do anything more than that. AI will do more, but it can be more unpredictable, and you don’t always know why it does what it does.
Even the developers don’t know how to use it correctly inside organizations. We’re all figuring this out together.
There is a lot of productivity data showing two types of effects. One, AI gets everybody up to an average level. So, if you know nothing about a topic, you can at least get to an average level very quickly. And two, people who have subject matter expertise and learn how to apply AI to those skills are supercharged. So, my argument would be, that you want the experts in any particular role to figure out how to apply it best to that role.
3. What is the best way to encourage experimentation and inquisitive attitudes while making sure people aren’t just playing for playing’s sake?
From what I’m seeing, this Uber example of everyone experimenting too much is not the problem organizations are dealing with. The problem is people don’t have time in their workdays to figure out how to use it, and they don’t have anyone to instruct them on how to use it appropriately.
Companies are encouraging experimentation, but I don’t think that experimentation is happening.
You have to start with a coalition of the willing. In every organization, there’s going to be a third of the people who are eager to learn a new tool, a third of the people who will jump on board once they see the initial wave of experimentation is over, and the last third who are going to resist no matter what you do. So, the trick is to get that 66% up to speed.
When super users do find the use cases that work, you need to highlight them across the organization, so people can learn from them.
That’s not happening at most organizations now. Everybody’s experimenting as individuals; nobody’s experimenting as organizations. When I meet with executives — for the last six months — that’s what I’ve been encouraging companies to do.
4. Have you seen any AI introduction strategies that are particularly good? How about particularly bad?
I hate to say it, but most often it’s bad examples. They’re using the old playbooks.
One bad practice I’m seeing is companies are providing a one-day training session and then expecting people to use it. Another is expecting the 20-somethings to figure it out because they are “AI native.”
Fortunately, we’re getting graduates familiar with using these tools in a professional setting as we ramp up our AI efforts. But I can tell you, most 20-somethings without the right training don’t know a lot about these tools, and even those that do don’t know enough about the organization to use it responsibly, reliably, and in ways that change workflows.
I’ve heard smart business leaders at very good companies focusing on these two strategies, and I just want to pull my hair out because it’s not going to work.
What works is when companies sustain a group of super users who keep meeting on a regular basis to experiment and learn with a subject matter expert who can focus them in a productive direction. That way, you can take one expert and magnify it in multiple ways across the organization.
Then, in round two, you get the people in that first cohort of “super users” to lead their own cohorts. That’s how you grow it over time.
Start with the “coalition of the willing,” build your innovation groups, and don’t do training as a one-off.
6. How should managers think about the savings from AI? Is it a short-term or long-term investment?
I think that’s the wrong question to ask, but it is what everybody’s talking about.
I tend to envision the benefits into three buckets: automate, informate, and transform. That’s sort of an (information systems) model that’s been around since the ’80s.
Automating is when you get a system to do the tasks automatically. That’s great, but everybody’s going to get that low-hanging fruit. Informating is when the system gives you different information so you can do your job differently.
The third is going to be changes that transform your work, so you can go about your job entirely differently because of these tools. My thesis is that’s where the companies are going to benefit, not in paving the cow paths and by doing the same type of business faster.
They should be asking, “What are the things that we can do fundamentally differently that’re going to change our competitive environment?”
Companies that win are going to be the ones that take those savings and then invest them back into those transformative options, because just saving and getting efficiencies and automating assume the business environment is going to stay static. It won’t.
That’s sort of the biggest gong I’m playing. Companies need to focus on what new competitive opportunities these tools can create for their organization.