Optimizing your AI costs: Don’t drive a Ferrari to the mailbox

Many middle market organizations have a Ferrari-to-the-mailbox problem when it comes to artificial intelligence. While a Ferrari can reach a destination faster than a less powerful car, it isn’t always necessary and the outcome is the same. The lesson is that objectives should drive tool selection. The best choice is not always the most sophisticated or expensive option, but the one that delivers the required outcome with the right balance of efficiency, cost and capability.  

Yes, it’s more fun to drive the Ferrari, just like it’s more exciting to use the latest popular, and likely more powerful, AI tool. But the value of AI is not simply productivity; it’s the creation of new opportunities for business growth. If the Ferrari goes 120 miles per hour, the driver can lose control and end up bobbing and weaving through traffic—or worse, in the ditch. By using the right AI tool at the right speed, businesses can control costs while getting the most possible business value. 

Putting the right cars in the company motor pool

Every company needs a motor pool with a range of tools of different functionality, not a garage full of Ferraris. In the early days of AI, many departments and even individual employees purchased their own tools, which created a web of disconnected platforms and strategies. Maximizing AI return on investment starts with making organization-wide tool decisions and establishing a process for adding new solutions.  

Organizations should start by determining what benefits a tool provides to the company in energy consumption, data usage and overall cost. The first step is to understand the application stack and how much of it is token-based versus subscription-based. Subscription-based tools often have a lower overall cost, which makes them a good fit for employees who need basic AI functionality. Because many vendors are moving to token-based pricing, it is important to monitor pricing changes. 

The next step is to evaluate current business needs against the tools that have already been purchased. Instead of jumping on the newest and shiniest option, consider whether existing tools can do the job just as effectively as a more powerful AI solution. If there appears to be a capability that is not already covered by your current platforms, determine the ROI for that use case. With that information, you can assess whether the business value of the task justifies the investment in additional tools.

Deciding who gets the keys to the Ferrari 

Once the organization understands which tools belong in the motor pool, the next challenge is determining who actually needs the keys. Instead of giving all employees access to all tools, the decision should be based on business need. Because tasks change over time, organizations should evaluate access regularly.  

This decision requires creating policies and governance around tool access. Organizations should fully understand all capabilities of lower-cost platforms before granting access to higher-level tools. Employees whose tasks can be completed with lower-cost tools should only have access to those options. However, businesses should create a process for employees to request access to more powerful tools and make a case if a project requires them. 

Many companies make this determination based on roles, such as all employees who hold a position of manager or higher get access to the high-level tools. This process is oversimplified and, unfortunately, almost always the wrong decision. Employees at lower levels are often the ones doing the work that needs extra functionality, more than leadership. 

Leadership should also consider employees’ resiliency and willingness to go on the AI journey when determining where to dedicate more advanced AI capabilities. By making the decision at an individual employee level based on their specific tasks and soft skills, organizations can make sure that only those employees who truly need the features have the keys.

By making the decision at an individual employee level based on their specific tasks and soft skills, organizations can make sure that only those employees who truly need the features have the keys.

Issuing driver’s licenses

Many organizations provide employees with multiple AI tools and leave it to individuals to decide which one to use, without providing them with proper training or reskilling. When employees have access to powerful and expensive tools without that knowledge, a common result is that they can write more emails of lower quality in less time. 

Receiving the car keys and knowing how to drive are two very different things. Companies often assume that employees, especially those at higher levels, have the skills needed to use AI. Similar to driver’s licenses, companies need to create a process to reskill and evaluate employees on their ability to effectively and accurately use AI. Because the technology changes so quickly, the time frame for that process should be much shorter than the driver's license renewal cycle, perhaps every three to four months for evaluations instead of once a decade.  

Many organizations fail at training by making it too broad and not actually teaching employees how to use the tools appropriately. Most employees need to learn the language of AI before being able to make the best decisions; it’s often a foreign language until they acquire the vocabulary. Reskilling should start at a basic level that focuses on terminology and concepts and then moves on to how to use various features.  

Because each role has a different need for AI, training should be tailored to employees, their business needs and their AI experience. Next, the training should focus on how to identify use cases that provide value, not just higher productivity. When businesses equip employees with both the critical thinking and technical skills they need to use AI effectively, they build the foundation to help the company continue deploying AI at scale.

Teaching employees when to drive the Ferrari

Knowing how to use AI is only part of the equation. Employees must also know when and where to use different AI tools. Because many employees will be given access to multiple platforms, training needs to include how to choose the right tool for the job.  

Often, more powerful tools have a lower learning curve, making it easier for nontechnical employees to use them through significantly easier prompting. By providing training on the lower-cost tools, organizations can gain value by giving employees the right skills to use the more appropriate tool for the task. 

Without education specifically focused on tool selection, employees often choose more expensive models, either because they are easier to use or under the assumption that faster is better. For example, a PowerPoint agent may be appropriate to build an internal presentation, while the extra cost of Cowork may be worth it for a keynote presentation at an industry conference. The reason for choosing Cowork isn’t simply speed, but the ability to create a presentation that is more likely to convert attendees into customers.

The goal is ROI, not horsepower

For the last few years, organizations could ignore inefficient AI usage because the costs were largely hidden inside subscriptions. However, tokenization has made every AI interaction visible. Now, the key challenge is not whether employees use AI but selecting the right tool.  

It’s tempting to focus on creating the most powerful AI system possible. However, the winners will not be the organizations with the latest technology. Instead, the leaders of the race will be businesses whose employees consistently use the right tool for the right business task.