Maximizing AI ROI: A Cross-Industry Guide
▼ Summary
– Enterprises should consider exchanging selective data access for service benefits or cost reductions rather than automatically prioritizing confidentiality.
– Business AI workflows require stability over constant updates, as frequent model changes disrupt back-office operations designed to be “boring.”
– Successful AI deployments focus on automating unique, mundane business tasks without needing the latest models to deliver value.
– AI systems should align with user needs and current business capabilities rather than vendor benchmarks to avoid inefficient spending.
– Design workflows for frugality by minimizing third-party service usage and reconfiguring processes to control costs effectively.
For any enterprise looking to maximize its return on investment in artificial intelligence, a strategic approach that balances data confidentiality with economic realities is paramount. While protecting trade secrets remains a top priority, the significant cost associated with each model API call means that selectively sharing data in exchange for service credits or price reductions can be a savvy financial move. Instead of treating the acquisition of an AI model as a standard procurement task, companies should explore opportunities for mutual growth. This involves collaborating with suppliers to advance their model’s capabilities while simultaneously accelerating your own business’s adoption and integration.
A second critical principle for AI success is embracing systems that are intentionally “boring” by design. The market saw a staggering 182 new generative AI models launch in 2024 alone. When a major new version like GPT-5 arrived in 2025, it rendered many models from the previous one to two years obsolete, disrupting the stable workflows of subscription customers who then threatened to cancel. Technology providers often assume customers will eagerly adopt the newest releases, failing to recognize that business operations place an immense premium on stability and predictability. This contrasts sharply with the behavior of video gamers, who frequently upgrade their custom-built systems to enjoy the latest titles.
This consumer mindset, however, does not translate to the steady, run-rate operations of a business. While individual employees might experiment with the newest models for creative tasks, core back-office functions cannot withstand the chaos of changing their underlying technology stack multiple times per week. These operations are meant to be reliable and, by their very nature, unexciting.
The most effective AI implementations tackle specific, often mundane business problems that are unique to the organization. They typically operate in the background, automating or augmenting routine but necessary tasks. For instance, AI can relieve legal or audit teams from the burden of manually cross-referencing reports, yet it wisely leaves the final decision in human hands. This approach combines the efficiency of automation with the critical judgment of people. Crucially, these applications do not require constant upgrades to the latest and greatest AI model to deliver significant value. Abstracting business workflows from direct model APIs provides an additional layer of long-term stability, allowing companies to update their core technology at a pace that suits their operational needs, not the vendor’s release schedule.
A third guiding principle involves applying “mini-van economics” to your AI strategy. The goal is to avoid inefficient spending by designing systems that align with actual user needs rather than being swayed by impressive vendor specifications and performance benchmarks.
Many organizations fall into the trap of purchasing new hardware or cloud services based solely on supplier-led marketing, instead of beginning their AI journey by assessing what their business can realistically consume. It’s essential to understand the current pace of adoption and the capabilities already in place. While the marketing for a high-performance sports car like a Ferrari is compelling, it’s important to remember that such a vehicle is subject to the same speed limits in a school zone and offers little practical storage for groceries. Similarly, in the AI world, every remote server and model interaction adds to the cost. Designing for frugality by reconfiguring workflows to minimize reliance on expensive third-party services is a fundamental practice for controlling expenses and maximizing value.
(Source: Technology Review)
