The market is officially three years post-chat GPT and many pundit bylines have turned to using terms like “bubble” to suggest reasons behind Generative A not realizing a material return outside of a handful of technology suppliers.
In September, MIT Nanda Report Waves were made as the soundbite was picked up by every author and influencer that 95% of all AI pilots failed to scale or deliver a clear and measurable ROI. McKinsey A similar trend was published earlier indicating that agentic AI will be the way forward for enterprises to reap huge operational benefits. at The Wall Street Journal‘ Technology Council SummitAI technology leaders recommended that CIOs stop worrying about AI’s return on investment because the benefits are hard to measure and would be inaccurate if they tried.
This puts technology leaders in a precarious position.
For decades, deployment strategies have consistently followed a cadence where tech operators avoid destabilizing business-critical workflows to replace individual components in tech stacks. For example, a better or cheaper technology doesn’t make sense if it jeopardizes your disaster recovery.
While the cost may increase when a new buyer takes over mature middleware, the cost of losing part of your enterprise data because you’re midway through migrating your enterprise to a new technology is far more severe than paying more for the stable technology you’ve run your business on for 20 years.
So, how do businesses get a return on investment in the latest tech transformation?
The first rule of AI: Your data is your value
Most articles about AI data are concerned with engineering tasks to ensure that an AI model is up against business data in repositories that represent past and current business realities..
However, one of the most widely deployed use cases in enterprise AI starts with specifying an AI model by uploading file attachments to the model. This step extends the AI model’s range to the content of uploaded files, speeding up accurate response times and reducing the number of gestures required to get the best response.
This strategy relies on sending your proprietary business data to the AI model, so there are two important considerations in parallel with data preparation: First, governing your system for appropriate privacy. And second, developing a deliberate negotiation strategy with model vendors, who cannot advance their frontier models without access to non-public data like your business data.
recently, Anthropic And Open Eye Large-scale deals have been completed with enterprise data platforms and owners because there is not enough high-value underlying data publicly available on the Internet.
Most enterprises automatically prioritize their data privacy and design business workflows to maintain trade secrets. From an economic value perspective, especially considering how expensive each model API call really is, trading selective access to your data for services or price offsets may be the right strategy. Rather than buying/on-boarding a model as a typical supplier/procurement exercise, think about the mutual benefits of your suppliers moving forward with the model and the potential for your suppliers to realize mutual benefits in adopting the model for your business.
The second rule of AI: Boring by design
According to The information is beautifulIn 2024 alone, 182 new generative AI models were introduced to the market. When the GPT5 hit the market in 2025, many models from 12 to 24 months earlier were declared unavailable unless subscription customers risked cancellation. Their previously stable AI workflows were built on models that no longer worked. Their tech providers thought customers would get excited about the latest models and didn’t realize the premium that business workflows put on stability. Video gamers are happy to upgrade their custom builds throughout the lifetime of the system components in their gaming rigs, and will upgrade the entire system just to play a newly released title.
However, the behavior does not translate into business rate operations. While many employees may use the latest model for document processing or content creation, back office operations cannot sustain changing the tech stack three times a week to keep up with the latest model drop. Back office work is boring by design.
The most successful deployments of AI have focused on deploying AI to business problems unique to their business, often working in the background or working to accelerate but overriding critical tasks. It combines the best of both worlds by removing legal or expense audits from manually traversing individual reports but keeping the final decision in the zone of human responsibility.
The important point is that none of these functions require constant updates to the latest model to deliver this value. This is also an area where abstracting your business workflows using direct model APIs can offer additional long-term stability while maintaining the option to update or upgrade core engines at the speed of your business.
The Third Principle of AI: Minivan Economics
The best way to avoid upside-down economics is to design systems to align with users rather than vendor specifications and standards.
Many businesses fall into the trap of buying new suppliers or cloud service types that their business can use, rather than what their business can do with the capabilities they have deployed today.
While the Ferrari marketing is effective and the automobiles are truly stunning, they drive the same speed through school zones and don’t have nearly enough trunk space for groceries. Keep in mind that each remote server and model is touched by a user layer to minimize costs and design for redundancy by reconfiguring workflows to minimize costs on third-party services.
Many companies have found that their customer support AI workflows add millions of dollars in operational run-rate costs and add more development time and cost to update implementations for OPEX forecasting. Meanwhile, companies that decided that a human could read less than 50 tokens a second—at the speed a human is walking—were able to successfully deploy scale-out AI applications with minimal additional overhead.
There are many facets to unlocking this new automation technology. The best guidance is to, in practice, design for freedom in the core components of the technology to prevent disruption to stable applications in the long term, and take advantage of the fact that AI technology makes your business data valuable to the development of your tech suppliers’ goals.
This content was developed by Intel. It was not written by the editorial staff of MIT Technology Review.