In the summer of 2023, technology executives and IT services executives will be hit by a lightning strike when they discover that AI automation is a valuable tool. McKinsey released “The Economic Potential of Generative AI: The Next Productivity Frontier” report. It was reminiscent of the 2010s, when Amazon Web Services launched a campaign aimed at Main Street C-suites: Why would any fiscally savvy executive allow their IT teams spend capex on servers and software if AWS only costs 10 cents per virtualized machine? Vendors know that aggressive advertising and reports about competitive risks for an industry sector will drive many calls to the C-suite from boards, and from the C-suite down to staff asking “What are you doing with AI?”. There was no time at that point to differentiate between “AI novelty” applications that were more Rube-Goldberg machines than tangible breakthroughs, and actual business benefits from applying AI. Today’s opportunity: Significant gains in automation
When business leaders react to an immediate panic, they often create new risks and mitigations. Two recent examples illustrate the risks of rushing to implement AI and publish positive results. The Wall Street Journal reported on April 2025 that companies were struggling to realize returns from AI. It covered MIT’s retraction of a technical report about AI in which the results that led it to publication could not have been substantiated.
These reports show the dangers of over-reliance on AI, without using common-sense safeguards. However, enterprise AI adoption is not completely off track. AI and related technologies are being used to automate processes in a variety of industries with incredible results. Where can you find the most value in applying AI to automate your business now that we have passed the “fear” stage?
Although chatbots have become almost as common as new apps for mobile phones, AI applications that result in automation and productivity improvements are based on the architecture and purpose of the AI system upon which they are built. The two dominant patterns in which AI gains are currently realized boil down to language (translations and patterns) as well as data (new formats and data searches).
Natural language processing
Manufacturing Automation Challenge: FMEA is critical and labor-intensive. FMEA is not always done prior to a manufacturing equipment failure, so it’s often performed in a stressful scenario. Intel’s global footprint of manufacturing plants separated by large distances, along with time zones and different preferred languages makes it even more difficult to pinpoint the root cause of an issue. The FMEA analysis is repeated for large fleets across these facilities, requiring weeks of engineering effort.
Solution :– Use already deployed CPU compute servers to perform natural language processing (NLP), across the manufacturing tool logs where local manufacturing technicians maintain observations about the tools’ operation. The analysis used sentiment analysis to classify positive, neutral, or negative words. The new system performed FMEA in less than one minute on six months’ worth of data, saving weeks of engineering and allowing manufacturing to service equipment proactively on a preemptive schedule instead of incurring unexpected downtime.
Financial Institution Challenge: The programming languages used by software engineers are evolving. The mature bellwether institutions are often formed by a series mergers and acquisitions, and they still rely on systems that are based upon 30-year-old languages that software engineers of today are not familiar.
Solution Use NLP for translation between the old programming languages and the new ones, giving software engineers the boost they need to improve the serviceability critical operational systems. Use AI to replace a risky rewrite, or a massive upgrade.
Example 2: Company product specifications with generative AI models.
The time required to reformat the product data of a company into a customer’s RFP format is a challenge that has existed across industries. Teams of sales and tech leads spend weeks reformatting the root data into the preferred PowerPoint or Word formats for different accounts. Customer response times were measured by weeks, especially when RFPs required legal review.
Using generative AI in conjunction with a data extracting and prompting technique known as retrieval augmented creation (RAG), companies are able to quickly reformat product information into different RFP response formats required by customers. The time it takes to move data between documents and document types, only to discover an unforced mistake in the transfer is reduced from weeks to hours.
HR automation challenge: Navigating through internal processes can be confusing and time-consuming for both HR and employees. The consequences of misinterpretation and access outages are extremely important for the company and individual.
Solution Combine RAG with generative AI and an interactive chatbot to determine access rights and identity based on employee-assigned assets. Provide employees interactive query-based chat formats that answer their questions in real-time.
Identifying your best AI use cases
With 80% to 90% AI proof of concept failing to scale, it is time to develop a framework based on caution. Consider starting with an assessment of your data strategy and governance. Find opportunities to compare AI-based automation at other companies through peer discussion. The best way to start a successful AI journey in your organization is with clear, rules-based processes and policies. Maintain tighter decision controls when you encounter disparate sources of data (e.g. unstructured video, structured databases, or unclear processes) to avoid unexpected data exposure or cost overruns.
As AI hype cycles cool and business pressure increases, it’s time to get practical. Apply AI to well defined use cases to unlock the automation benefits which will be important not only in 2025 but for many years to come.
Intel produced this content. This content was not written by the editorial staff of MIT Technology Review.
