In 2024, 60%* of enterprise Generative AI investments are drawn from innovation budgets. As you plan, will these costs still be considered innovation, or will they shift to OPEX?
If your aim is to include Generative AI in your OPEX, it suggests a focus on proven ROI—while innovation budgets often allow more flexibility regarding returns.
In this newsletter, we provide key considerations before integrating enterprise Generative AI costs into your 2025 OPEX budget.
Early Stage? In the short term, select a use case with a clear business impact. Medium term: set up a pilot project to launch quickly. Test the use case and measure user adoption.
Key element to remember: Scaling up a pilot may seem straightforward initially, but the final stretch can be complex. The first 80% of development appears manageable; the remaining 20% consumes 80% of the effort.
To prepare, stress-test potential vendors by asking how they would handle more than six critical issues that we identified in 90% of projects. Find these questions in our Mastering AI Adoption recap by Squirro CEO Dorian Selz
Organizations typically identify 10 potential use cases, reflecting ambitious goals. When considering OPEX, prioritization hinges on tangible ROI—through enhanced productivity and operational efficiency.
According to the Menlo VC report*, the dominant use cases for Generative AI include:
What does this mean for organizations concretely in terms of tasks?
Last week, we shared five real-world Generative AI use cases from highly regulated industries. Spanning finance to government, these examples demonstrate impactful change and enhanced efficiency. Select your use cases and access our detailed ROI data in our case studies.
Deciding whether to build in-house or source from vendors? In 2024, companies are almost evenly split: 47% build, while 53% buy.
A few years ago, our approach for companies deciding to build their own solutions was to suggest they go ahead and gain firsthand experience with their engineering teams. In most cases, these companies contacted us a few months later, having consumed numerous engineering hours and resources, seeking support for the six specific challenges we mentioned earlier. Remember? The first 80% of development may seem straightforward, but the remaining 20% consumes 80% of the effort. That’s why we recommend stress-testing potential vendors—or your team—by asking how they would handle these six issues we at Squirro have identified in 90% of cases.
Today, our approach has evolved, and we strongly suggest considering the buy approach. Your perspective might change too:
Retrieval-Augmented Generation (RAG) now dominates AI design patterns, with 51% adoption. However, basic RAG isn’t sufficient for enterprise needs. To build robust, trustworthy AI that employees can rely on, you need to implement the following:
Get Your Free Technical Guide on GenAI for Knowledge Management! You'll discover the key tech foundations that make it all possible 👇
These insights aim to support your budget planning for 2025. Still have questions? Contact our expert team—we’re here to help you navigate this exciting period with confidence.
*Data sources: Squirro’s experience and "Menlo VC 2024: The State of Generative AI in the Enterprise" - Non-exhaustive list of recommendations.