What two years of GenAI has taught us about unlocking value

A lot has happened in the two years since generative AI burst into the enterprise consciousness. The technology arrived on a wave of hype, excitement and FOMO — fear of missing out.

Most technologists were curious at minimum and wasted no time doing what they do best — embracing the latest innovations with a test-and-learn approach to see how it works and whether it might have internal applications. That then evolved into a formal pilot — then pilots, plural — as more internal teams and functions sought to test whether generative AI could make their operations and ways of working simpler and more efficient.

Since then, generative AI has reached near ubiquity as a top-five strategic priority and innovation initiative in organisations across the spectrum of industries and sectors. This has led to transformative use cases, including: more efficient clinical trials to allow potentially life-saving drugs to reach patients in need much sooner; improved health outcomes for patients by creating curated treatment plans; and real-time multilingual natural language communication support that can bridge language barriers between regional teams, creating new opportunities to collaborate and build business value.

This transformative trend is set to continue through 2025, with the technology augmenting an even broader range of tasks and making people more productive.

A key metric of success will be the ability of organisations to translate a higher proportion of their generative AI work into innovative and transformative outcomes that have real-world, measurable value. Currently, while some organisations have achieved tremendous value from generative AI, others are still trying to figure out where the value of the technology lies.

While the end goal has always been about realising value, it’s clear the two-year anniversary for the technology is where the rubber hits the road.

Generative AI’s continued development and evolution may assist in this regard. While the GenAI revolution started out with a limitation of producing only text-based output, today the technology is multimodal and capable of generating content in many different formats, including still images, audio or video. This is opening the door to many more creative use cases, and the potential value is also exponentially larger now than even a year ago. The challenge for organisations is to find it, and to apply it successfully, and at scale.

Research by Everest Group shows that of more than 1200 GenAI PoCs in 2023, less than 18% reached a production stage. Organisations that can’t achieve appropriate value from their use of the technology risk becoming a statistic.

Lessons from successful organisations show the pathway to realising value is being methodical in the approach to and embrace of generative AI. This means determining which model is best suited for which purpose; what is the most efficient infrastructure set-up to power GenAI initiatives and ambitions; and how to improve the accuracy of the model’s output and outcomes.

The value of a methodical approach

If there is one criticism that can be levelled at some of the earlier experimentation with generative AI, it’s that there was no method to the madness. GenAI was implemented by different teams and functions, often in isolation of others. As excitement about the technology took over, GPU capacity or access was purchased or leased to get experiments underway. In the initial excitement, there was often no overarching strategy or coordinated approach to adoption.

This led to organisations encountering a range of challenges. Some of these are technology-related: such as experimenting with public models that lacked explainability, security or privacy protections, or trying to apply AI to every business problem, regardless of whether an alternative solution existed and was better suited. Organisations that moved too fast often also found themselves in the crosshairs of regulators, particularly if sensitive data was entered into the AI, or the AI hallucinated problematic responses; or in the crosshairs of executives and the board, when well-intentioned experiments delivered only siloed or limited value, and the path to expand GenAI was still unclear.

The hangover of this approach — of going too big, too fast on generative AI — is clear. Having said that, other organisational cohorts that took a different approach — such as those that proceeded with excessive caution, or who simply sat back and did not move forward at all — may also ultimately have landed in the same place, unable to unlock the value of generative AI.

This shouldn’t dissuade organisations from chasing generative AI, given the enormous potential value from getting it right. But it does reinforce the need for a more measured approach to adoption.

Value comes to organisations that prioritise a methodical approach. Unlocking GenAI’s potential must start with a well-constructed plan, eliminating any guesswork and avoiding ‘me-too’ approaches. From there, define the problem that is to be solved and specify what outcomes are expected. Be selective in using GenAI: make sure that you’re using it in a sensible manner, and only to solve problems that couldn’t otherwise be solved. A skilled technology partner can assist in this process with services and capabilities that enhance exploration, discovery, understanding and assessments of options.

The value of efficient delivery

In addition to being methodical, leading organisations have learned that one of the keys to achieving value from generative AI is to operate the capability efficiently and cost effectively. This means having an efficient infrastructure set-up to support the needs of generative AI models, promoting efficient resource utilisation, and balancing cost against value that can be generated. Recent ESG Research found only slightly more than one-third (37%) of organisations believe their infrastructure and data ecosystem is well-prepared for implementing GenAI solutions.

Generative AI can get expensive fast, because it requires tremendous amounts of data, compute and IT infrastructure. Such intensive workloads also require a considerable amount of energy and cooling capacity to function. This will have an impact on the organisation’s sustainability goals, and this will need to be factored in.

A data landscape spanning edge, core and hybrid cloud environments is crucial. The IT infrastructure must be meticulously right-sized, cost-effective and sustainable, addressing critical needs in storage, computing (for both training and inference) and networking. A key value realisation aid here is to work with data experts to establish and implement robust data management solutions and strategies that address data security and integrity wherever that data may reside.

Top image credit: iStock.com/Kitinut