As I have written a couple of times before, 2024 will be the year of truth for GenAI. Truth in the sense of proving the delivery of real business value to technology users.
Why did the Metaverse disappear as fast as it came up? Or blockchains, for that matter? These technologies didn’t deliver a significant value add, let alone a killer application that proved their necessity.
This does not necessarily require a “killer application” but there certainly need to be supported business scenarios that go well beyond trivial ones like the summarization of opportunities, cases, and accounts, or the creation of mediocre content. Even meeting summaries and minutes with action items are not that exciting. Other scenarios include creating job descriptions, semi-automated vetting of resumes in applicant tracking systems, and (as a response to that?) the increasingly common practice of enlisting GenAI support to automate the creation of said resumes and cover letters. We are talking low-hanging fruit here.
On the other hand, organizations of all kinds are looking at GenAI, or even AI in general, as a transformational technology that helps them transform their businesses. The currently available GenAI scenarios that provide this big value of doing their work very differently or even creating new capabilities for them. We are yet to see this scope of AI.
On top of this, a recent report by investment bank Goldman Sachs asked whether GenAI involves too much spend for too little benefit. One of the experts interviewed for this study said:
“Given the focus and architecture of generative AI technologies today … truly transformative changes won’t happen quickly and few – if any – will likely occur within the next 10 years”.
Another one added:
“AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn’t designed to do”.
The cost of training and running a contemporary GenAI system is excruciatingly high, from the high prices that the currently dominating Nvidia chips are commanding to the high energy and utility costs that running AI data centers are incurring. You may remember that end of March 2024 Microsoft and OpenAI announced to build a $100 billion data center. In comparison, a typical data center comes in at far less than a billion dollar. These very high costs are also why vendors increasingly look at shrinking their models while retaining their power. One of the last examples is OpenAI’s GPT4.o mini on July 18, 2024.
With all this, one thing is clear: There is not one AI system that is fit to solve all problems. Instead, we live in a multi-modal and multi-model world and probably will continue to do so. In fact, another recent study found that the average number of AI models that are used productively in the participating organizations is at a staggering 158.
This essentially means that organizations are counting on specialized models, which is partly because many vendors deliver AI models that are embedded into their applications.
To be able to use all these models in conjunction, it is necessary to orchestrate them. Or else, they are merely helpers to make the same — often broken — processes more efficient.
In consequence, this may very well mean that the AI killer application is the orchestration of different, purpose-specific AIs that then collaboratively solve business problems.
But what does this mean for my business?
Obviously, this has some implications for business leaders who want transformational – instead of incremental – change. By their very nature, harvesting these benefits will take some time and require a strategic and long-term approach.
The need for a multi-modal and multi-model approach translates to the need for orchestrating several different models to enable their collaboration toward business goals. Raising this additional value makes the AI game very strategic and even more of a platform and ecosystem game.
That makes the AI platform decision that every company needs to take, even more important. And make no mistake, it is a platform decision as a pureplay best-of-breed strategy will drive integration cost to an extent that they will be prohibitive for achieving and maintaining an agile, adaptable business strategy, if interfacing cannot be achieved and maintained via a GenAI-driven system.
That is driven, as in the AI is in control, not supported, as in the AI helps.
In addition, an AI platform strategy helps reduce the complexity that inevitably comes with this orchestration need.
Based on this AI platform, the IT architecture that supports the foreseeable business needs and is flexible enough to work with surprises will be developed. The architecture must define how different AI models that may be offered by different vendors securely and safely interface with each other.
Once the platform is decided and the architecture is designed, it is about defining the right business challenges with their associated KPIs that shall get tackled. The business challenges need to be reasonable enough to have an impact on the business (as opposed to PoC type of implementations) and the KPIs need to enable the measurement of success along the desired value axis. Ideally, these KPIs can also be translated into cost (reduction) or revenue/profit (increase). Prioritizing these business challenges within a program then helps to incrementally build out the AI platform while validating it at every step along the road.
Some execution strategies
Execution towards this needs to be thought about in five dimensions:
- Investments need to balance between immediate needs and long-term objectives. While the delivery of high-value use cases may take some time, it is important to not forget about low-hanging fruit on the way. It helps to partner with leading AI experts to stay abreast – or ahead – of technological advancements.
- A solid IT strategy with a platform approach and the CIO having a seat at the table are key. While the lines of business own the requirements, IT needs to govern how solutions to these requirements are delivered. This strategy will help avoid costly best-of-breed solutions and point solutions that hamper business agility and drive cost.
- Corporate culture needs to embrace innovative thinking and experimentation. It also needs a culture that does not condemn failures. This allows the teams to creatively solve business challenges or create new business opportunities.
- Personnel needs to be continuously upskilled and trained to be able to work with a blend of external and internal talent. Self-reliance is important.
- Last, but not least, a strong governance framework needs to be established to ensure ethical use, compliance, and a continuous alignment with business goals. This implies a constant monitoring of the performance of all AI models. Especially compliance will become a critical topic.
Considering these five points will help in avoiding some of the pitfalls and complexities of an orchestrated AI landscape while opening up the opportunity of harnessing the full potential of GenAI to sustainably transform their businesses.
