With this fresh technology, the answers relate not to “if” but “how” – whether to build or buy, the investment required, the risks to manage and the talent to assemble – and will depend on a company’s specific circumstances.
The democratization of Generative AI through ChatGPT and similar widely available tools has sent the hype cycle into overdrive. CIOs across industries are getting a lot of “advice” from their CEOs, executive peers, consultants, and technology vendors on the hottest areas ready for the application of GenAI. Everyone has an opinion — and is likely doing their own GenAI experimentation on the side.When these sorts of game-changing technological advances begins to take hold, there is often a call for a new C-level role to take charge like a Chief Digital Officer, Chief Data Officer, or more recently a Chief AI Officer. In my opinion, that’s almost always the result of a leadership vacuum and a CIO who’s failed to embrace the new leadership opportunity.
But CIOs, especially those with enterprise process and operations responsibilities, are one of the only executives with an end-to-end purview across of the enterprise and an understanding of the organization’s key pain points. That puts them in the best position to realistically assess GenAI opportunities.
Now is the time for CIOs to take the technological bull by the horns to identify and implement the right GenAI solutions for their enterprise. Savvy CIOs have a right to own this space. But to do so, they must first understand which GenAI solutions make the most sense, both in terms of business relevance and feasibility. There are some foundational questions CIOs must answer to move forward with GenAI in the enterprise.
What GenAI Is and Isn’tBefore we get to those questions, it’s important to clear up some common misconceptions about GenAI. There’s long been much overuse and misuse of the term AI and additional confusion about GenAI specifically.
Many companies have been using AI for a number of years now, often alongside machine learning, for use cases involving forecasting, predictive modeling, and structured data analytics.
GenAI is different. Its primary capability, as the name implies, is to generate text, video, images, or code that mimics a human creation. Some key GenAI use cases include:
- Distillation or Summarization. The foundational models of generative AI (such as the large language model that powers ChatGPT) can sift through enormous volumes of unstructured data to boil content down to its essence. This is particularly useful in areas such as legal contract analysis, contact center sentiment analysis, and product recommendations.
- Intelligent Conversations. GenAI tools can produce underlying content to manage or mimic human interactions. The output can be audio or text and can be applied to elevate “chatbot” capabilities.
- Content Generation. Trained GenAI models can also be used to generate static content. This can be particularly useful for creation of email responses, product marketing content, legal contracts, or computer code.
By Mary K. Pratt
4 Key Questions to AnswerHaving a better understanding of what GenAI is and its potential use cases in the business is a good starting point to determine where these capabilities might add value. But there are other questions a CIO must consider to get an accurate idea of feasibility and where to begin.
1. Should I build or buy foundational models?
Underpinning any GenAI application is the foundational model(s) that enables the solution. These models must to be trained on vast amounts of data in order to generate text, video, images, audio, or music. Depending on the particular business application, a CIO could use one or more foundational models.
One of the primary decisions to make is whether to buy or adapt existing foundational models or build one on your own. Without a clear strategy concerning which foundational models you are likely to use over time, this can become a very expensive decision in the long run.
There are some key factors to think about that will help CIOs zero in on the right approach:
- Privacy or Intellectual Property. How strategic or sensitive is the underlying data you’ll be using to train the model? Companies that are not comfortable sharing their data as part of an opensource foundational model may find they need to take a different tack. CIOs must also consider where they have the necessary rights to use the data they need in a foundational model.
- Data Availability. What data do you need to create the foundational model? CIOs will need to assess whether they have access to sufficient quantities of that data to create an effective model. What’s more, they’d need to tag and catalog all the structured and unstructured data necessary for the model. In many cases, a company may not have the ability to do this for themselves.
- Quality & Accuracy. The quality and accuracy of some publicly available foundational models may not be adequate to meet shareholder, client, regulator, and employee expectations. This may force a company to adapt a public model or build their own.
- Effort. It takes much more time to build a foundational model versus using something off the shelf. Competitive pressures may force this decision.
- Security. It’s essential to put in place the necessary security controls – access, authentication, or otherwise – to create, manage, and operate GenAI models in the enterprise.
The pricing mechanisms for GenAI capabilities and tools are evolving and will surely change over time as new usage models emerge. For now, however, CIOs need to be aware that the costs associated with GenAI applications as well as the cloud infrastructure required to support them will continue to rise rapidly in the near term.
Today’s foundational models have different pricing levels depending upon use case, but are typically based on the volume of content being managed. But it’s not just the cost of foundational models that CIOs must factor into their business cases. There are many additional expenses to calculate, including labor (in particular, the highly skilled talent required), data provisioning and cataloging, model training, and systems integration. In addition, hyperscalers other large infrastructure and SaaS providers have made significant investments building their GenAI platforms, and they’re likely to recoup these costs through various pricing schemes.
CIOs would be wise to map out a longer-term strategy to start. While there’s value in testing out a variety of GenAI applications, without a clear strategy governing which applications, models, and data are will be developed, CIOs will find themselves with a complex, multi-vendor, and therefore more expensive environment.
IT leaders should take the time to project all GenAI expenses, applying the same rigor and prioritization processes they do with the rest of their IT investment portfolio with a clear and transparent view into costs and benefits.
3. How will I manage the risks?
There are several well-documented risks associated with the use of GenAI. While there may be an enterprise risk management function and owners of specific risk categories, CIOs must to lead the way here as well, as the responsibility for managing this risk ultimately falls to IT. The ownership of privacy risks may typically reside with the chief legal counsel, for example. But given how intertwined privacy, data, and security are, the CIO will need to establish the right operating and governance models to ensure that these risks are identified during the building and operation of any GenAI applications.
As a best practice, CIOs should convene and chair an GenAI Risk Committee with the active participation of all the appropriate risk domain owners. Some key risk areas to manage will include GenAI quality and accuracy, inherent bias and model governance, data privacy, intellectual property protection, and cybersecurity.
4. Do I have the right talent to make this happen?
Recruiting and retaining talent with the right skills and experience to work with GenAI is a critical challenge. The skills and capabilities required may be significantly different than those needed for traditional systems development and implementation.
A number of capabilities that may need to coalesce to effectively implement GenAI systems, including UI/UX, machine learning/prompt engineering, data science, domain-specific data engineering, foundation model engineering, data cataloging and annotation, security, and systems integration. The hardest to source roles will be ML/prompt engineers and foundational model AI engineers, and CIOs will need to create a plan for hiring, training, or developing these skills. The rest may already exist within larger enterprise technology teams or can be acquired via third party consultants.
An Opportunity for the CIO
While there are key questions to consider and many unknowns in the rapidly evolving GenAI space, there is a clear opportunity for CIOs to take the lead. And now is the time. Those CIOs that do so will enable their companies to drive effective enterprise transformation with these powerful new capabilities and will also better position themselves to take on broader leadership roles and responsibilities.