Generative AI is very high on the agenda of large and small enterprises alike. Recent CEO surveys show almost 80% of CEOs believe AI is likely to significantly enhance business efficiencies in their organization. Indeed, because of its accessibility and versatility, Generative AI has separated itself from the broad set of new emerging technologies like blockchain, NFTs and the metaverse, and is widely seen to sit amongst some of the large breakthrough technologies the world has seen, like the internet or mobile.
First movers should have a disproportional advantage and are expected to separate themselves from the followers significantly. Worse still, laggards are expected to lose competitive advantage and durable differentiation in their markets. The case for a double-down focus on Generative AI seems to be settled.
And yet, getting a Generative AI initiative off the ground in a large corporation is harder than it sounds. We saw this in the early years of the digital transformation wave – a very large part of the investments and effort never paid off – and this is likely to happen with Generative AI as well. Looking across the Generative AI efforts at enterprise across vertical industries and learning from the early lessons there, here are a few things to get right:
Strategy is Core
True innovations come at the intersection of disciplines, and with Generative AI it is no different. The power of Generative AI will not come from technology alone – it will take the business and IT teams working together, joined at the hip, to define the best use for any Generative AI projects. Picking the right business problem to address is critical for success – but Generative AI cannot be the answer looking for a question – the teams must start with the questions first.
In many ways, Generative AI is the last piece of a jigsaw puzzle to fall into place after years of innovation in data, machine learning and AI, making the puzzle now complete. But defining Generative AI projects too narrowly to exclude the other aspects of AI – predictive, classification, and machine learning – would be like standing a chair on one leg. Indeed the best use cases we have seen of Generative AI use it for only a piece of the overall solution – and the strategy must therefore invoke a broader engagement of and investment in AI and Data.
Generative AI not only opens up new use cases for enterprises but also can accelerate and scale existing ones. Much of its applications today in enterprises revolve around sales and marketing, customer conversations and knowledge management, business analytics and software development. But we also see so many non-traditional, and outstanding use cases – from creative design to agriculture and farming, to healthcare and medicine. Even more importantly, every knowledge worker will eventually find value in partnering with generative AI and we will see value come from the embedment of generative AI into everyday tools.
Therefore, core to the success of these efforts is the role strategy plays in it – bringing the right owners to the table, picking the right use cases, being very sharp and pointed with capital allocation, driving adoption, and mitigating risks, and applying the learnings in an iterative cycle across company initiatives. Some enterprises may jump right in and reinvent the way that works gets done, while others may prefer to be more measured and experiment with a few use cases first – but in all cases the enterprises that get the technology and strategy right will win over the ones that get just the technology right.
Guardrails and governance are a must
There are appropriate questions on the table about data privacy and security, loss of confidential and proprietary information or advantage, fairness, explainability, accuracy, and hallucinations in results. In addition, applicable regulation is still being shaped, and there are implications for employee motivation, brand perception and consumer opinion. All these risks need to be understood and managed with the proper oversight.
Therefore, rolling out generative AI in an enterprise context requires a structured framework and active governance throughout. Getting the right operating model around this is critical since this is not a one-time effort. And answers are evolving – just in recent months for instance we have turned around two of the largest issues from when we first started – ensuring data doesn’t leave our corporate firewalls – and we don’t inadvertently train an otherwise publicly available LLM with proprietary information from our book of business and intellectual property. Emerging solutions need to be constantly evaluated and if appropriate, incorporated into the governance architecture.
The learning from enterprises that are doing this well is that governance as a responsibility is centrally owned but participation is distributed. What this means is that a diverse team can bring in inputs from the various areas of the company (e.g. data science, engineering, legal, cybersecurity, marketing, design, and other business functions as well as the broader community and industry) but this can’t be a loose self-aggregating team with a self-defined charter – it needs to be managed centrally with direct oversight from the most senior levels of the company.
Build the right technology stack
The ecosystem around Generative AI is early in its evolution and is evolving rapidly, particularly in enterprise use cases. As a result, it is important to evaluate the options that exist and are emerging – and pick the components that best meet your business needs. Three areas require significant attention – data platforms, optimally tuned LLM or foundation models, and security and governance infrastructure.
It’s important to recognize that as AI itself becomes a commodity in the long run, it’s the proprietary data of a corporation that will become the strategic asset around which AI can be orchestrated to drive differentiated value. In fact, if there is anything that Generative AI has done, it has put the spotlight back on Data. Setting up the right enterprise data platform, master data management and data engineering capabilities is key to success.
Enterprises need more advanced Generative AI that is tuned for their industry and enterprise environments rather than the ones consumers use, for instance ChatGPT. This reduces hallucinations, improves accuracy, and increases relevance to enterprise use cases. But there are three very different ways to achieve this technically – from building a custom LLM from scratch – to starting with one as a foundation model and further customizing it – to using a generally available LLM but layering over it a prompt-tuning or prompt-engineering layer. Each have their pros and cons, and it’s the use case and the specifics of the situation that drive what’s best for each enterprise.
Finally, technologies and frameworks that allow the corporation to be intentional about data integrity, confidentiality, and privacy are key. Economics are also a concern in the long run – inferencing is expensive, training is super expensive – so monitoring and management technologies to keep the usage on track is important. And some use cases might require tokenization or other additive technologies to ensure content that is leaving the firewall is adequately protected. All these choices need thoughtful evaluation and execution. It’s important to architect the right technology stack.
Build the right Talent
It’s often said that in the end, talent (and culture) are the only sustainable long-term differentiators for businesses – and this is even more true now with the advent of Generative AI. Three areas in particular are key – human in the loop, prompt engineering, and hiring and rewarding for attitude not skills.
The role of the “human in the loop” becomes critical with Generative AI, and as the AI becomes a copilot and augmentation for teams in the organization, employees need to be able to take advantage of it – skilling becomes a key enabler. Prompt engineering is a new discipline altogether that enterprises are finding themselves to be severely deficient in – it’s the ability to bridge the gap between the “understanding the business” and “working the AI”, and requires thoughtful approaches to cross-skilling.
Finally, the best companies are changing the way they hire, reward and retain talent – with a focus on attitude instead of skills. There is an increasing recognition that the skills needed today will change in the future, as they did from the ones in the past. Focusing solely on skills leaves enterprises underprepared for the opportunities of tomorrow and sends the wrong message to employees looking to grow. Instead, talent strategies that focus on the right attitudes – embracing change, endorsing the new with curiosity and humility ensure that enterprise talent remains ever fresh.
The new career paths are all about learning, unlearning and relearning. Evolving your long-term HR policies in the back may seem a far cry away from driving urgent AI adoption and acceleration in the front, but the truth is it will be just as important for the success of the enterprise in the time of AI.
Every company needs to build their own strategy, governance, tech stack and talent model for Generative AI. Now is not the time to wait and see.