The launch of ChatGPT grabbed the interest of business leaders everywhere, catapulting GenAI from its technology silo into high-level business strategy discussions. Since then, it’s become clear that GenAI is here to stay and will add value across all industries. Even so, enterprises may hesitate to move forward with GenAI initiatives because they are unsure which use cases to prioritize, or are worried about the risks around data privacy and algorithmic bias.
Enterprises will find it difficult to resolve these questions about use cases and risk in the abstract. GenAI is an extraordinarily iterative technology — more like a musical performance than a machine. Think of the GenAI model as the instrument, the enterprise’s data estate as the sheet music, and the enterprise itself as the musician. The outputs of GenAI will be intimately tied to the quality and characteristics of an enterprise’s unique data estate. Rather than coming up with some high-level theory of how GenAI will transform their business, enterprises need to start simply practicing with GenAI. Launching early, low-risk GenAI prototypes and proofs-of-concept (POCs) is the fastest way to gain the experience necessary to craft a more high-level GenAI strategy that delivers significant competitive advantages while building robust risk-management frameworks.
Starting a GenAI Prototype or POC
When it comes to GenAI, the most important thing is to simply get started. To identify appropriate GenAI POCs, enterprises should consider how tactical use cases can enhance their near-term efficiency and operational advantages. They should also consider the ability to scale, prioritizing tactical use cases that seem likely to contribute to key business processes and competitive advantage in the long term, but without letting the perfect be the enemy of the good. The business case for a GenAI POC — and how to measure the impact on that business case — should be clarified as part of the discovery process. Clear definitions of success are crucial, as they will decide which POCs are worth scaling.
For organizations facing cost-optimization pressures amid macroeconomic turmoil, GenAI has arrived at the perfect time. Many of the more obvious tactical GenAI use cases — for example, rapid text summarization and automated code generation and testing — are tailor-made for enabling measurable efficiencies and cost-reductions in the prototyping phase.
As GenAI foundation models proliferate, enterprises have numerous options for running GenAI prototypes. Even the simplest prototypes will require understanding the general suite of large language model (LLM) capabilities and finding a reasonable match between the LLM and the use case. Given the costs and skills required to train a foundation model from the ground up, most prototypes will take alternative approaches. These approaches may include working with established players like Google and OpenAI through corporate partnership models, engaging with smaller GenAI startups looking to scale solutions for corporate clients, or standing up a team to deploy open-source models in-house.
Managing GenAI Risks
A prototype-based GenAI strategy inherently manages one of the core risks of any technology transformation: investing too much, too fast without a clear path to business impact. But other risks (including very real legal and reputational risks) need to be managed as well. Fortunately, especially for early prototypes, organizations can begin with “minimum viable governance” frameworks. These relatively simple checklists allow experimentation with new GenAI tooling. Any enterprise exploring GenAI will already have data controls and risk management frameworks in place; by refining those frameworks to address GenAI-related risks specific to their use cases, enterprises can run sandbox experiments with lower risk using offline models on their existing infrastructure.
Of course, governance and risk management frameworks will need to mature as GenAI use cases mature. A prototype setting is a perfect opportunity to develop “responsible AI” checklists that both inform the viability of emerging use cases and identify future needs for risk policies and frameworks.
Use cases that involve communicating with customers or employees, for example, will need strong anti-bias guardrails that draw on emerging approaches like constitutional AI. Any GenAI project that intends to publish GenAI-generated material for public consumption will also need automated checks for copyright infringements. Enterprises can socialize frameworks to address these risks in an iterative way, implementing learning and development activities as prototypes become mature use cases that impact multiple team and business units.
Constraints are GenAI’s friend. But rather than building guardrails for all potential GenAI use cases on day one, enterprises can de-risk their GenAI prototypes by constraining GenAI to very discreet tasks, measuring how well GenAI performs those tasks, and preventing GenAI from doing anything outside of the defined problem space.
An Early Look at GenAI Impacts
Globally, McKinsey’s recent influential report on the economic potential of GenAI forecasts that GenAI could add between $2.6 trillion and $4.4 trillion in value annually across more than 60 business use cases. A Goldman Sachs analysis, meanwhile, predicts that GenAI will contribute to a 7% global GDP increase over the next 10 years.
But even today, early adopters are already experiencing measurable benefits from GenAI, which should give enterprises confidence to ramp up their GenAI experiments. In the consumer industry, generative AI is already helping brands build stronger relationships with customers. We have also observed firsthand how GenAI-led transformations are beginning to deliver concrete value. One media client, for example, used GenAI to automate the generation of contextualized abstracts, resulting in a direct cost savings of 33% in addition to other productivity improvements.
It’s Time to Practice
GenAI outputs will be intimately tied to the quality and characteristics of an enterprise’s unique data estate; the results and ROI, therefore, may not be entirely predictable. Some experiments may be barely audible, and others will resonate across the enterprise.
While the first GenAI prototype may seem like cautious attempts to tune a new instrument, enterprises will quicky find themselves with an orchestra. In time, an enterprise should aim to achieve a GenAI flywheel in which time/effort savings from one initiative can fund further GenAI initiatives. Building a GenAI center of excellence, meanwhile, will help both upskill talent and serve as a hub for exploring the art of the possible. Leaders need to imagine how their organization will function when GenAI is available in every business process.
But just as no amount of reading sheet music turns a student into a concert violinist, no amount of abstract strategizing will achieve remarkable GenAI outcomes. Delivering outstanding business performance through GenAI starts with practice.
