While traditional AI has focused on detecting patterns, honing analytics, and classifying data, generative AI deals with the creation of new content—something that companies have long been eagerly awaiting as a productivity enhancement as well as a way to improve the customer experience.

At our Future of Data and Analytics roundtable, executives from a number of sectors, including the apparel, health, and manufacturing industries, shared their experiences with this newly emerging technology, as well as enablers such as data quality and governance. They also discussed their hopes of obtaining additional value from generative AI in the future.

Will generative AI prove to be an invaluable business tool or a gimmick?

While futurists envision sophisticated uses for generative AI, our roundtable participants are just beginning to find use cases for it.

At present, they are applying generative AI to contextualize insights from extensive sets of internal and external data. For example, ChatGPT can be used to sort through email threads and provide a summary to the user in order to save them time; it can also then go one step further and generate editable responses based on the user’s writing style. The goal, as a participant from a healthcare company put it, is to “sift through lots of internal data as well as customer and purchased data to deliver insights where and when needed, without jargon.”

It is also being employed as a writing tool to generate quarterly earnings releases and other formulaic writing, with an eye to increasing its usage, as a roundtable participant from the apparel industry put it, “to adopt generative AI for productivity to create documents, spreadsheets, and decks.” The intent of these use cases is to increase employee productivity by helping business users get insights easily and quickly to free up their bandwidth so they can get on with higher-value work, changing their ways of working for the better as an addition to the suite of digital products and services people use.

Generative AI shows great promise as a way to enable business strategies and activate more channels to serve as growth avenues. It can be employed to augment the decisions people make by offering personalized offers to them as they shop in brick-and-mortar locations. In doing so, retailers are demonstrating an ability to do something that online stores have long done, narrowing the gap between the online and in-person shopping experience. Data holds the promise of activating more new, data-driven channels in the future, enabling business strategies such as go-to-market.

In working with generative AI, it is important to build and scale product portfolios across geographies, markets, and functions. “Scaling is a challenge,” noted an executive from a manufacturing conglomerate. “We need to step back from the ‘everyone can do it’ idea because we don’t want the sort of failures that will lead to a loss of trust or enthusiasm.” Successes should build upon successes; high-quality data is needed for these successes to be achieved. As the manufacturing executive said, “Everything depends on how good the data being used is.”

In sum, our discussion uncovered three key points around how generative AI augments decisions and strategies and helps to materialize a data-driven enterprise:

Generative AI can contextualize insights from extensive data sets.
Generative AI can increase employee productivity and grow the top line.
Building and scaling are crucial when introducing and expanding generative AI usage.

However, several key questions remain, including:

How can we decode the model in such a way that teams understand how best to use it?
How can we enable governance to help trust the insights?

As generative AI begins to be implemented, companies will need to ensure sound data governance processes are in place to ensure data is used correctly.

With great power comes great responsibility: why effective governance is needed when working with generative AI

While some dream of companies in which data is freed up to power democratization and self-service across the business, the reality is that such a situation would be more akin to anarchy than a utopia. Effective data governance is a must for a number of reasons.

In order for data assets to be trustworthy, there must be a strong system of governance around them. At some organizations, seemingly everyone is clamoring to put data science to work. But while anyone can generate data-driven insights, there has to be trust in these insights. So much depends on how good the data platform is; a strong governance model and well-defined ways of working are needed to help establish and maintain high-quality data.

In recognition of the importance of governance, 10 companies, including OpenAI, TikTok, Adobe, and the BBC, have signed on to a new set of guidelines on how to build, create, and share AI-generated content responsibly. These guidelines call for the builders of the technology, as well as the creators and distributors of synthetic media, to be more transparent about the technology and to disclose when people interact with content created through generative AI technology.

Data governance should also define the policies and procedures for data security. Sound processes need to be established and maintained to help protect data from unauthorized access or corruption. Procedures are also needed to define the actions to be taken without delay in the event of any sort of data breach or theft.

Another factor influencing trust in data and its outputs is how diffuse or consolidated the business’s data platforms and toolsets are. There is a strong bias toward consolidation, as it enables the business to use its data resources more efficiently and effectively. Of course, most legacy data doesn’t match with this desire, and merger activity and divestitures add to the challenge of creating and maintaining a single source of truth. Data assets need to be connected, and the data foundation continuously reworked and structured to ensure data quality and usability. Without this sort of structure and cleanliness, AI struggles. As one roundtable attendee from the manufacturing sector noted, “We can achieve beautiful results with something like using ChatGPT to summarize and even reply to email threads because it’s working from a complete dataset and a single language. Once you bring in different languages and more datasets or incomplete information, it becomes less effective.”

It’s clear that organization-wide discipline is needed to succeed with generative AI. It has to start at the base, with the quality of the underlying data, and be reinforced through proper data structuring and stewardship via strong governance. It’s a challenge that requires investments in the form of support from the top and sufficient budgeting, but the rewards bring a tremendous potential ROI.

Our discussion uncovered three key insights around the need for governance when working with generative AI:

A strong system of governance helps to build trust in data assets.
Consolidated data platforms and toolsets help to maintain that trust.
Organization-wide discipline is needed to succeed in generative AI efforts.

But some key questions remain:

Is simple reporting using unstructured commands really the future?
If it is the future, how do we ensure consistent output across the organization?
How can we ensure that budget is secured and leaders are receptive?

While some users in the organization may feel like governance is too restrictive, generative AI is far from a mature technology and caution is warranted while it continues to evolve.

Generative AI is still in its infancy and users need to learn to walk with it before they can run

Emerging technology can be managed in traditional ways; the wheel needn’t be reinvented just because new capabilities are being introduced. For example, CRISP-DM, the industry standard methodology for data science life cycle processes since 1999, remains in wide use today for analytics work, including generative analytics. This time-tested research/production/deployment model works well to ensure that the new technology works as intended.

Terminology can be a sticking point with generative AI; because people will put queries forth using different words, it is inevitable that different answers will be generated. As one roundtable participant commented, “Business terminology needs to be standardized because AI doesn’t recognize context and may give bad conclusions if questions aren’t phrased precisely.” Some are hopeful that we will soon be able to employ AI to offer proprietary information on things such as plant-specific inventory levels through voice as easily as we currently do with digital assistants at home to find out the weather forecast, but it is clear that different languages, multiple data sets, and incomplete information all reduce generative AI’s effectiveness. Standardization is clearly called for.

As always with technology, end users are impatient and want the newest product to be delivered yesterday with full functionality and zero bugs. The reality is that tackling something such as generative AI calls for a phased approach that starts by addressing a relative handful of use cases in a couple of domains and gradually scales up to more use cases and domains.

As the work scales up, there will invariably be some rework required and more data assets to integrate. Real-time visibility and connectivity among systems are crucial to obtaining insights from generative AI. Future requests that rely upon generative AI will become increasingly granular, such as customers asking vendors to provide the carbon footprint of different SKUs. In some instances, businesses will be reliant on custom applications powered by generative AI to tackle this sort of question. But as our participant from the fashion industry reminded us, “Generative AI will also be integrated and built into technologies we’re already buying and using like SAP and PowerBI.”

Finally, the power of a tool such as generative AI means that change management should be instilled so end users can learn to make the most of this new capability. This change management effort includes ensuring that open communication channels are in place, resistance is identified and mitigated against, and the new capabilities are harnessed so as not to be abused. Where needed, the change management process can also help to institute a growth mindset in employees and introduce data literacy programs for those who will be new to AI usage as we look to change and improve employees’ ways of working.

Our discussion uncovered four insights around the need to improve and better promote generative AI:

Traditional methods of technology management can be leveraged.
A phased approach around specific use cases helps as a proof of concept that promotes buy-in.
As we scale the proof of concept to include more use cases and domains, some rework may be required to integrate new data assets.
Change management is a must when introducing generative AI capabilities to the business.

Your industry

Your needs

About us

Careers

Insights

Your industry

About us

Careers

Insights