Amidst the ever-evolving landscape of data-driven decision-making, a new entrant has taken center stage: Generative AI. This cutting-edge technology has the potential to revolutionize how we extract insights and make business decisions from data. However, that potential comes with an equal weightage of open questions.
Despite the buzz…
What realistic steps should data leaders take to integrate generative AI into their plans?
How will teams avoid the hype and quickly reap the benefits?
In this article, I share my perspectives on Generative AI's capabilities and limitations while providing tangible insights to empower data leaders to navigate the fine line between innovation and hype.
Near-term, I expect the application of generative AI within data and analytics to primarily involve improving end-user experience, simplifying business operations and assisting engineering teams with programming-related tasks. For example, a data engineer might call upon a language model to create data ingestion and transformation scripts, configuration templates, or SQL queries.
Similarly, an analyst could leverage the capability to produce DAX code tailored to Power BI, while a data scientist may generate Python code to facilitate various machine learning functions.
Within 12 months, you will be asked to provide an AI-driven Analytics Assistant to almost every role in your organization. At this point, your data teams will go far beyond writing SQL, creating data pipelines, or building models. Today's users already have more questions than a dashboard can answer, and they'll prefer a conversational UI over a scripted click sequence.
Once provided with a natural language response to basic queries, those users will want to go deeper, uncovering hidden patterns in data and receiving updates proactively. This is where large language model (LLM)-driven conversational interfaces will be indispensable.
LLMs can aid your data teams by comprehending user intent, providing automated troubleshooting support, and generating narratives. Imagine a co-pilot-like experience for your users – delivering personalized analysis while developing a deeper understanding of the data.
LLMs can accelerate the creation of knowledge graphs, ontologies, or data structures derived from structured and unstructured sources. To support this, your data teams must acquire new skills in prompt engineering, model validation, and content validation. They may also need to generate inputs for reinforcement learning to fine-tune models.
But to assume that generative AI applications will replace Data Engineers, Analysts, and Scientists in the near term would be … implausible…. Keep reading to understand why.
Generative AI apps are far better suited for deep analysis and pattern recognition than typical reporting and analytics. LLMs are probabilistic, creative, and inductive - they thrive in scenarios where precision is unimportant and definitive answers do not exist, such as categorizing content, responding to inquiries, or translating languages (including programming languages).
However, the very nature of analytics-driven business decisions necessitates pinpoint accuracy, precise response, and time-based analysis tailored to specific data and knowledge models, where LLMs prove inadequate.
While a generative AI app may craft a basic sales report from a neatly structured dataset, it will not support analysis of a market spanning various regions, product categories, and fluctuating consumer preferences, all streaming in from multiple data sources with unique formats, frequency, and granularity.
In these (everyday) scenarios, the precise, rigorous, and deductive nature of a Data Analyst who understands the nuances which govern how data is applied proves far superior.
Generative AI apps can assist in areas where analysts reach the limits of their creative thinking. These apps can form multiple plausible hypotheses to explain anomalies while leveraging the abilities of an analyst to conduct further analysis to test these hypotheses before arriving at a conclusion.
These apps can also effectively explain predictions, or the process followed to reach a decision, thereby increasing trust and accelerating decision-making.
So, where do you start?
A recent indicates that 45% of companies saw an increase in AI investment in 2023, and 68% of executives believe that generative AI benefits outweigh its risks. Despite this enthusiasm, organizations need help to get started: Only 19% of teams are actively running pilots. Generative AI can be overwhelming as it opens a massive scope of opportunity.
Notably, three disruption patterns surpass the capabilities of other existing AI methods: Content Consumption, Content Generation, and Acceleration in Tech-enabled innovation. As such, a good starting point would be to consider how these patterns may be applied within your organization to solve genuine business problems.
Here are a few steps you can take to embark on your Generative AI journey:
1. Add a conversational UI to your Data & Analytics apps: You likely have access to everything you need to do this today – all you need to do is ask your team to do it. Challenge your team to provide a conversational interface for all user queries. Start by using chatbots to offer automated troubleshooting, guiding users with a co-pilot-like approach (e.g., helping users locate reports or explaining metrics).
As confidence grows, use them to enable natural language querying and perform deeper analysis. Eventually, integrate these analytical chatbots into your organization's collaboration tools like Microsoft Teams or Slack.
2. Train your data teams to be story writers: As chatbots replace traditional dashboards, information isolation may increase due to dependence on specific queries that the users ask. In this scenario, your data teams will need to assume the role of data journalists to address this gap.
Leveraging Generative AI's content generation capabilities, they can distribute timely updates, presenting recent trends from market analysis, unforeseen shifts in customer engagement metrics, or novel insights from user surveys in the form of engaging narratives. Each update should be intricately interconnected, crafting a compelling storyline that underscores prevailing patterns across the organization.
3. Identify candidates for a Generative AI pilot: The possibilities are endless with Generative AI. Some of your users may already be considering or experimenting with its applications.
Imagine auto-populating metadata catalog, generating synonyms, providing transparent calculations for better decision-making, enhancing data scientists' efficiency by using synthetic data for stress testing and feature engineering, or simulating diverse scenarios with risk scores for informed decisions. Consider conducting a Generative AI environmental analysis within your organization to identify high-value use cases and prioritize areas for a pilot initiative.
About the Author:
An entrepreneur, mentor, and change-maker, Shashank Garg is passionate about all things Data. Renowned for his role as a trusted advisor to multiple Fortune 500 CXOs, he has played a pivotal role in leveraging Data & Artificial Intelligence to spark transformative shifts in businesses. As the co-founder and CEO of Infocepts, Shashank spearheads a data solutions firm that champions enhanced business outcomes through more effective use of Data & AI-driven, user-friendly Analytics. With a mission to bridge the gap between the worlds of Business and analytics, Shashank motivates his team to enable data-driven transformations with a sharp focus on tangible business value.