A personal assistant for BI activities: How to start producing value with Generative AI today
Generative AI in Business Intelligence
Ever since the sky-rocketing market launch of ChatGPT by OpenAI at the end of 2022, millions of private and business users have had first-hand experiences of the potential value of Generative AI and its impressive content-generating capabilities [1]. Potentially transforming core business activities such as report creation or programming, executives now face the question where to implement generative AI to generate value. In fact, executives expect to see improvements of 7 to 9 percent in areas such as customer interaction or operational efficiency from the adoption of generative AI within the next three years [2].
While some companies currently may lack Generative AI maturity to launch Gen AI products to its customers, internal decision-making processes offer great potential for efficiency gains or new insights. At Capgemini Invent, we have analyzed the potential value-adding impact that Gen AI can have on Business Intelligence process in companies, answering the question: How should companies start to use Generative AI to improve their decision making?
In our opinion, Gen AI can essentially take on the role of a personal assistant to every employee looking to create insights from company data, a key driver to enabling a data-driven organization. To maximize the value of this assistant, we have therefore developed three concrete recommendations along the dimensions of technology, processes, and people.
Benefits
“The current wave of generative AI is a subset of artificial intelligence that, based on a textual prompt, generates novel content” [3]. This spans across diverse domains, including natural language processing (NLP) and visualization. A key driver of the recent popularity of these models are large language models (LLMs) such as GPT-4, PaLM or LaMDA which are trained on large amount of text data and consist of billions of parameters. LLMs are designed to comprehend and generate text that resembles human language, enabling a highly intuitive interaction experience.
To understand the benefits of Gen AI in Business Intelligence, we follow the four phases of our standard BI development cycle:
Let’s illustrate the impact of Gen AI on each phase with the example of a standard dashboard for regional sales managers, containing financial KPIs such as volumes sold, revenue generated per product category and sales representative.
In the phase of the requirements engineering, Gen AI can supplement business analysts to create suitable frameworks and user stories by describing typical user tasks and paint points. It can suggest relevant KPIs, visualizations and data sources based on the respective scope improving the workflow of translating business needs into BI applications. For example, an analyst could prompt an LLM for typical KPIs of a regional sales manager, creating a long list which only needs to be validated with the manager. The LLM could also suggest additional data sources such as weather data or information on regional demographics to improve the steering of the sales force.
Generative AI excels at writing code based on natural language. It can speed up the data preparation phase by reducing the level of required data model knowledge and effort to prepare the data. In our example, the data engineers can provide the required sales data quicker and could utilize the saved time to optimize the solution design of the data pipelines.
In the insights creation phase, Generative AI can support the translation of the data model into the most relevant KPIs and visualization which allow the user to answer their business questions. Recently, Microsoft announced the PowerBI Copilot which creates an entire report with the desired functions and needs by using text input. In addition to this, chatbot-like interaction with the Copilot is possible, with which one can also ask data analysis questions, but also enable the possibility of individualizing reports [4].
Lastly, Generative AI improves the workflow of daily decision making by allowing business users to extract relevant BI insights in a self-service approach. Reports and summaries on BI insights can be created in real-time promoting a close collaboration between BI and decision-making. Q&A functionalities facilitate a more accessible experience to data-driven insights, leading to a reduced dependency of business users on technical resources like data analysts and quicker access to information. Our regional sales managers will be able to decide themselves the appropriate format such as data tables or presentations. Generative AI models can even suggest and run their own root cause analysis leading to new decision-making processes.
Challenges
However, the integration of Generative AI into Business Intelligence processes is challenging because of the fast-paced development of Gen AI, uncertainty around the implementation, the complexity of the technology and unresolved legal and societal considerations.
Most obviously, a lack of expertise on the new technology among management and employees makes the integration into existing BI processes both, technically and legally, challenging. It requires highly sought-after specialists in AI, infrastructure setup, model deployment and integration into the prevailing tech stack.
Once a Generative AI model is utilized in the BI process, the challenge of checking result validity remains as the calculations conducted are hard to reproduce due to the “black box” nature of many of the underlying models [5]. Generative AI MVPs are still prone to errors in its interpretation of the provided information which can lead to invalid results and conclusions. In our example, the financial reports on sold units and revenues will require a result validation before they are used in managerial decision-making or audit-relevant reporting.
In addition, AI models bear the risk of inadvertently generating biased content. Hence, the question of accountability and trust in the results needs to be addressed to ensure that managers can rely on Generative AI models to assist in the data-driven decision-making process. International legal frameworks such as the EU’s AI Act (in development) and GDPR can serve as guidelines for a compliant application of Gen AI.
Finally, for both open-source approaches as well as commercial Gen AI services, the licensing agreements must be investigated closely to assess legal circumstances. Special attention, among other, should be put on copyrights of the output, handling of sensitive, proprietary or personal data as well as potential data breaches.
As the risks of GenAI are met by tremendous transformation potential, Capgemini has conducted further research into solution approaches to overcome challenges [6].
With respect to Generative AI integrations into the BI process, first leaps need to be carefully planned, including a risk assessment and performance monitoring to sidestep pitfalls and ensure reliability outputs to leverage GenAI in the decision-making process.?
领英推荐
Our recommendations
Generative AI is set to revolutionize task management, acting as a personal assistant to handle various responsibilities. By enhancing accessibility to analytics and technology, it enables a broader audience to benefit while shifting focus towards interpersonal tasks or fundamental questions. For our regional sales manager, the gained time can be used for discussing the new insights with the sales representatives themselves. To maximize the benefits of Generative AI for the company, we recommend concrete actions in the areas of technology, processes, and people.
Technology
When it comes to the technology itself, the market for Generative AI is very new and volatile. When deciding whether to build, partner, or buy a solution, it is crucial to consider the company's prerequisites and experience with related technologies such as NLP. More mature organizations might want to try to develop a custom solution themselves while less mature ones should start their journey with small, external solutions such as plugins or features of standard business software.
We recommend starting with a smaller-scale adaptation of Generative AI, tailored to the company's prerequisites and needs. Experimenting with new features in existing software, such as PowerBI’s Copilot, serves as a practical entry point with no costly downside. Deeply integrating Generative AI solutions across the organization and into the existing IT landscape, on the other hand, is as complex as the implementation of other business essential technologies such as CRM or ERP systems.
Processes
As outlined above, there are many use cases for Gen AI in Business Intelligence. For a structured approach, we recommend jointly establishing a use case process in your team on which tasks can be improved with Generative AI. This can be as simple as leveraging LLMs for SQL query generation up to automatically created recommendations in the form of reports on complex business questions.
The regular process should follow an easy three-step approach, e.g., in the form of a focused workshop:
People
As a new technology with a disruptive character, BI organizations need to familiarize their employees with Generative AI and the potential they offer. To facilitate this, we recommend promoting the usage of these new technologies actively among employees by giving them safe access to the tools and encouraging them to explore their applications in their respective tasks. For the beginning, a premium access to a commercial software with limited expenditures is a good start.
To optimize this, organizations should train their employees on the capabilities and limitations of Generative AI. Furthermore, it is essential to provide guidance to employees on how to stay compliant, especially when sensitive or personal company data is involved in the analyses.
These trainings for both developers and business users can showcase the value of these innovations and address doubts or mistrust that may arise. This creates a more receptive environment for their Gen AI application and enables the company to make better, more data-driven decisions.
If you're interested in integrating Generative AI in your business and get most of the benefits, contact the Insights Superpower team!
Thinkers50 Radar | AI Co-Thinkers | Walk the Talk | VP, Co-lead of Capgemini Invent's Management Lab
1 年Thanks Colja Maser. What's your view on the (ex) Code Interpreter, on its likely devopments and in general on OpenAI strategy about data analytics tools?