A Strategic Guide to Identifying the Right Use Cases

Introduction

As businesses plan to implement generative AI, the number of use cases can seem daunting. The possibilities are endless–from text content creation and document summarization, to code generation, opportunities abound. However, as you explore generative AI’s business potential, are you falling into the trap of treating it as a one-size-fits-all solution? As they say, when all you have is a hammer, everything becomes a nail. But, generative AI is more than just a hammer. It’s only one of the tools in your organizational toolbox with various uses, and understanding which ones fit your business best can make all the difference.

The growing interest in generative AI, driven by apps like ChatGPT, Claude, Gemini, and Copilot, has captured the imaginations of many. There’s certainly a lot of enterprise experimentation going on–a survey from VentureBeat suggested that more than 70% of organizations are testing the generative AI waters, but less than 20% are planning to spend more on the technology.^[1] Given the immense productivity gains to be had, what is holding organizations back?

Given the hype, identifying suitable use cases and deployment approaches perplexes even the most adept executives. We all understand that data, analytics, and AI are crucial for informed decision-making. However, seeing beyond the buzzword bingo and staying up-to-date with the latest trends and best practices remains challenging.

This article provides a guide to identifying appropriate use cases and deployment frameworks for generative AI. By understanding what generative AI is good at and what trips it up, you can ensure that you’re not treating it as just a hammer but rather, leveraging its strengths to drive innovation and growth in your organization.

Types of Generative AI Models

Did you know there are at least nine different types of generative AI models? In the scientific paper, ChatGPT is not all you need. A State of the Art Review of large Generative AI models, the authors categorize generative AI modes according to their inputs and outputs.^[2]

The nine categories of large generative AI models based on their input and output formats are:

1. Text-to-text models, probably the most familiar, generate textual content based on textual input. An example is ChatGPT, which can perform tasks like question-answering, text generation, and text summarization.
2. Text-to-image models generate images based on textual descriptions. DALL-E 2 and Stable Diffusion are examples of text-to-image models that can create realistic and artistic images from textual prompts.
3. Text-to-audio models generate audio content based on textual input. AudioLM is an example of a text-to-audio model that can create audio from textual descriptions.
4. Text-to-video models generate video content based on textual input. Phenaki is an example of a text-to-video model that can create videos from textual descriptions.
5. Text-to-code models generate code based on textual input. Codex is an example of a text-to-code model that can generate code from textual descriptions.
6. Text-to-scientific text models generate scientific text based on textual input. Galactica is an example of a text-to-scientific text model that can create scientific articles and papers from textual descriptions.
7. Text-to-algorithm models generate algorithms based on textual input. AlphaTensor is an example of a text-to-algorithm model that can create algorithms from textual descriptions.
8. Image-to-text models generate textual descriptions based on image input. Flamingo is an example of an image-to-text model that can create textual descriptions of images.
9. Text-to-3D models generate 3D images based on textual input. Dreamfusion is an example of a text-to-3D model that can create 3D images from textual descriptions.

Now that we understand the different generative AI model types and their capabilities, let’s turn to the use cases.

Don’t Forget Traditional AI

Back to my hammer-and-nail analogy, the hubbub surrounding generative AI is at an all-time high. As organizations ponder the different applications and approaches to implementing generative AI, they should not forget regular, old, traditional AI and analytics. Ultimately, your solution will likely amalgamate traditional and generative AI. My article titled Generative AI vs. Traditional AI: What’s Better? provides a more comprehensive overview but I’ll provide a summary of other, battle-hardened, algorithm types available.^[3]

The different algorithms and use cases include:

• Prediction and forecasting: used for making predictions about future events or trends based on historical data, such as risk prediction, customer churn, and sales/demand forecasting.
• Planning and optimization: focuses on optimizing resources and processes to achieve specific goals, including operations research, optimization, and route planning.
• Segmentation and classification: involves categorizing data or objects into distinct groups based on shared characteristics, including clustering, customer segmentation, and object classification.
• Recommendation systems: provides personalized suggestions to users based on their preferences and behavior, such as recommendation engines, personalized advice, and next best action.
• Anomaly detection: involves identifying and flagging unusual or abnormal events within data, including abnormal transaction detection, outlier detection, and monitoring.

These uses represent a broad landscape of AI applications that will likely be combined with generative AI.

Identifying Use Cases for Generative AI

Aligning Generative AI with Business Objectives

Before beginning any data, analytics, or AI endeavor, you must start with a clearly defined business objective. It’s OK to have different functional areas across your business test and experiment with generative AI tech, but implementing an enterprise-wide solution is an entirely different matter.

The first step in aligning generative AI with business objectives is clearly understanding the organization’s needs and goals. The following figure provides a good starting point for understanding the process.

Figure 1.1: Understand the Business Decision to be Made

The key to making generative AI work is to truly understand how your business will change due to implementing generative AI. If we’re generating content at scale but have a bottleneck at a downstream step in the process, then your organization will fail to realize the true value of generative AI.

In addition to the above, organizations should also identify the key performance indicators (KPIs) and key results that generative AI can deliver. For more on KPIs, please see my TowardsDataScience article The Art of the AI KPI.^[4] Don’t skip this step; metrics drive behavior!

Once business needs and goals are clearly defined, the next step is to assess how generative AI can contribute to achieving these objectives. This involves identifying areas where generative AI can provide significant value, such as automating repetitive tasks, improving decision-making processes, or enhancing customer experiences. Business leaders should pay special attention to the potential benefits and risks of implementing generative AI in these areas. After all, it’s prone to confabulations, perpetuating bias’ and stereotypes, infringing on intellectual property (IP) rights, and leaking sensitive information. For more on AI risks, see my article The 12 Hidden Risks of ChatGPT and Generative AI.^[5]

After identifying potential use cases for generative AI, carefully evaluate the feasibility of implementing these solutions. This includes assessing the availability of required data, the organization’s technical prowess, costs, and implementation timeframes.

Suggested Actions for Business Leaders

• Develop a clear understanding of your organization’s business objectives and the challenges and opportunities in the current business landscape.
• Move beyond experimentation and identify potential use cases for generative AI within your organization, focusing on areas where it can provide significant value.
• Evaluate the feasibility of implementing generative AI solutions, considering factors such as data availability, technical capabilities, costs, and implementation timelines.
• Engage relevant stakeholders in this process to ensure a comprehensive understanding of the organization’s needs and the potential impact of generative AI.

Data Quality and Quantity

Generative AI models’ success relies heavily on the quality and quantity of the data used for training. However, since building a LLM from scratch is prohibitively expensive and time-consuming, most will start with a foundation model (FM), which means the quality of the LLM data is already set and out of the organization’s control.

However, to provide context to generative AI, many organizations are turning to retrieval augmented generation (RAG) to allow for your organization’s data to be used as a knowledge base. See my article, Generative AI’s Force Multiplier: Your Data, for more details.^[6] Having pristine data quality for fine-tuning and RAG is entirely within your organization’s control. I’m not saying it’s easy, but it is a solvable problem that your company can address.

High-quality data is essential for fine-tuning generative AI models or for RAG so they can accurately and effectively perform their intended tasks. To help with this, there are two things to consider. The first is understanding where your organization’s data is and the quality of it. For most organizations, analysts and data scientists spend months searching for and gaining access to the right data for their projects. How do they do this? They generally phone a friend and meander through the corporate data jungle until they find what they need. It’s a sad state of affairs, but true. However, smart companies use a data intelligence platform like Alation to help ensure that all of the data is cataloged and of a known quality level.

The second factor to consider is data collection. If the data needed to make a business decision does not exist or is scattered about, data engineers and data scientists will have to collate and shape it so it’s in the right format for the algorithms to process. When doing this, teams should focus on collecting relevant, diverse, and representative data to ensure that the models can generalize well and avoid biases. Additionally, the data should be clean, consistent, and error-free to prevent the models from learning incorrect patterns.

The amount of data required for training generative AI models can vary depending on the specific use case and the complexity of the task. While some tasks may require large amounts of data, others may only need a smaller, more targeted dataset. Business leaders should work with their data engineering and data science teams to determine the appropriate amount of data needed for each use case, considering factors such as the model’s complexity, the desired level of accuracy, and the availability of computational resources.

Data privacy and security are also important considerations when working with generative AI models. Business leaders must ensure that the data used for training adheres to relevant privacy regulations and ethical guidelines. This includes obtaining proper consent for data collection, anonymizing sensitive information, applying AI guardrails, and implementing robust security measures to protect the data from unauthorized access. For many analytics and data science teams, the data usage policies are separate from the data. Again, this is where a data intelligence platform can come in handy–it can apply governance policies at the point of consumption.

Additionally, business leaders should be aware of the potential risks associated with generative AI models, such as the generation of fake or misleading data, and take appropriate steps to mitigate these risks. Please see my three-part blog series for more details.

1. AI Oversight: Crafting Governance Policies for a Competitive Advantage
2. AI Oversight: Implementing Governance Policies for a Competitive Advantage
3. AI Oversight: Strategic Imperatives for Successful AI Governance

Suggested Actions for Business Leaders

• Conduct a thorough data assessment to identify available data sources, evaluate their quality, and determine their suitability for generative AI use cases.
• Implement robust data governance practices to ensure the ongoing collection, management, and maintenance of high-quality data.
• Collaborate with data engineering and science teams to determine the appropriate amount of data needed for each use case and develop strategies for acquiring and preparing the required data.
• Address data privacy and security concerns by adhering to relevant regulations, obtaining proper consent for data collection, anonymizing sensitive information, and implementing robust security measures.

Ethical Considerations and Risks

Generative AI models can inadvertently perpetuate biases, generate misleading or false information, infringe on privacy rights, and leak sensitive information. Business leaders should be aware of these potential ethical issues and proactively address them. This includes understanding the limitations of the models, assessing the potential impact on various stakeholders, and ensuring that the use of generative AI aligns with the organization’s ethical principles and values. See my TowardsDataScience article Generative AI Ethics for more details.^[7]

In addition to ethical concerns, deploying generative AI solutions can pose various risks to organizations, such as reputational damage, legal liabilities, and security threats. Business leaders should conduct a thorough risk assessment to identify and evaluate these risks, considering factors such as the sensitivity of the data used, the potential consequences of model errors, and the organization’s risk appetite. This assessment should inform the development of risk mitigation strategies and the establishment of appropriate safeguards.

To address ethical concerns and risks associated with generative AI, business leaders should develop and implement strategies that promote responsible and safe use. This includes establishing an ethical framework for the use of generative AI, engaging in ongoing risk assessments, and implementing robust governance and oversight mechanisms. Organizations should also invest in employee training and education to ensure that those working with generative AI models know the ethical considerations and risks and are equipped to address them effectively.

Suggested Actions for Business Leaders

• Develop an ethical framework for the use of generative AI that aligns with your organization’s values and principles.
• Conduct a thorough risk assessment to identify and evaluate the potential risks associated with deploying generative AI solutions.
• Implement robust governance and oversight mechanisms to ensure responsible and safe use of generative AI models.
• Invest in employee training and education to raise awareness of ethical concerns and risks and equip employees with the skills to address them effectively.
• Engage in ongoing risk assessments and monitor the ethical implications of generative AI use to ensure that your organization’s approach remains aligned with its ethical framework and evolving best practices.

Conclusion

While generative AI has gained significant attention, it is crucial for IT leaders to understand when and when not to apply generative AI for specific use cases. After all, creating a promotional marketing email is quite a bit different than using generative AI for interpreting your organizational policies–just ask Air Canada.^[8] The hype surrounding generative AI can lead to its misapplication, resulting in higher complexity, failure, and diminished business value.

Organizations can make more informed decisions about the best approach for their specific needs by systematically evaluating use cases and considering alternative AI techniques. Use cases in prediction and forecasting, planning and optimization, segmentation, and classification are not currently a good fit for generative AI models, but they will be used in combination with generative AI to solve business challenges.

Combining generative AI with other AI techniques will create more robust systems that mitigate some of generative AI’s limitations, such as its tendency to generate inaccuracies and hallucinations. By developing the ability to combine the right AI techniques, organizations can build AI systems that have better accuracy, transparency, and performance while also reducing costs and the need for data.

To avoid failure and maximize the value of AI in their organizations, IT leaders shouldn’t forget the other tools in their toolbox, which include analytics, traditional AI, and generative AI to address business issues. Don’t swing the generative AI hammer just because it’s there, not everything is a nail.

---

If you enjoyed this article, please like it, highlight interesting sections, and share comments. Consider following me on Medium and LinkedIn.

---

Please consider purchasing my latest TinyTechGuide: Generative AI Business Applications: An Exec Guide with Life Examples and Case Studies.

If you’re interested in this topic, check out TinyTechGuides’ latest books, including The CIO’s Guide to Adopting Generative AI: Five Keys to Success, Mastering the Modern Data Stack, or Artificial Intelligence: An Executive Guide to Make AI Work for Your Business.

---

^[1] “More than 70% of Companies Are Experimenting with Generative AI, but Few Are Willing to Commit More Spending.” 2023. VentureBeat. July 25, 2023. https://venturebeat.com/ai/more-than-70-of-companies-are-experimenting-with-generative-ai-but-few-are-willing-to-commit-more-spending/.

^[2] Gozalo-Brizuela, Roberto, and Eduardo Garrido-Merchán. 2023. “ChatGPT Is Not All You Need. A State of the Art Review of Large Generative AI Models.” https://arxiv.org/pdf/2301.04655.pdf.

^[3] Sweenor, David. 2023a. “Generative AI vs. Traditional AI: What’s Better?” Medium. November 27, 2023. https://medium.com/@davidsweenor/generative-ai-vs-traditional-ai-whats-better-f2f9e86a61ef.

^[4] Sweenor, David. 2022. “The Art of the AI KPI.” Medium. September 13, 2022. https://medium.com/towards-data-science/the-art-of-the-ai-kpi-3d6ed8a03994.

^[5] Sweenor, David. 2023. “The 12 Hidden Risks of ChatGPT and Generative AI.” Medium. December 16, 2023. https://medium.com/@davidsweenor/the-12-hidden-risks-of-chatgpt-and-generative-ai-e7bb1ec45f00.

^[6] ​​Sweenor, David. 2023a. “Generative AI’s Force Multiplier: Your Data.” Medium. September 20, 2023. https://medium.com/@davidsweenor/generative-ais-force-multiplier-your-data-3763e8ed59df.

^[7] Sweenor, David. 2023a. “Generative AI Ethics.” Medium. July 28, 2023. https://medium.com/towards-data-science/generative-ai-ethics-b2db92ecb909.

^[8] Belanger, Ashley. 2024. “Air Canada Has to Honor a Refund Policy Its Chatbot Made Up.” Wired. February 17, 2024. https://www.wired.com/story/air-canada-chatbot-refund-policy/.