Chapter 4. How to kick off your generative AI strategy

You need a well-thought-out generative AI strategy if you want to get maximum value from your efforts. This strategy should cover key aspects like defining the app's purpose, identifying the target users, understanding the risks, getting legal measures in place, and selecting the tools. By following these steps, you can develop a robust strategy for building an effective and engaging generative AI app.

Step 1. Create an AI tiger team

An AI tiger team brings experts from around a company together not only to guide strategy, but also to deliver a high-quality generative AI app efficiently and effectively that meets business goals and user needs. Therefore, including a diverse set of roles can ensure the project's success.

The team should have a product manager who defines the app's vision, strategy, and roadmap, ensuring alignment with business goals and user needs. There should be domain experts from the business to provide subject matter expertise relevant to the app's use case and help guide the development process and validate the app's outputs. And important to any GenAI app project are lawyers and advisors who can address potential legal and ethical concerns, such as intellectual property rights, content moderation, and bias mitigation.

The team will require at least one software engineer to build the app's architecture, integrate the AI models, and ensure the performance, maintainability, and scalability of the app. It will also need a UX/UI designer to create intuitive and engaging user interfaces, ensuring a seamless user experience and effective communication of the app's generative outputs.

One or more AI/ML engineers and data scientists are a crucial part of any AI tiger team for developing and implementing the AI models and algorithms that power the app's generative capabilities, The team should also include testing and QA specialists to identify and fix bugs, ensure the app meets performance and quality standards, and verify the accuracy and appropriateness of the generated content.

The exact composition of your tiger team might vary depending on the complexity and scope of the GenAI app—and how you plan to develop it. If you use low-code, for example, you might not need as many engineers, designers, or testing and QA.

Step 2. Create a GenAI playground

Building a GenAI application involves exploring and experimenting with a variety of models and techniques. Therefore, the next step is to set up a GenAI playground–a single place for comparing and testing different models. Playgrounds enable developers and others in an organization to explore and learn about generative AI by engaging with different cutting-edge foundation models and prompts in a single practical and educational setting.

There are playgrounds available from AWS, OpenAI, Github, NVIDIA, Hugging Face, Quora, and others. These playgrounds are web-based, and some offer APIs that enable the integration of parts of them into some development platforms. However, some require developers to import data, which can be problematic if the data isn’t test data. To enable their employees to experiment with GenAI safely, companies as well-known as Walmart and CBRE have built their own safe internal playgrounds so that developers, IT, and the business can get hands-on GenAI experience AI without risk of data leaks or exposure. If this appeals to you, here’s what you need to do.

Define your goals

Determine what you want to achieve with your AI playground. Do you want to create text, images, music? Do you need translation capabilities? Will you be embedding generative AI in your applications or are you building a GenAI application? Your goals will guide the choice of tools and models.

Choose your AI models and tools

Select the generative AI models you want to experiment with. For text generation, models like GPT from OpenAI or Claude from Anthropic are suitable. For image generation, consider DALL-E or similar models. Popular frameworks include TensorFlow, PyTorch, and Hugging Face's Transformers library.

Set up your environment

Ensure you have a robust development environment. You'll need a host machine with a good GPU or access to cloud-based services like AWS, Google Cloud, or Azure. You should install Python on your system because most AI libraries are based on Python and virtual environment tools like virtualenv or conda to manage dependencies.

Install necessary libraries and obtain API access

Install the libraries required for your chosen models. For some models, like GPT-4, DALL-E, and Claude, you may need to get API keys.

Write your playground code, test, and refine

Create scripts to interact with the AI models. After you’ve finished, continuously test your playground, refine the models, and tweak the code to improve the quality of the generated content.

By following these steps, your organization can set up a generative AI playground tailored to your needs and start exploring the capabilities of different AI models. Or, if you’re using a low-code platform, you can skip some of these steps. No matter how you make your playground available, it will be your most valuable resource for learning and honing GenAI skills, and it provides a place for exploring use cases.

“Through the power of curiosity and play, developers learned what they needed to do to integrate. We did not add staff to our team or send anyone to training. We were able to scale quickly without having to scale the engineering team itself. Plus, we enabled the playground for business staff members, business technologists, and anybody else who wanted to play and learn.”

Joanne Markow, Vice President, Digital Transformation, OutSystems

Step 3. Explore possible use cases

There are numerous potential use cases for GenAI in your business and industry. The specific application you build will depend on factors such as company goals, target audience, and the nature of your products or services. To determine the most relevant and valuable use cases for your organization, the tiger team should collaborate closely to identify and prioritize opportunities.

Start by determining whether you should build a GenApp for one of the use cases. The popular cross-industry use cases described in “Where GenAI embedded in apps is making the most impact” can help you determine if you want to improve an area of your business. You might also prefer to address a use case specific to your industry. GenAI applications are being used successfully in most of the major industries. Here are just a few examples.

Generative AI use cases in the financial services industry

In the world of finance, there are several use cases for GenAI. For example, a generative AI fraud detection application can review alerts to determine which are false and remove them. GenAI-driven chatbots can handle customer inquiries with finesse, providing swift and precise answers on their mobile banking apps. There are GenAI personal banking assistants that help customers manage expenses, plan for the future, and simplify budgeting.

GenAI use cases in the insurance industry

For insurance, GenAI is rewriting the underwriting playbook, assessing risks with accuracy and fairness while using natural language to explain them. Claims processing, once a manual ordeal, now races to resolution, thanks to GenAI automation. Carriers are already rolling out generative AI into their mobile and apps so customers can obtain quotes, file claims, and track policy statuses using prompts and natural language instead of “canned” bot questions.

Generative AI use cases in the manufacturing industry

Manufacturing has many processes and activities, leaving a lot of room for GenAI applications. Imagine product managers telling an app you’ve built what they want to make, and it generates different designs and features for them to choose from. Another could predict how much inventory to keep based on production plans and sales forecasts, along with the best time to add more to avoid running out or overstocking.

Generative AI use cases in government

GenAI opens up a world of possibilities for government. You could build a GenAI app that delivers realistic 3D models and reports on the impacts of different designs to help urban planners create more livable, sustainable, and equitable communities. Or, a GenAI system could use real-time data from 911 calls, traffic cameras, weather sensors, social media feeds, and more to generate predictive models and recommendations for resource allocation, evacuation routes, and treatment priorities.

Step 4: Identify risks

There are potential GenAI pitfalls that can affect the reliability, security, and ethics of your apps: bias, hallucinations, cyberthreats, copyright infringement, and opaqueness. Here’s what to look out for and what you can do to avoid these issues.

Legal risks

GenAI models can generate content that closely resembles existing works, such as articles, images, or code. This affects intellectual property rights and could even be. If your application generates content that is too similar to copyrighted material, legal challenges and financial penalties could be on your horizon. How can you stop this theft? Use techniques that detect and filter out potentially problematic content and make sure you have the necessary licenses and permissions.

Privacy and security risks

GenAI models are trained on massive datasets, many of which have sensitive or personal information. This can lead to privacy breaches. Additionally, GenAI models are vulnerable to black hats maliciously managing input data to generate misleading or harmful outputs. Implementing robust data protection measures, such as encryption, access controls, and data governance policies, as well as adversarial training and detection techniques, can protect data.

Governance risks

GenAI apps are capable of generating content that isn’t factual (hallucinations) and making biased or unfair decisions. For example, an AI model trained on a medical image dataset can learn to identify cancer cells, but if there are no images of healthy tissue, the AI model may incorrectly identify that tissue as cancerous. Also, because GenAI tools like ChatGPT, Claude, Sora, and DALL-E 2 are freely available, they can be used in ways that weren't intended, such as making private company data public—either in the tools themselves or in web content.

Because of these risks, it’s important to make sure everyone involved in developing and using AI is on the same page and following best practices. Plan to use fact-checking mechanisms, human oversight, or confidence scoring to filter out unreliable outputs to prevent hallucinations.

Ethical risks

If an AI model is trained on data with biases, such as gender, racial, or cultural stereotypes, it may generate content that perpetuates them. This can lead to discriminatory or offensive output that can harm your users and damage your brand. An example is a tech giant’s AI software for job applications that discarded any engineering candidate who went to an all-women’s university because the model had been trained on the resumes of its all-male team. Diverse and unbiased training data and techniques like bias detection and correction can mitigate this risk.

Transparency risks

GenAI models are complex and opaque, so it can be hard to see how they arrive at their outputs. This is problematic for legal, financial, or medical applications that require audits and traceability. To address this risk, you should develop explainable AI techniques, implement human-in-the-loop oversight, and establish clear accountability frameworks.

Knowing these risks and using the mechanisms, tools, and techniques described here can help your development team produce unbiased, factual, secure, original, and explainable GenAI applications—for all kinds of use cases in all kinds of industries.

Step 5: Assess readiness

No one should jump right into a GenAI application development project without taking a good hard look at what is needed. To assess your organization's readiness for building generative AI applications, your tiger team should start by asking questions about your current talent, tech, and learning status, looking for gaps and determining how to address them.

Determine IT and business readiness

Does your IT organization have the right skills and experience in AI, ML, and software development? These roles are critical for designing, building, and maintaining GenAI systems. What about domain expertise? You should also consider the business knowledge in the areas where you plan to embed GenAI.

Check your tech and data stacks

How about the necessary computing power and infrastructure? GenAI can be resource-intensive. So it’s important to make sure you have the hardware and cloud services to support it, as well as the ability to scale up as needed.

Are your data assets in good shape—diverse, high-quality, and well-organized? GenAI relies heavily on data, so having a robust and relevant dataset is essential. Think about both the quantity and quality of your data, and whether it covers a wide range of scenarios and edge cases. These are all key ingredients for successful GenAI projects.

Identify gaps

What are you missing that could hold your project back? Maybe you need to invest in training or hiring to build up your team's expertise in GenAI technologies and best practices. Look for opportunities to collaborate with academic institutions or industry partners to tap into additional knowledge and resources.

Perhaps your data needs some work to be ready for prime time—cleaning, labeling, and ensuring it covers a wide range of scenarios. This can be a time-consuming process, so plan accordingly. Or it could be that you need to upgrade your tech stack to handle the demands of GenAI, like high-performance computing or specialized software tools. Consider whether you have the budget and resources to make these investments.

Include governance and ethical considerations too. A big gap can be not having policies in place to ensure responsible and unbiased GenAI development. You should not move forward until you address that issue and ensure that everyone on your team is aligned with these principles. Be honest about where you're falling short, so you can make a plan to fill those gaps and get your organization GenAI-ready. With the right preparation and mindset, you'll be well on your way to unlocking the power of GenAI for your business with just the right app.

Step 6: Identify a pilot

Just like you shouldn’t rush into a GenAI project without assessing your readiness, you should also try not to take on too much at once. Therefore, you should start with a pilot project. But it can’t be just any pilot either, because choosing the right one is critical for testing your GenAI solution. The pilot needs to have a clear business goal and be feasible with your technology. Here are other key criteria that can help you identify a suitable pilot.

Have a clear problem to solve

Don't start without a specific business outcome in mind. Pick a well-defined problem. You want something that's not too complex, but still meaty enough to showcase the potential of GenAI. Maybe it's automating a repetitive task, like generating product descriptions or customer service responses. Or perhaps it's creating new content, like social media posts or marketing copy. The key is to find a use case that's relevant to your business and has clear success metrics. You need it to demonstrate value—and even ROI.

Define measurable outcomes

Know how you'll measure success before you start by identifying the key performance indicators (KPIs) that matter most to your business. These could be things like time saved, content quality, or user engagement. Next, set specific, quantifiable targets for each KPI, such as "reduce content creation time by 30%" or "increase click-through rates by 10%." Finally, make sure you have the tools and processes in place to track and report on these metrics throughout the pilot.

Secure access to relevant data

Every pilot needs a good dataset for training. It should be diverse and representative of the real-world scenarios your GenAI application will address. It’s possible that you will need to do some data collection and curation before you can get started. And don't forget about data quality—you need it to be clean, consistent, and properly labeled.

Consider the user experience

Designing a smooth and intuitive UX will be key to getting buy-in from your users and stakeholders. Think about who will use the application and what they need from it. Put yourself in their shoes. Think about how they'll interact with the AI-generated content or outputs, and what they'll expect in terms of quality, relevance, and ease of use. Make sure the GenAI features integrate with existing tools and workflows and provide clear guidance on how to use them effectively. Remember the human touch and provide a way for users to provide feedback and get support if needed.

Step 7: Choose technologies

Chapter 2 and Chapter 3 of this ebook provide information on the majority of tools needed for traditional generative AI app development, deployment, maintenance, and updates—and demonstrate the difference low-code can make. To choose the technologies for your pilot and your full GenAI-powered application, your GenAI tiger team should review all that information in depth and do their own research into solutions.

You should also take into account the future of those tools, platforms, and capabilities. The GenAI application development world is still in flux, so take a look at what appears to have staying power. That will go a long way in choosing the generative AI development path that’s right for your organization.