GenAI product and service implementations: How to plan for success

In a recent interview with VentureBeat, Vijoy Pandey, SVP and Head of Outshift by Cisco, and Reynold Xin, Co-founder and Chief Architect at Databricks, discussed the challenges that enterprises face when pursuing Generative AI (GenAI) product and service implementations.

GenAI continues its trend of wide adoption, and the impact it’s bringing can be felt across all industries. However, as enterprises are eager to ride the wave, what are the key considerations that they need to address for successful GenAI integration?

Identify generative AI business use cases for your enterprise

The potential for GenAI integration is so broad that many decision-makers in the enterprise are overwhelmed and paralyzed. As there are so many GenAI use cases, they need to find the best ones for their business. 

Pandey: "What you should do as an organization is concentrate on what your differentiation is, on how you bring differentiated experiences to your customer. Concentrate on solving the use cases that bring those differentiations to your customers. Don't try and solve the entirety of the stack."

When determining GenAI use cases for your enterprise, knowing your business differentiation is key to maintaining focus. From there, we recognize that enterprises have two primary use categories for leveraging GenAI: productivity and product.

GenAI use cases for productivity enhancement

Think about the countless tasks that your organization’s employees perform each day. When GenAI can be used to increase productivity in those tasks, you should consider pursuing it. Pandey highlighted some examples of how GenAI is radically impacting productivity and efficiency in business operations:

• Software development: Using GenAI to generate code, developers can quickly complete simpler coding tasks to free up resources for focusing on more complex problems.
• Recruitment: HR managers use GenAI to summarize résumés and automate contacting potential candidates for follow-up.
• Search: GenAI tools serve as enterprise assistants, powering the search of internal company data with an efficiency that far surpasses that of classical search mechanisms.

GenAI use cases for product innovation

With GenAI, enterprises can build innovative services and products that amplify their differentiation as a company.

For example, Reynold highlighted the problem of data discovery for their customers at Databricks, a data intelligence platform. Their customers—with millions of tables and data sets—struggled to find their data or understand it. To address this pain point, Databricks built an application that uses GenAI to examine existing datasets, infer the nature of the data within, and then create textual descriptions of that data for use in a search engine. This integration of GenAI radically improved the data discovery process for customers.

Prioritizing the use cases based on risk and validation

After identifying suitable use cases for GenAI implementation, how do you know which one to pursue first? To answer this question, Xin frames this exploration by considering two axes:

Xin: “When you think about all the different use cases, plot them onto a map. There are two axes to this map. One axis asks, ‘What is the cost of getting it wrong?’ The other axis asks, ‘Can I validate whether it’s wrong or not?’ If you plot your different use cases onto these two axes, you want to start with the ones where the cost of getting it wrong is low, and validating the effectiveness of your application is easy. If you start with those, you have a much higher chance of becoming successful.”

The diagram above visualizes this framework. As enterprises begin their GenAI implementation journey with low-risk applications that are easy to validate, they build up the tools, expertise, and processes for success. Then, building upon this early success, enterprises can pursue higher-value GenAI applications while striving to remain in the lower-left quadrant. 

Navigating the data management challenges for GenAI success

Building services and products on GenAI underscores the importance of high-quality, secure data and proper data management. Enterprises aiming for success in their implementations must navigate these challenges thoughtfully.

On the question of data quality and data management challenges, Pandey and Xin had this insightful exchange:

Pandey: “GenAI use cases are now pervasive across our organization, which is forcing us to look at our datasets—whether they’re clean, normalized, can talk to each other, and are categorized well. Those are questions which we were punting before because there used to be no need to think about it holistically.”

Xin: “Your data—how you answer all of these questions—become the fundamental pillars for your GenAI applications.”

Pandey: “This is just going to pivot all of us toward improvement in our handling of data.” 

One of the challenges that your enterprise will need to address is model transparency and explainability. AI models can often be seen as a “black box.” You will need to gain clarity on how your GenAI models make decisions (transparency) and make these models interpretable (explainability). A clear path here will help when it comes time to validate—and improve—the accuracy and effectiveness of your GenAI implementation.

The success of any GenAI product or service implementation is closely tied to the data used to train or fine-tune your models. Therefore, your enterprise must tackle the challenge of ensuring high-quality data and securing it well.

• Is your data accurate?
• Are your data pipelines secure?
• Are you monitoring the movement of your data through the pipelines?
• Will you have the capabilities to establish data lineage and provenance so that you can retrace your steps to figure out how your data got to where it is (in case you need to backtrack or fix)?
• Will you adopt responsible AI practices?

Ultimately, enterprises that will be successful in their GenAI implementations are the ones that approach the endeavor as a data and applied machine learning problem. 

Ensuring long-term relevance and compliance

The final segment of the interview wrestled with what enterprises must do to ensure that their GenAI implementations would remain relevant and compliant over the long term. GenAI is a rapidly evolving space, and the regulations around the use of GenAI are evolving, too. In light of this, what should enterprises do to ensure agility and build resiliency? Pandey’s guidance was to take a software-centric approach to the problem. 

Pandey: “We need to start by building out an abstraction layer. Let me abstract away the complexity associated with the transitions or the rapid evolution that’s going to happen in this space. That’s step number one. Once you do that, then there are other aspects that you can plug in. For example: Compliance within an evolving regulatory space. If you can accommodate that ever-evolving factor by abstracting away its complexity, then you set yourself up to adapt.” 

The question of governance and data privacy is unquestionably an extremely challenging problem. In the interview, the discussion highlighted the need for collaboration across teams—data scientists, developers, security, and IT—in order to achieve effective GenAI data governance.

Next steps for enterprises

How should your enterprise get started with a GenAI product or service implementation? In the interview, Xin recommended starting with off-the-shelf solutions to validate your ideas and quickly determine product-market fit. Once you need to deploy to scale, encountering the challenges of cost, privacy, and data security, then you can start to build a custom solution that is cost-effective and tailored to your needs.

Pandey closed the interview with a call to organizations to lean into their strengths:

Pandey: “Focus on what you would consider the top of your stack. But things like security, scale, and responsible AI can be farmed out to vendors that can actually provide them at scale to you.”

Outshift by Cisco is becoming a leader in global discussions around responsible AI and the successful use of GenAI in the enterprise space. Learn more about Outshift’s position on trustworthy and responsible AI and the centrality of customer trust to the use of AI in the enterprise.