Years from now, when we reflect on the proliferation of generative AI (GenAI), 2024 will be seen as a watershed moment – a period of widespread experimentation, optimism, and growth, when business leaders once hesitant to dip their toes into untested waters of innovation, dove in headfirst. In McKinsey’s Global Survey on AI conducted in mid-2024, 75% predicted that GenAI will lead to significant or disruptive change in their industries in the years ahead.
While much has been learned about the advantages and limitations of GenAI, it’s important to remember we’re still very much in a stage of evolution. Pilot programs can be ramped-up quickly and are relatively inexpensive to build, but what happens when those programs move into production under the purview of the CIO’s office? How will function-specific use cases perform in less controlled environments, and how can teams avoid losing momentum before their program has even had the chance to show results?
Common Challenges Moving From Pilot to Production
Given the enormous potential of GenAI to improve efficiency, reduce costs, and enhance decision-making, the C-Suite’s mandate to functional business leaders has been clear – go forth, and tinker. Business leaders got to work, toying around with GenAI functionality and creating their own pilot programs. Marketing teams used GenAI to create highly personalized customer experiences and automate repetitive tasks. In customer service, GenAI helped power intelligent chatbots to resolve issues in real-time, and R&D teams were able to analyze huge amounts of data to spot new trends.
Yet, there is still a lot of disconnect between all this potential and its ultimate execution.
Once a pilot program moves into the orbit of the CIO’s office, data is scrutinized much closer. By now, we’re familiar with some of the common issues with GenAI like model bias and hallucinations, and on a larger scale those issues become big problems. A CIO is responsible for data privacy and data governance across an entire organization, whereas business leaders are using data that might only pertain to their specific area of focus.
3 Key Things to Think About Before Scaling
Make no mistake, business leaders have made significant progress in building GenAI use cases with impressive results for their specific function, but scaling for long-term impact is quite different. Here are three considerations before embarking on this journey:
1. Include the IT & Information Security Teams Early (and Often)
It’s common for functional business leaders to develop blinders in their day-to-day work and underestimate what’s required to expand their pilot program to the broader organization. But once that pilot moves into production, business leaders need the support of the IT and information security team to think through all the different things that might go wrong.
That’s why it’s a good idea to involve the IT and information security teams from the beginning to help stress test the pilot and go over potential concerns. Doing so will also help foster cross-functional collaboration, which is critical for bringing in outside perspectives and challenging the confirmation bias that can occur within individual functions.
2. Use Real Data Whenever Possible
As mentioned earlier, data-driven issues are among the biggest roadblocks in scaling GenAI. That’s because pilot programs often rely on synthetic data that can lead to mismatched expectations between business leaders, IT teams, and ultimately the CIO. Synthetic data is artificially-generated data created to mimic real-world data, essentially acting as a stand-in for actual data, but without any sensitive personal information.
Functional leaders won’t always have access to real data, so a few good tips for troubleshooting the problem would be: (1) avoid pilot programs that might require additional regulatory scrutiny down the road; (2) put guidelines in place to prevent bad data from corrupting/skewing pilot results; and (3) invest in solutions using the company’s existing technology stack to increase the likelihood of future alignment.
3. Set Realistic Expectations
When GenAI first gained public prominence after the launch of ChatGPT in late 2022, expectations were sky-high for the technology to revolutionize industries overnight. That hype (for better or worse) has largely endured, and teams are still under enormous pressure to show immediate results if their GenAI investments hope to receive further funding.
The reality is that while GenAI will be transformative, companies need to give the technology time (and support) to start transforming. GenAI isn’t plug-and-play, nor is its true value only limited to clever chatbots or creative imagery. Companies that can successfully scale GenAI programs will be the ones who first take the time to build a culture of innovation that prioritizes long-term impact over short-term results.
We’re All in This Together
Despite how much we’ve read about GenAI recently, it’s still a very nascent technology, and companies should be wary of any vendor that claims to have figured it all out. That sort of hubris clouds judgment, accelerates half-baked concepts, and leads to infrastructure problems that can bankrupt businesses. Instead, as we head into another year of GenAI excitement, let’s also take the time to engage in meaningful discussions about how to scale this powerful technology responsibly. By bringing in the IT team early in the process, relying on real-world data, and maintaining reasonable ROI expectations, companies can help ensure their GenAI strategies are not only scalable, but also sustainable.