Before rushing to invest in generative AI (GenAI), organizations must pause and take a step back. GenAI is powerful and has shown potential to revolutionize multiple industries – but it’s not a silver bullet.
Now that we’ve finally gotten past the hype phase, it’s time to look at the realities of what GenAI can and can’t do and how it can affect your organization.
Along the way, I hope to dispel a few C-suite misunderstandings – but not to put you off GenAI. There’s much more to be gained from using GenAI thoughtfully rather than mindlessly.
Truth 1: Large language models alone do not solve business problems.
GenAI and LLMs are features that augment your existing processes. To gain real value, you’ll need to integrate them with other technologies.
With or without LLMs, it’s still about solving real-world problems for specific industries and people in a way that delivers real-world value. That requires end-to-end systems and technologies seamlessly moving from data management and discovery to model development and deployment.
Truth 2: LLMs cannot be directly used for analytics and decision intelligence.
LLMs are a conversational interface, nothing more. Analytics still drive insights, not LLMs or GenAI.
Prompt engineering is limited and can only get you so far without more enterprise capabilities. All it does is create natural language text to instruct AI what to do.
Truth 3: Retrieval augmented generation (RAG) architectures are slow and difficult to operationalize.
From its beginnings, GenAI had serious problems with the truth. It made up references and facts; nothing short of your in-depth research could counter its made-up “truths.”
RAG architecture improves factual accuracy, helping ensure that responses are grounded in facts. That makes it a giant leap forward for natural language processing.
But RAG architecture isn’t perfect. Besides being slow and having operational challenges, RAG won’t necessarily find the right data if there’s any ambiguity in what you’re asking. And under the best of circumstances, the LLM output may be superficial or incoherent.
You still need people to bring RAG back to reality.
Truth 4: Three things you can’t short-shrift when starting your GenAI program.
- Enterprise data orchestration: If your organization is like most, you have data siloed within individual departments: marketing, finance, operations, etc. GenAI requires gathering data from every department and looking at all those different departments as one company. Only by working with one data set can you analyze and make decisions that work across the entire organization.
- Prompt governance: GenAI prompts must necessarily be particular and clear. Prompt governance sets rules and restrictions for making a request. For example, you may have to restrict questions about company finances according to a person’s need to know or set other guidelines so inquiries follow established rules.
- LLM governance: Integrating GenAI into existing processes requires having a platform that empowers your people to integrate LLMs safely. Like prompt governance, it’s not always about restricting the use of LLMs to certain people. It’s more like setting guidelines that benefit the entire company.
Truth 5: Organizations with strong data quality and data management rigor will be the most successful with GenAI.
GenAI is new and exciting and can quickly take over every business conversation about the future.
Here’s the reality, though: GenAI runs on data. And just like the fuel you put in your car, quality data will take you where you want to go and safely guide your strategy for the future.
Faulty data, however, will impede performance at every step.
The time to invest in data quality is before you majorly invest in GenAI.
Truth 6: Prioritizing Trustworthy AI from day one is important
GenAI creates not only new opportunities but also new risks – compliance, reputation and IP-related. Define and publicize a vision for responsible AI with clear principles and policies across fairness, transparency, safety, privacy, sustainability and regulatory compliance. Start with these four items:
- Data integrity: Use robust mechanisms to ensure data integrity so you can trust that the information driving your decisions is precise and dependable.
- Ethical practices: Foster trust in AI, creating a foundation for ensuring that your AI-driven decisions are fair and equitable and do not unintentionally discriminate against any group.
- AI governance: Create with XOps frameworks like DataOps, ModelOps, and DecisionOps to align processes, technology and people.
- Continuous monitoring and auditing: Set up constant monitoring and auditing processes to track AI performance and compliance over time.
Truth 7: Cost controls and oversight are important.
GenAI may play an important role in your organization’s future, but there’s no reason to mortgage that future by writing a blank check for its implementation.
GenAI requires new investments in cloud expenses and operations, as well as costs for data security, talent, infrastructure, hardware and dozens of other categories you may have never considered.
But GenAI is a long-term investment, not a quest for single-shot success stories.
What’s important today is building a technology architecture that can be scaled and standardized for future success.
That requires a fully integrated approach to implementation. SAS works with all types of organizations to create an AI blueprint that begins with learning about your company’s needs, strategies and budget so that every step is outlined and expectations agreed upon before implementation begins.
You’ll save money in the long run and avoid wasting money on systems that don’t fit with a fully integrated vision for the future.
Truth 8: Never underestimate the power of synthetic data.
GenAI may be the future, but it’s essential that it not be compromised by faulty data or invasions of people’s privacy. Language models are very data-hungry, and we will eventually run out of data.
That’s the joy of synthetic data. It allows you to generate data where there isn’t enough, starting from your real data. It also helps you avoid releasing private information and even gives a voice to underrepresented populations to avoid biases.
If you don’t have enough data or cannot use the data you do have because of sensitive information, synthetic data is the way to go.
Truth 9: If you think your competitors have GenAI all figured out, think again.
There certainly are success stories, but individual successes don’t always equal competitive advantage, much less long-term advantage.
The early failures we’ll see in 2025 will result from jumping in without long-term strategy, but the businesses that learn from these challenges will emerge as leaders in this space.
The GenAI journey is just beginning, and the organizations that take a thoughtful, structured approach will be the ones that transform their industries for the better.
