Generative AI: Unleashing the Power of Innovation or Stumbling at the Gate?

Generative AI is one of the most exciting modern technology capabilities that many of us have witnessed in our careers. We can’t project too far into the future to know what the breadth of the use cases will be that apply to our different segments of the industry — but we do know enough to realize that it’s here, it will be impactful, and our customers and other stakeholders will have high expectations of the value it will deliver. It might be too early to determine whether your company is ready to adopt generative AI, but it’s absolutely the time to ask, “Will we be ready?” And if you’re not, what do you plan to do about it? 

This statement is generally accepted and obvious but what does it really mean? There are multiple dimensions to dependency on trusted data.

Dimension number one is internal governance. While we are all trying to figure out how to use these new capabilities, rest assured there are others who are trying to figure out the right guardrails to put in place to ensure responsible use. An acceptable use policy is probably being re-written; there are debates going on about software development; and there is already tension between risk and benefit.

One of the answers is the ability to demonstrate mastery of the data. Where did it come from? Is it allowed to be exposed to a public large language model (LLM)? Is it allowed to be exposed to an internal-only LLM? If so, how will you manage granular access to the data, which, instead of being accessed via an application or a dashboard, is being accessed via a natural language prompt?

Dimension number two is trust in the output. If you expect to “ask your data a question” and get an answer that you can act on, how confident are you that the response is trustworthy enough? Who in your company is asked to figure this out? Even if there isn’t an official charter, it’s likely that the eyes in the room will focus on you, as the CIO or CDO, to solve this problem and effectively start to enable productive and responsible use of these new capabilities.

Think generative AI is optional? Industry experts say “no.” Sure there will be advanced, highly differentiating use cases that the best companies will deploy, but there will be myriad basic use cases that are available to everyone. Not accessing these use cases will cause companies to fall behind. The differentiating use cases will require putting enterprise data into the AI model. And that means your data must be ready. 

The visual analytics fiasco from 10 years ago after business users saw their first demo of Tableau is still etched in the minds of many IT folks. Everyone was amazed at the ability to easily visualize complex datasets, filter, drill down and zoom out to make raw data actionable and consumable by management. It was nirvana — the answer to all your problems!

Then the user came home, excited to put Tableau on top of their data to get answers to their every question, only to be woefully disappointed in the outcomes because their data was bad. That’s because the company didn’t take the time to define business terms, build calculations into the data model and solve data access rights challenges.

Or simply put — they didn’t have a data management strategy that was adequate to fulfill their visual analytics ambitions. The gap between expectations set by the vendor and the reality that could be achieved by a company that had deficient — or nonexistent — data management practices was vast. And guess who more than likely let this happen on their watch? Either the CIO or CDO.

And it will be on the CIO or CDO again when it comes to generative AI, only ten times worse because the expectations are so incredibly high. My advice? Start now if you haven’t already and be ready.

Which vendors? The vendors who build generative AI into their application that you already use — Microsoft 365, Salesforce, ServiceNow, etc. — will do a decent job of making sure that the definitions, access rights, etc. that support the app itself will also support the gen AI solution. This will be beneficial, of course, but the differentiating use cases will require YOUR data from multiple sources to be made available to a LLM from Microsoft, Google or someone else. These vendor solutions recognize the need to include “your data.”

In fact, one of these companies has a tiny box at the very bottom of their reference architecture diagram that says “your data” with a simple line into the complicated set of components that make up their generative AI solution. I would argue that “your data” isn’t a tiny piece; it’s the piece that matters. The piece that will bring data and AI to life for your company. Who is going to manage that? Not your vendor.

Even though it’s unclear now what the differentiating use cases will be, you need to be prepared. And that means focusing on data management. 

The success of AI is dependent on the availability of holistic, trusted, governed data. Here are some key data management capabilities you need to implement now so you’re not caught off-guard and left in the dust:

1. As enterprises start adopting generative AI and large language models (LLMs), their unique value and differentiation will come from fine-tuning, operationalizing and customizing LLMs with their data. That’s why you need a solution that finds, prepares, cleanses and governs data.
2. Being able to drive scalable and responsible adoption of generative AI will depend on how effectively you monitor and manage data quality, privacy and compliance. This means having data governance frameworks and tools in place to ensure you are fueling LLMs with trusted data.
3. AI also needs intelligent data management to quickly find all the features for the model. This includes data lineage to understand where data is coming from and going to; and data privacy to ensure sensitive data is used appropriately for training and model development purpose.

As a CIO or CDO, it’s your responsibility to drive awareness and investment toward data management to take your organization into new, unchartered directions. I urge you to act now, so when the time is right, you can unleash the power of LLMs on your data to create these differentiating use cases. Companies that act decisively are poised to succeed. Those who do not will fail. Which side do you want to be on?

In a future blog, I’ll share what we’ve accomplished at Informatica to prepare for the future. In the meantime, take a listen to how AI has impacted the data landscape and why I believe generative AI will positively impact the workforce.