Without the proper data management tools and processes in place, enterprises struggle to figure out which genAI use cases to target because they don’t know if they have the right data – and the right data governance and quality management controls – to enable use cases they are considering.
If you’re struggling to take advantage of generative in your business, your core underlying problem is probably a lack of sufficient data management tools and processes.
That’s the short version of the main roadblock that enterprises encounter along their genAI journeys, based on my experience working with businesses to define and execute data management strategies to power genAI. Keep reading for the longer version as I unpack why implementing enterprise genAI can be so challenging and what businesses must do if they want to take full advantage of genAI.
How enterprises aim to use genAI
It’s no secret that enterprises of all types are eager to take advantage of generative AI technology. As of late 2023, 66 percent of organizations surveyed by IDC reported that they were exploring genAI. Likewise, Gartner found that, in the same time period, 55 percent of businesses were piloting genAI projects or (in limited cases) already had them in production.
Yet, there’s a big difference between experimenting with genAI or implementing it in a handful of areas and taking full advantage of it across the business. It’s easy to talk about genAI or identify potential use cases; putting them into practice is the real challenge.
See also: Bridging the Gap: Scaling GenAI for Real Business Growth & Impact
Roadblocks on the enterprise generative AI journey
There are many reasons why fully embracing generative in the enterprise can prove so difficult. Here’s a look at the top issues I’ve witnessed in my work helping businesses establish the data foundation for successful genAI initiatives.
Not knowing which use cases to target
Generative AI can potentially do a bunch of different things – from helping employees draft emails or reports to powering chatbots that engage with customers, to documenting products, and much more.
But whether a given business will actually benefit from a particular genAI use case – and whether it can successfully implement the use case in the first place – depends on factors that some enterprises struggle to decipher. They may not know whether they have the right data to enable a specific use case, for example. They may also not know which use cases will deliver the greatest ROI.
As a result, enterprises end up struggling to decide which genAI use cases to focus on, stunting their genAI implementation plans.
Uncertainty about which genAI models to use
There are many genAI models available today. Some are proprietary, and some are open source. Some are easier to customize than others, and some excel at certain tasks where others come up short. Some require you to share sensitive data with third parties, while others let you keep all of your data in-house.
Given all of these choices, it can be challenging to decide which model or models to use and how to use them. For example, you might find yourself weighing the pros and cons of taking a pre-trained open-source model and retraining it on your data. This approach gives you more control over how the model works, a greater ability to customize it, and the peace of mind of not having to share data with a third-party vendor. But this strategy also requires more effort than using a model that is available as a service. In addition, if you use an open-source model, you must ensure that you have sufficient data (and data of sufficient quality) to retrain the model.
On the other hand, perhaps you will see better ROI if you use a proprietary model. These are typically easier to work with because they don’t require you to operate the model on your own. The drawback is that proprietary models are harder to customize, and using them usually requires sending your business’s proprietary data to a third-party vendor – so there may be major security and data privacy issues to weigh.
The bottom line here is that evaluating different model options requires a deep understanding of genAI from both technical and business angles, and it can be very tough to bring together the right types of expertise to provide accurate insights.
Difficulty predicting genAI ROI
The cost of investing in genAI technology can vary depending on how you go about it. For instance, developing and training your own model typically requires a higher upfront investment than using a third-party model. Still, no matter what your approach is, investing in genAI is not cheap.
At the same time, predicting how much money genAI investments will save by increasing revenue and/or productivity can be challenging because the technology remains so new, and it’s hard to figure out the potential ROI. For instance, an enterprise might not know whether the cost of collecting, transforming, and managing all of the data it will need to power custom genAI solutions will be justified by the savings that those solutions produce.
Given this uncertainty, some enterprises hesitate to embrace genAI fully because they’re simply not sure if the investment will end up being worth it.
Focusing on tools but not processes
There is an increasingly large ecosystem of tools available to help companies manage genAI, as well as the data that genAI models rely on. However, tools alone don’t move the needle when it comes to putting genAI strategies into practice. Businesses also need processes to accompany the tools.
For instance, imagine that you’re an enterprise trying to figure out whether you have the right types of data to develop a genAI chatbot. You could buy a data discovery tool that creates an inventory of all of your data. That’s one step toward building your chatbot, but it won’t get you all of the way there. You also need processes that allow you to collect the data, manage its quality, perform any relevant transformations, and so on before you can feed the data into the genAI model that powers your chatbot.
My point here is that it’s easy to focus on tools, but tools alone are not a complete solution for implementing genAI.
Solving the genAI data challenge
If you read between the lines, you’ll notice that there is a common theme that runs throughout all of the enterprise genAI challenges I just described: Data management. Without the proper data management tools and processes in place, enterprises struggle to figure out which genAI use cases to target because they don’t know if they have the right data – and the right data governance and quality management controls – to enable use cases they are considering.
They also can’t decide which models to use or how to retrain or customize them. Nor can they accurately predict the cost of investing in genAI because they don’t know how expensive it will be to implement the requisite data management processes.
This is why implementing an effective data platform is the single most important step that enterprises can take to enable genAI. A data platform means a holistic set of tools that allow businesses to discover, monitor, govern, secure, and transform all of the data they own. Building on that foundation, they can enable DataOps and MLOps processes that allow them to collect whichever data they require to support a given genAI use case, feed the data into their models, and deploy the models into production.
To be sure, there are other challenges to overcome as well. For example, enterprises need to acquire personnel with the requisite talent to build and train genAI services and/or integrate with third-party genAI vendors to deploy the solutions they require. They must also invest in infrastructure to power genAI model training and inference.
But all of the above hinges on having the right data management solutions in place. That’s where enterprises too often stumble – and it’s where they should be focused if they want to stop simply talking about genAI and start putting it into actual use.
