NVIDIA Is Increasingly the Secret Sauce in AI Deployments, But You Still Need Experience

I’ve been through a number of briefings from different vendors from IBM to HP, and there is one constant: they are all leaning heavily on NVIDIA for their AI services strategy. That may be a best practice, but NVIDIA doesn’t do deployments. As companies like Accenture spin up 30K people on its effort, there will be a big difference in terms of quality when it comes to deployment teams.

In addition, service entities tend to tell their customers what they want to hear, not necessarily what they need to know, because the service folks don’t want to upset a new or existing customer. But what you know will come back to bite you, so it is important you not only validate the company doing the work and its relationship with NVIDIA, but the experience of the team assigned to your project. If any of those factors aren’t adequate, your deployment is likely to fail. Given failure rates currently exceed 80%, there are a lot of teams out there that clearly don’t know what they are doing…yet.

The Necessity of Assuring the Team

Years ago, I drove a move in my firm from Lotus Notes to Microsoft Exchange. It wasn’t a popular move, but I’d seen the writing on the wall and didn’t want to be stuck with a soon-to-be obsolete messaging and collaboration platform. I wasn’t the one who selected the deployment team that was expert in Lotus Notes and knew our company, but I had never deployed Exchange before. It was a nightmare because we ended up paying for that education. Ironically, we were acquired, and the acquiring company took us back to Notes until it became obsolete and then had their own wonderful migration experience again.

(PeopleImages.com – Yuri A/Shutterstock)

One thing you learn about email that will likely also apply to AI is that you can piss off a lot of people if the deployment goes wrong because everyone uses that tool, including the top executives and often the board. AI is evolving to not only be a productivity enhancer but also the digital face of the company. At some point, likely soon, your phone interactions with customers will be mostly through AI and your employees will be using AI tools broadly in their jobs.

So, if the AI is non-viable, hallucinating or has been corrupted, the impact will be adverse on your image, brand, employees and especially your executives who won’t find any of this funny. As a result, you not only need to assure the technology provider is working with NVIDIA (they are currently the most experienced, though AMD is coming up quickly, and Intel is ramping up its own programs), you also need to ensure not only that the company providing the implementation support knows what they are doing, but the team you’ve been assigned is also experienced in both the technology and the kind of implementation you need done.

Assuring Your Vision

Before selecting a technology or a deployment service provider, one of the first things you should do is fully understand what you want to do and what the various AI solutions can do. Too often we pick the technology and the service provider and then flesh out the project, especially with new technologies like AI. That is exactly backwards, because the knowledge of what you need done will directly inform on the technology and implementation partner you should use to get it done. If you haven’t fleshed out the project, how can you intelligently select a vendor or a deployment team?

(amgun/Shutterstock)

Put a different way, in the health care realm, you don’t select a medical expert and then get the diagnosis. The diagnosis is what points you to the expert you need. Same thing holds true when it comes to AI deployments. Because AI is so new, none of the firms have expertise in all industries or all companies. So, you need to fully understand and be able to articulate what needs to be done in order to properly interview the candidates bidding on the project. And it should go without saying that it is virtually impossible to set a budget before you have fleshed out what you want done.

Wrapping Up: Be Methodical With AI

AI is huge and getting bigger, but we are still early into deployments, and most are failing still. To assure the success of your project, you need to understand what you need done, what tools are available and designed to do what you need done, a vendor to supply those tools, and an implementation partner or service to deploy them. Each of these steps needs to be assured before any work is done. Otherwise, the project is likely to be doomed.

With projects of this high level of visibility it is better to get them done right than it is to get them done quickly. Yet too often we prioritize speed over quality. Setting a solid foundation is good advice for both buildings and projects like this.