Vendors would have you believe that we are in the midst of an ai revolution, a revolution that is changing the very nature of the way we work. But the truth, according to several recent studies, suggests that there is much more nuance than that.
Enterprises are extremely interested in generative ai as vendors drive potential benefits, but turning that desire for a proof of concept into a working product is proving much more challenging: They face the technical complexity of implementation, whether due to technical debt. from an older technology stack or simply lack people with the right skills.
In fact, a recent Gartner study found that the top two barriers to implementing ai solutions were finding ways to estimate and demonstrate value (49%) and a lack of talent (42%). These two elements could become key obstacles for companies.
Consider that ai-benchmark-survey/”>a LucidWorks studioan enterprise search technology company, found that only 1 in 4 of respondents reported having successfully implemented a generative ai project.
Aamer Baig, senior partner at McKinsey and Company, speaking at the technology-lessons-learned-from-the-early-innings-of-ai“>MIT Sloan CIO Symposium in May, said his company also found in a ai“>recent survey that only 10% of companies are implementing generative ai projects at scale. It also reported that only 15% were seeing any positive impact on profits. That suggests the hype could be well ahead of the reality that most companies are experiencing.
What is the support?
Baig sees complexity as the main factor holding companies back, even with a simple project requiring 20 to 30 technological elements, with the right LLM being just the starting point. They also need things like proper data and security controls, and employees may have to learn new skills like rapid engineering and how to implement IP controls, among other things.
Old technology stacks can also hold companies back, he says. “In our survey, one of the top obstacles cited to achieving generative ai at scale was actually too many technology platforms,” Baig said. “It wasn't the use case, it wasn't the availability of data, it wasn't the path to value; In reality they were technological platforms.”
Mike Mason, ai director at consultancy Thoughtworks, says his company spends a lot of time preparing businesses for ai, and its current technology setup is a big part of that. “So the question is: how much technical debt do you have and how much deficit? And the answer will always be: it depends on the organization, but I think organizations are increasingly feeling the pain of this,” Mason told TechCrunch.
Start with good data.
A big part of that readiness deficit is the data piece: 39% of respondents in Gartner's survey expressed concern about a lack of data as a top barrier to successful ai implementation. “Data is a huge and overwhelming challenge for many, many organizations,” Baig said. He recommends focusing on a limited set of data with an eye toward reuse.
“A simple lesson we've learned is to focus on data that helps you with multiple use cases, and that usually ends up being three or four domains in most companies where you can really start and apply it to your high-priority goals. “Matching business challenges with business values and delivering something that really hits production and scale,” she said.
Mason says a big part of being able to run ai successfully is related to data preparation, but that's just one part. “Organizations are quickly realizing that in most cases they need to do some ai prep work, some platform building, data cleansing and all that kind of stuff,” she said. “But you don't need to take an all-or-nothing approach, you don't need to spend two years before you can get any value.”
When it comes to data, companies must also respect where it comes from and whether they have permission to use it. Akira Bell, CIO of Mathematica, a consulting firm that works with companies and governments to collect and analyze data related to various research initiatives, says his company must tread carefully when it comes to putting that data to work in generative ai.
“As we look at generative ai, there will certainly be possibilities for us and we look at the data ecosystem that we use, but we have to do it cautiously,” Bell told TechCrunch. This is partly because they have a lot of private data with strict data usage agreements, and partly because they sometimes deal with vulnerable populations and need to be aware of that.
“I came to a company that is really serious about being a trusted data steward and, in my role as CIO, I have to be very entrenched in that, both from a cybersecurity perspective and from the way we treat our customers and their data, so I know how important governance is,” he said.
She says right now it's hard not to be excited about the possibilities that generative ai brings; technology could provide significantly better ways for your organization and your customers to understand the data they are collecting. But it's also her job to move cautiously without hindering real progress, a challenging balancing act.
Find the value
Just like when the cloud emerged a decade and a half ago, CIOs are naturally cautious. They see the potential that generative ai brings, but they also need to address core issues like governance and security. They also need to see the actual return on investment, which is sometimes difficult to measure with this technology.
In a January TechCrunch article about ai pricing models, Juniper CIO Sharon Mandell said it was proving challenging to measure the return on investment in generative ai.
“In 2024, we're going to put the genAI hype to the test, because if those tools can produce the kinds of benefits they say, then the return on investment on them is high and can help us eliminate other things,” he said. So she and other CIOs are running pilots, moving cautiously and trying to find ways to measure whether there really is a productivity increase that justifies the cost increase.
Baig says it's important to have a centralized approach to ai across the company and avoid what he calls “too many skunkworks initiatives,” where small groups work independently on a series of projects.
“Enterprise scaffolding is needed to ensure product and platform teams are organized, focused and working at pace. And, of course, you need visibility from senior management,” he stated.
None of this is a guarantee that an ai initiative will be successful or that companies will find all the answers right away. Both Mason and Baig said it's important for teams to avoid trying to do too much, and both emphasize reusing what works. “Reuse directly translates to speed of delivery, keeping your companies happy and making an impact,” Baig said.
Regardless of how companies execute generative ai projects, they should not be paralyzed by governance, security, and technology challenges. But they should not be blinded by the hype either: there will be many obstacles for almost every organization.
The best approach might be to launch something that works and shows value and build from there. And remember, despite the hype, many other companies are struggling too.
