Smart companies will experiment with small generative AI applications to gain the skills and confidence to try bolder projects.
Just like big data back in 2013, we’re in the “everyone’s doing it, no one knows why” phase of generative AI (genAI). A recent McKinsey survey found that 65% of enterprises are “regularly using genAI.” Promising! In Elastic’s recent earnings call, the company noted that over 1,000 customers are paying to build genAI applications. Wow! Each of the big cloud companies, as well as Oracle, has talked up how genAI is driving cloud spend. Amazing!
Maybe. Maybe not.
Peel back the headlines and we’re still seeing genAI as aspirational, not necessarily transformational for most companies. For example, while touting all its customers building genAI applications, Elastic CEO Ash Kulkarni also said, “We are not modeling significant revenue contribution from genAI this year.” In other words, 1,000 companies are not paying very much, largely because they’re not doing very much. That’s not a slight on Elastic; rather, it’s the reality of where we’re at with genAI today. The clouds are mostly fattening their AI revenues through training models, rather than enterprises using those models to draw inferences from that data in applications.
In other words, if you have yet to transform your business with AI, you’re not alone. You have time.
Still early for genAI
I wrote about this recently and won’t belabor the same points (i.e., rather than big genAI projects, the enterprises finding real success tend to be doing better search through retrieval-augmented generation (RAG). According to the McKinsey survey, enterprises have yet to figure out where exactly to use genAI. Only two use cases (“content support for marketing strategy” and “personalized marketing”) were cited by at least 15% of respondents. There are some IT help desk chatbots (7% of respondents) and design development (10%), but for the most part, everything else is largely a rounding error.
Enterprises are kicking the tires, to put it nicely.
Other data from the survey creates more questions than it answers. For example, the report said, “Respondents most commonly report meaningful revenue increases (of more than 5%) in supply chain and inventory management,” yet just 6% of enterprises in that market report regularly using genAI. If it’s working so well to drive revenue, wouldn’t more companies be doing it?
Again, this isn’t to suggest that genAI, and AI more broadly, won’t have a significant impact. Rather, it’s indicative that we are early in the adoption cycle.
Get started. Break things
I suspect one key reason that sales and marketing is the biggest area for genAI within enterprises, according to the McKinsey survey, is the perception that these are areas a company can “get wrong.” I don’t mean that these areas are unimportant. I just mean you’d probably rather have an LLM hallucinate on an early version of marketing copy than your income statement. According to McKinsey, the top genAI performers tend to be those that “have experienced every negative consequence from genAI we asked about, from cybersecurity and personal privacy to explainability and IP infringement.” They’ve been burned by genAI and learned from the experience. It’s best to learn the ropes with activities that are behind the firewall and relatively low risk.
These same high performers run more genAI workloads than their peers (they use genAI in three functions on average; less-experienced companies average two) because they’ve figured out how to manage the risks of rough edges. They also have more advanced risk-mitigation strategies, says McKinsey, and then become “more than three times as likely as others to be using genAI in [more advanced] activities ranging from processing of accounting documents and risk assessment to R&D testing and pricing and promotions.” They’ve also run into problems with data: 70% of high performers cite problems with data, including figuring out data governance processes or lacking sufficient training data.
You don’t run into these problems (and learn from them), if you aren’t willing to experiment and risk breaking things.
Returning to Elastic’s 1,000 customers paying to build genAI applications, this is great news for Elastic, as well as the industry, regardless of near-term financial impact. As the company’s executives said, genAI will be “a significant growth driver for us in the long term,” even though “customers are still in the early stages of the adoption cycle.” The way all enterprises are going to go from early tire-kicking to enterprise transformation is to start small, break a few things, and gain the experience and confidence to go bigger with genAI.
More by Matt Asay:
