Why the Gartner Hype Cycle is a Bad Way to Look at GenAI

This popular approach can lead to inaction at exactly the wrong time

Imagine the following meeting, which may or may not have happened, at the Macy’s board of directors in September 2001:

Gartner senior analyst: I just want to say our Hype Cycle framework created in 1995 has been proven right once again! E-commerce is overblown. Amazon’s market capitalization once a ridiculous $29 billion is now still overpriced at $2.3 billion. Amazon has yet to make a profit. Macy’s solid market capitalization of $3.2 billion and measured approach to exploring e-commerce, is the best course of action and exactly the approach our Hype Cycle analysis recommends…Today, Macy’s market capitalization is $5.2B; compared to Amazon at $1.9T.

When executives face a wave of powerful new technology many people have found refuge in the Gartner Hype Cycle which posits five stages for any new technology: “Innovation trigger, peak of inflated expectations, trough of disillusionment, slope of enlightenment and plateau of productivity — as the five stages” to guide executive’s thinking. See the illustration below from Gartner. The Hype Cycle is exactly the wrong way to think about GenAI today because it gives busy executives a reason to avoid analyzing and deploying this new technology, when action, experimentation, and learning are needed. The very language of the framework invites delay; who wants to chase a wave of inflated expectations?

With GenAI we have a massive new technology that has had the fastest consumer adoption of any technology in history, represents a new form of interaction with computers where we can “talk” with the machine in plain language, and delivers significant new capabilities almost weekly as the titans of technology battle for supremacy. Gargantuan legacy firms like JP Morgan, McDonalds, WalMart, Epic, Saudi Aramco, and many many others have already deployed production systems with GenAI less than two years after it became popular, and they have done it, not in some backwater task or division, but in the very heart of critical tasks — ordering and franchise operations for McDonalds, shopping and store operations for WalMart, doctor note capture for Epic — all vital customer or operations functions.

Even in a recent study of IT employees that highlights 98% of GenAI projects were paused, understandably, to make sure they complied with IT policy. 98% of those same firms are also experimenting with use, and 64% have GenAI used in most or all of their departments. (Source: PageDuty Survey.) Those statistics seem to show pretty wide adoption already at least in experimentation mode: production systems still hover around 5% right now according to most surveys.

A defender of the Hype Cycle may say, “but expectations are overblown”. Of course they are. Every powerful technology has sycophants, fake promoters, and ignorant hustlers just as any young rising star sees “friends”, relatives, and hangers on attach themselves to their meteoric rise like so many barnacles attached to a speedboat. Movement attracts fakers — but that does not make the movement fake.

Executives should use a contingent framework that helps them assess how urgent this new technology is for their firm. We wrote about our WINS framework in the September, 2023 Harvard Business Review in which we said if a lot of your cost base is made up of the creation and improvement of Words, Images, Numbers and Sounds (WINS) and that work is already digitized then your firm is “in the crucible” — and you need to learn about it, experiment with it and deploy it now, today, immediately. Tyler Perry the self made billionaire producer stopped an $800 million dollar build out of a new studio in the Atlanta area after he saw the Sora demo. He knows that entertainment is an industry in the crucible and he’s too smart to wait for the Hype Cycle to run its course.

I know it is easy to use the hype-cycle as a crutch to think about things. The poor IT executive is bombarded with every new technology either from the vendors or from the the business leader he or she serves. The Gartner Hype cycle also does map to some things well. I’d argue that crypto, and fin tech were wildly over hyped. At the same time other very important technologies like CRISPER and GenAI are not overhyped and the market should adjust accordingly. A very important thing to remember is that when there is a true, fundamental, technological change as there was with the steam engine, electricity or stored program computers — those who adopted it early became the giants of the next wave. Those who waited for the trough of disillusionment were crushed by the new competitors who did the hard work of figuring out what was relevant to them, experimented and invented a better more productive future on the back of their courage and inventiveness.

In fact, thinking about GenAI thought the narrow lens of technology and as another IT project, is a recipe for failure. GenAI as a technology needs to be married up with organizational capability for prototyping, piloting, creating production systems and driving toward a portfolio of improvement — what we call the four P’s of progress. The process of codifying the massive knowledge and potential locked up in unstructured data, new data types, flexible process and procedures — well beyond robotic process automation is a whole new field of improvement. There are and will be economies of scale and scope to the combined technological and organizational system.

I think the right way to think about GenAI is that it is an organizational learning process — of content, tool, and application. In 1936, Theodore Paul Wright described the learning effect on production costs in aircraft manufacture and proposed a mathematical model. Bruce Henderson the founder of BCG generalized the idea into the experience curve in which he stated: “costs decline by some characteristic amount each time accumulated experience is doubled”. We don’t know yet what the cost decline to volume ratio will be, but given that LLMs progressively structure often unstructured data, and incorporate the feedback of use and interrogation into their learning, my belief is that there will be a predictable learning effect over time as we gain more experience with these powerful tools.

If I’m right, experience with these tools is essential because unit costs decrease with cumulative volume — which can be a hard thing to catch if you are trying to be a fast follower. Put another way, when you hear the words “Gartner Hype Cycle” make sure you are not applying it to a fundamental technology with potential experience curve effects because you may be dooming your organization to staying behind now and even more in the future.

I believe the right path is to take at least the following five steps:

1. Encourage experimentation with public company information with no-code LLM tools like CustomGPT.ai and Chatbase where teams can create a driven chatbot with public company data in a week for under $500, just to learn.

2. Ensure you have clear GenAI corporate use policies that are updated monthly, with examples — e.g. Can I use Perplexity at work, etc.

3. Find, fund and support GenAI rebels and learners.

4. Ensure everyone on your team — including the c-suite — uses an LLM for at least three hours exploring a practical use case in a domain that they are an expert in, to make this technology real for them.

5. Think through the near term operating leverage you can get on WINS tasks, and imagine what type of learning curve you could have to create true competitive advantage over time. Remember Henry Ford shared a lot of the surplus he created with his customers AND his workers. That strategy of sharing the surplus may be relevant here too.