When will Generative AI Deliver?

In the eighteen months since Generative AI entered public consciousness, both technology firms and industry pundits have talked about little else. It's to be expected really. Intelligent machines have long caught the public imagination, and the late 2022 launch of ChatGPT heralded the first glimpses of a possible future that included them.

The initial hype around AI was based on the idea that it would automate basic knowledge work, allowing businesses to cut their workforce. There was even talk about single person unicorn start-ups. In practice, the technology is nowhere near mature enough for that world. In the meantime, tech firms are spending billions to train new models with little prospect of an immediate return. There is a serious risk that interest in the AI category could be doomed long before the technology is actually ready for mass deployment.

Increasingly, that last point is becoming a critical concern for both vendors and their business customers. Generative AI has its uses, but those aren't pervasive enough or significant enough to justify a business case. You know that a new technology has a problem when Gartner Analysts are "recommending CFOs not to bother calculating the ROI", while The Economist reports it has "almost no economic impact." These are outlets read by CEOs saying that the technology isn't cost effective.

That message is getting through to Wall Street, who have been interrogating tech firms about the scale of their AI investments. The answers from big tech CEOs haven't been to the liking of analysts. Microsoft's share price took a dive following their quarterly earnings call, during which the CFO said that AI investments would deliver a return across a 15 year period. That's not the kind of timeframe that Wall Street is looking for across any investment, let alone one with as uncertain prospects as AI.

Efficiency Gains

Generative AI is not about to disappear. It has millions of users across a wide range of industries. Some of those uses will have a viable use case, particularly for high-volume, low-value content creation workflows in marketing or customer service teams. In these scenarios, it's replacing agency costs with a technology cost which will make CFOs happy.

The biggest benefit is probably realised by the type of middle-ranking executive who spends all day in meetings and needs help to get 'real work' done. It might save an overworked middle manager a couple of hours per week, which in turn will slightly accelerate project delivery, and boost employee satisfaction. However, managers logging off earlier every day doesn't offset the significant cost of the technology.

The trouble is that the most useful benefits of Generative AI don't really impact the bottom line. For many small businesses, automatic meeting transcripts and report summaries are nice to have rather than critical productivity tools. Only large organisations have sufficient numbers of people who need the technology to justify the high price tag of a CoPilot license. Everyone else can get by with the free chatbot bundled with Office 365 or Google Workspace.

The Next Step

Every technology vendor is releasing reports promoting the time savings and efficiency benefits of Generative AI. I read one from HubSpot just last week. However, they all focus on businesses using AI in production, ignoring those who decided not to pursue the technology for cost reasons. The biggest group of businesses is probably those still on the fence about Generative AI. Many of their workers will occasionally use the free version of Microsoft CoPilot or Google Gemini, but not more than that. Any Generative AI projects will still be in testing, due to the questions of cost and accuracy.

It's not surprising, therefore, that technology firms are constantly promoting new use cases for Generative AI technology. OpenAI recently claimed that next year's version of ChatGPT will be capable of independent decision-making. Microsoft have been pushing the ability of CoPilot to generate workflows within Power Automate, never mind that those AI-generated workflows require significant customisation before they can be used in production. There will be a time when both these use cases can become a practical reality in business - but it's not the timeline currently being teased by Sam Altman and Satya Nadella.

Revolution Delayed?

Technology revolutions never progress in the way that analysts expect, and many use cases for new technology turn out to be more inefficient or unaffordable than traditional methods. Witness the rise and fall of blockchain in recent years. Generative AI is different, but not revolutionary. Going further back in time, the rise of personal computers led to many secretarial and PA roles becoming redundant. For now, AI represents the next step in that cycle, with traditional admin work becoming more automated.

However, it won't be touching every worker across every industry in the way imagined eighteen months ago. It's a valuable tool across a range of sectors and job roles, particularly in marketing where it is seeing active adoption. Generative AI has already impacted the marketplace for copywriting and translation. There are efficiency benefits from the technology too. The open question is when those benefits will justify the high cost of training AI models. Until they do, Generative AI will become an upsell feature for the top tier of enterprise software. The likes of Oracle and Microsoft are already taking this approach. For once, technology firms will have to lower their ambitions.