How Accenture is minding the AI enterprise adoption gap

As the market braces for a possible bursting of the AI bubble, research from Accenture suggests that enterprise investment in AI is unlikely to be swayed by this with 46% of leaders saying they would continue to increase AI investment even in the event of a market correction. Many of the AI technology vendors are focusing on their partnerships with consultancies and SIs to help their enterprise target markets cross the chasm of enterprise adoption for AI. 

 Fer Lucini, Accenture’s Global Machine Learning Lead's view is that we are in a period of quiet enterprise work. He argues: 

Every enterprise customer is doing things with AI, however, the majority of the technology available is consumer-oriented and enterprises think about things differently because they have to make choices about what is valuable, how to do it safely and put it into the daily working lives of employees. It has never been easier to build a demo but it is really difficult to make this new technology work inside a company.

To the question of whether the problem lies with the shape of enterprise data, he says: 

The data itself doesn’t change very quickly and so choices must be made about which Gen AI models work the way you need, and which data will suit your purposes. The amount of data is not always the problem, but it depends how complicated the data is. The net value of the data is realised by applying technology such as Gen AI but only if you have clarity about the problems you want to solve because you cannot move the data very quickly.

People are making really sensible decisions at the moment by taking the time to play around with different use cases. Take ChatGPT – everyone can use it and imagine its potential, which is great. The downside is that companies then end up with a very long list of things to do with the technology. We are currently in the middle of a quiet enterprise period as choices are being made while a little data work and a little AI work is undertaken. While that is happening organisations should fire the imagination of the entire workforce using Copilot so that when the business is ready, the workforce is converted to the approach.

At the moment we need to resolve safety and governance issues with a technology that is going to change every three months. We are learning to learn with a new technology that has no manual. At the moment what we do with it is propagated by the availability of use cases. We need to be patient with the majority of humans that want things to proceed in a structured way. In the enterprise space people are quietly working on these things.

We are not in the realm of AI replacing humans

Another issue that the Accenture research highlights is the gap in levels of optimism about AI between leaders and the workforce. Lucini’s view on this is:

If you put yourself in the place of a CFO, it is reasonably easy to spark their imagination about how this technology could enable innovation. If you put the technology in front of an employee, the view is different because technology is seen as a way to get specific things done, and if you do not make clear what specific things this technology will do, it is difficult for employees to get a clear impression of what is about to happen.

Also, there is a lot of over-stating of what the technology is capable of. For example, while Model Evaluation and Threat Research (METR) shows that that AI can double the length of the tasks it can do every six months. We are not talking about tasks that a human would be doing. We are breaking human tasks into tiny pieces, feeding them to a model and repeating this exercise eight times typically before the technology can complete these tiny tasks with a 50% success rate.

When irresponsible headlines get down to the workforce you generate fear and anxiety, whereas what is really happening is that workers retain their agency to solve their specific problems and we need them to translate the technology into the specifics of their role so that it can assist them with the daily tasks they do.

So, are SI partners the missing link that the tech vendors are hoping they are for AI market development? Lucini suggests: 

At the end of the day, we need to get the workforce engaged with the technology as a flexible tool to help with problems. As partners we are training people to see what it can do in their environment, tailoring training to specific jobs and people. To do this we ask people to prompt and a minority will fly with the concept. The rest of us need a bit of help, to not be afraid of prompting.

You can start by asking the model itself to create a prompt based on what you want it to do. In this way, the human has driven the exercise and the technology is clearly in an assistance role. Learning this way reinforces the idea that the human is in control and illustrates the limits of the technology. This type of learning process breaks down fear of the technology and reveals it for what it is a bunch of tools that can learn. If each worker is asked to think of three shortcuts that would like to implement using the technology to make their job easier, the workforce is engaged while the company is working on business implementations.

My take

This is a multi-headed problem, exacerbated by the fact that we have gone so fast on the consumer side while in the enterprise we are still trying to understand, apply and learn about the technology as it continues to evolve.