Artificial Intelligence (AI) makes big promises, but does it deliver? According to a recent study conducted by the BCG GAMMA, the BCG Henderson Institute, and the MIT Sloan Management Review, just 10% of companies report a significant financial impact of implementing AI. Despite this low figure, adoption of AI in business continues to increase. AI is perceived as a strategic advantage and a way to manage risk in increasingly complicated economic times. So why have so many struggled to glean financial benefits from AI, and what are those 10% doing right?
What do we mean by AI?
AI can now be used in a wide variety of processes — from low-level automation, augmenting human experts and producing novel insights. For example, Alibaba uses AI for everything from predicting-commerce purchasing decisions, reducing traffic jams in smart cities and maximising agricultural efficiency by monitoring crops.
According to a survey by Fountech Solutions, 36% of firms were concerned about their level of AI progress compared to competitors. Media hype and the huge number of AI-powered tools that make big claims contribute to the pressure to implement AI quickly. However, with so few companies achieving a worthwhile return on investment, is it time to stop the AI hype?
The most obvious challenges to AI adoption are the most fundamental. Accessing the right quantity and quality of data has always been an issue, and the skills gap in the data science industry continues — with demand far outstripping supply. However, according to the BCG GAMMA, the BCG Henderson Institute, and the MIT Sloan Management report, even when companies got the basics right, just 20% of them saw significant financial benefits (calculated according to company revenues). A somewhat disappointing figure given the cost and time involved in AI implementation.
The key to success: organisational learning
According to the report mentioned above, ‘Learning with AI’ was the most important factor influencing the financial impact of AI — the likelihood of achieving significant benefits increased to 73% in these businesses. This organisational learning had three key elements including:
Continuous learning: businesses that facilitated multiple methods of mutual learning (AI teaching humans, humans teaching AI, and independent AI learning) were five times more likely to earn significant financial benefits than those that used a single method. An example of this mutual learning can be seen from how AI is used by traders in financial institutions. In this environment, the AI first observes trader behaviour to learn which data to prioritise, then the AI autonomously creates a set of rules based on what it has learnt, and finally the trader learns from the AI-derived insights.
Context dependent human-AI interaction: the weighting of human-AI interaction varies from the AI autonomously making decisions with minimal human supervision to human employees making the decisions and only using AI to gain further feedback on likely outcomes. Using a variety of different methods is more successful because the appropriateness of each method depends upon the wider context. Generally, in simple environments where the AI has all the appropriate data, then the AI can be given more control. In contrast, in more complex environments that require human creativity and interpretation to understand the likely outcome, AI should be relegated to a lesser role.
Structural and cultural change: one of the biggest challenges to implementing AI is underestimating the difficulty of the transition. Businesses that extensively changed many processes across their organisation were five times more likely to achieve significant financial benefits than those who only made small changes. Making sure employees are practically able to use AI technology and understand both the advantages and limitations of such tools is vital.
It’s clear that to reap the financial rewards of AI, companies must commit to an ongoing culture of organisational learning. If this culture represents a radical change from the status quo, then AI adoption may be challenging in the short term. However, discomfort is a small price to pay in order to secure the competitive advantages of AI and thrive in our increasingly digital world.