Building a framework for AI implementation
Blog: Capgemini CTO Blog
In the first article in this short series, I discussed how to build a strategy for an AI-driven business. This time, we’re looking at how to move beyond strategy, to a framework for implementation.
The first decision to make is how fundamental you want to be. At its lowest level, artificial intelligence (AI) is highly mathematical, and is the exclusive domain of mathematicians and data scientists. If a company wants to develop a framework that is operating with fundamental algorithms, then the skills it has will need to be commensurate. This in-depth approach may be key to a new area of product design. It could be in drug development, or in manufacturing industry. For example, Tesla and other automotive manufacturers are working on autonomous driving at the fundamental level. They are innovating: creating new techniques that will, over time, be subsumed into software frameworks or into third-party products.
There are many software frameworks for AI that codify all kinds of algorithms and accelerate development (a good example is Google’s Tensorflow). These are great for the data science and engineering teams that need to apply various machine learning algorithms or deep learning neural networks from first principles.
However, there are many domains where the AI development work has already been done. You’ll find there are a significant number of startups that are dealing with the application of AI to specific business problems. For instance, if you need to trap revenue leakage, screen job applications, or improve demand planning, there is a specialist business out there that aim to tick that box for you. The goal of the organization is to take and apply this AI to a modified operating model, so as to achieve the best possible outcome.
There is also an in-between. Automated machine learning, or AutoML, takes away some of the low-level data science activities in a bid to accelerate AI solution development, and to apply it to real-world problems.
So there are options. It therefore makes sense that organizations consider the breadth of AI applicability and construct a framework accordingly. This can extend from base level data science, to AI applications that are applied to a specific set of requirements. The framework should include, and indeed foster, the need to adapt.
The AI market is still in its infancy. There are hundreds of new entrants every year. The framework should therefore include a means to accommodate such fluidity, so as to add new, or replace existing, AI components.
The framework also needs to address the means by which a foundation of trust can be created – in other words, the means by which data is governed and managed across a hybrid environment of centralized, distributed, and decentralized computing.
Managing outcomes
The fundamental rule in managing outcomes is to understand the problem AI aims to solve – and that means getting a clear handle on the business outcomes that could be enabled as a result. However, perhaps more important than this is the need to get to grips with how a current way of working can be reimagined, so it can be driven by AI. This may lead to a fundamentally different product or service.
Outcomes take many forms:
- Strongly personalized recommendations help increase customer retention (e.g., Netflix)
- The perfect recommendation helps promote cross-selling (e.g. Amazon)
- Pattern matching can prevent fraud by identifying anomalies in, for example, procurement activities, resulting in revenue protection
- Computer vision can identify products with visible quality issues, leading to improved product quality
- Analysis of sensor data can predict failure points in machinery, which in turn can be used to optimize maintenance schedules, thereby increasing asset up-time
- A chatbot can effectively answer customer queries in a positive self-service interaction, increasing customer satisfaction, and hence potentially increasing customer retention
To effectively manage outcomes from AI, it is important to understand the broader impact being made on the business. Measurability is key: being able to calibrate the impact on sales (revenue growth), variable costs or working capital elevates the impact of AI to the level of C-Suite metrics. This is where they belong.
In the third and final article in this series, we’ll be bringing everything together, and considering how to implement the strategy and the framework we’ve described so far.
Download Capgemini’s TechnoVision 2020: Future Thinking Simplified, a report that helps business executives anticipate and assess emerging technologies as part of their strategy creation.
Lee Beardmore has spent over two decades advising clients on the best strategies for technology adoption. More recently, he has been leading the push in AI and intelligent automation for Capgemini’s Business Services. Lee is a computer scientist by education, a technologist at heart, and has a wealth of cross-industry experience.
Leave a Comment
You must be logged in to post a comment.
