Taking AI from the Lab to Real Life

Operationalizing artificial intelligence and data analytics

Artificial intelligence needs to shift out of the labs and into everyday life in enterprises. Some leading deployments help show us the way.

Enterprises from many industries and around the globe continue to embrace artificial intelligence (AI). In a survey of IT decision-makers that CCS Insight conducted this summer, a remarkable 75% of US and European organizations said they’re now using, testing or researching the deployment of machine learning in their businesses, up from over 70% in 2018.

But there’s a challenge: for the vast majority, AI remains an experimental workbench technology. I speak with many organizations that concentrate their AI projects on proof of concepts or implementing point solutions with a narrow focus. Justifiably so, in my view, because, short of experience and skills, they need to start with small projects to learn and iterate. However, this means that most machine learning models built today fail to make it into production. Fewer than 10% of companies using AI have yet to fully put it into operation within their business processes or have organization-wide strategies. Some estimates suggest that at most, one in five AI solutions become operational.

The path to implementing AI is far from straightforward, so how can enterprises bring the technology out of the lab and into everyday life in their businesses? Let’s take a look at some leading enterprise deployments, which offer some great tips in answering this question.

AES: Improving Maintenance by Transforming Wind-Farm Inspection

Fortune 500 global power company AES provides our first example. AES generates and distributes sustainable energy in 15 countries (see this video). As part of its wind-energy business, it manages eight wind farms, each containing 50 to 300 wind turbines spread across a wide area.

The company conducts annual inspections of this infrastructure using drones. Human inspectors check images for signs of damage or defects such as cracks and lightning strikes, a process that could take up to two weeks. AES now relies on customized and automated computer vision technology to analyse the inspection images — an approach that has improved the speed, accuracy and safety of its maintenance inspections and radically transformed a key business process.

This is an illustrative case because of AES’s focus from the outset on business metrics and its goal of having a continual cycle of operational improvement. This is a big step in the successful implementation of AI, as it helps ground projects in operational success. For AES, the use of computer vision slashed inspection times from two weeks to two days, eliminating 50% of the images gathered from needing human review. This translated to savings of a staggering 57,000 work-hours in the process and over $10 million annually, enabling staff to focus on higher-value projects.

ANZ: Better Customer Insight and Business Decisions with Data Analytics and Governance

Australia and New Zealand Banking Group (ANZ) offers another important example. The multinational banking giant has been using data analytics to extract better personalized customer insight to help its institutional customers make better strategic business decisions. These include on matters of liquidity, risk, and cash management, and on helping with decisions such as store locations, inventory and strategy.

For a company of its size in the industry it operates, ANZ has significant regulatory obligations, with its Institutional Banking division operating in 34 markets globally. Additionally, trust in banking companies has been low since the financial crisis of 2008. As a means to address this, security, governance and compliance requirements are now top considerations, especially when it comes to data analytics.

ANZ has managed to overcome some of the most significant hurdles in these areas by deploying a managed public cloud service, which, for data analytics specifically, serves as the repository for its data pipelines, ensuring the security of its customer data and reliability of its applications. As part of this approach, ANZ has been using analytics and data science technology to analyse aggregated, anonymized data sets, including credit card data, to gain these insights. This process used to take days, but now takes just seconds according to Google.

ANZ shows that companies can get ahead of the game if they look to embed good data management and governance into their data analytics and AI programmes from the get-go. One of the most important challenges emerging in data and AI is in its governance. How can enterprises address potential bias in their data sets or rising concerns about customer privacy? How can they ensure systems are transparent, explainable, compliant and, ultimately, instil trust with users? Our surveys show that trust in AI is the biggest barrier to the implementation of the technology and that data security, privacy and transparency have become the most important requirements for investment in machine learning tools.

Enterprises cannot begin to tackle these challenges without a good data governance foundation because AI is only as effective as the data it is fed. This is also essential to reducing problems later on, and engendering trust in AI which enables faster deployments, wider adoption and more-responsible innovation in your business.

UPS: Forecasting at Scale with Data Analytics and AI

A final case study comes from UPS. The global transportation company is applying data analytics and machine learning models to an astonishing 1 billon data points per day to help it power the most precise and comprehensive forecasting capability in its history.

The data collected ranges from package weight, shape and size, to facility capacity and delivery routing and scheduling throughout its network. The insights extracted from the data help UPS inform everything from how to load delivery vehicles more efficiently, to making more targeted adjustments to operations and minimizing forecasting uncertainty during holiday periods. Through more intelligent routing and delivery processes, the technologies have also helped the company cut its fuel consumption by up to 10 million gallons a year.

UPS’ strategy reveals a very high level of operational dependency on data analytics and AI compared with many organizations. At the heart of this is a strategic collaboration with a major cloud provider; in this case, Google Cloud. This is another important aspect of operationalizing AI. As more companies like UPS embed AI into operational processes, we’re starting to see them choose a strategic supplier for cloud and AI solutions to help them achieve their business and innovation goals and above all, help them collaboratively to build and apply AI in their operations. This was a topic I covered in Choosing a Cloud for Business Innovation and Change.

Three Tips to Start the Journey to Operationalizing AI

I believe AES, ANZ and UPS are great deployments because they highlight three important considerations for customers that have yet to shift beyond narrow projects or take AI from the lab into operational processes.

  1. Continuous measurement. Focussing on the right business and operational metrics will dictate whether projects succeed or fail. Embedding these indicators into projects allows you to continually measure and optimize success in an ongoing, iterative cycle of improvement. Ensure defined goals and purpose, and that you collect the right metrics to support these aims. This will help you influence each phase of your journey as you expand AI in your operations.
  2. Embed governance into your AI processes early. CCS Insight predicts that governance will become the biggest requirement for AI in 2020, based on the growing need in enterprises for support in the ethical design of applications, the management of data bias and drift, tools to enhance the interpretability of models and help in deploying secure and compliant algorithms. Look to embed governance processes from the beginning when designing AI projects underpinned by a solid foundation of data governance. Remember, performance and trust in AI will depend on the quality of the data practices that you feed into it. Therefore, think of governance not as a process that stifles innovation, but as one that improves the performance, output and above all, trust in the technology and as a fundamental for operationalizing AI in your business.
  3. Choose the right cloud. Whatever stage of the cloud journey your organization is in, selecting the right cloud provider to support your data and AI strategy is vital. Indeed, our latest IT decision-maker survey found that 57% of respondents prefer a single or preferred cloud supplier for their machine learning strategy. Look for a provider that continually invests in a technology stack that takes the risks out of an AI strategy, sparing you the complexity of implementation and letting you focus on how the new tools solve your most important problems. It’s also important to consider a provider that offers support beyond the technology, that can best work collaboratively, share best practices and understand your operational context to best deploy AI responsibly.

Reaching AI’s Potential

AI is now being applied to solve a wide range of important problems, from ambitious goals in fighting climate change, to more specific challenges such as money laundering and improving demand forecasting. But for the technology to reach its full potential, organizations must spend time and effort developing a realistic plan to operationalize the AI solutions they select.

As more businesses define their strategies and structure their AI programmes in the coming 12 months, they should consider some of the lessons learned from companies that are further along their journey, like AES, ANZ, UPS and others. They will help you reap the benefits from operationalized AI in your organization.