AI and Analytics PwC 2022 AI
Business Survey

1. Think better, faster, longer term: Make decisions with AI support

AI-supported decision-making is so powerful for a simple reason: it can enable you to incorporate and analyze far more information than you (or any human being) could do on their own. To help optimize pricing, for example, AI can input reams of historical data on product sales, margins, supplier costs and customer satisfaction, then produce rigorous estimates of possible future scenarios. It can project how competitors and suppliers might change their prices in response to your decisions. AI combined with Internet of Things (IoT) sensors can forecast machines’ performance and maintenance needs, enabling better operational decisions. With the rise in “AI at the edge” — AI combined with edge computing, so AI algorithms are executed at or near the device level — these decisions can be lightning fast, too.

Some complex business decisions are still being made without much use of AI, but that may soon change. AI analysis has much to offer M&A, for example, potentially automating parts of due diligence, predicting likely regulatory responses and projecting a combined company’s results under various conditions. Opportunities for AI to support ESG (environmental, social and governance) decisions also abound. An estimate of an investment’s carbon footprint may be far more accurate, for example, if AI models project future energy supplies, weather patterns and second-order impacts on your supply chain. 

What companies can do

• Start with outcomes. As you consider new AI models in decision-making, don’t start with the data you have. Instead, start with the business outcome you seek, then look for the data and analytics to back it up. Consider which decision-makers will use the model to achieve this outcome, where the model will fit within the decision-making process, how it will integrate with the cloud, and how you will monitor, scale, improve and eventually retire it. 

• Let AI make your data actionable. Once you’ve identified the data you need, let AI help you find and use it. Many companies are drowning in unstructured, “messy” data. Whether with documents, images or video, AI can wade through this ocean of data, extract exactly what decision-makers need and put this data — and only this data — in front of the right people at the right time.

• Focus, then scale. A holistic approach doesn’t mean “everything at the same time.” An effective way to use AI in highly complex decisions, such as ESG, is to start with a specific element, such as a single facility’s carbon footprint. You can then scale up to other facilities and ESG factors.

2. Simulate everything: Seize AI’s virtual power

It’s close to unanimous: 96% of survey respondents plan to use AI simulations, such as digital twins, this year. AI simulations are powerful, because they can do more than provide detailed, real-time insights into current performance. They can increase the speed and help lower the risk of your future operations. By modeling huge numbers of scenarios in parallel, simulations let you quickly project likely events and “game out” your leading real-world actions without taking any real-world risks. For example, when you bring together simulations of suppliers, customers, competitors and the weather, you can better predict supply chain pricing dynamics and disruptions — and have a plan in place to navigate them.

A holistic approach offers a particular advantage for more complex simulations, such as forecasting market conditions and addressing supply chain challenges. With time, simulations may also help overcome talent challenges. Already, nearly two-fifths of “holistic” companies are using AI simulations to help hire and train employees. AI-powered virtual reality simulations enable better virtual recruiting, access to talent in far-flung geographies, better monitoring of remote workers and the upskilling of even hands-on roles. 

What companies can do

• Create synthetic data. Machine learning models require huge amounts of data — which simulation models can create. For facial recognition, for example, instead of acquiring images of faces from multiple angles, contrast levels and brightness, simulations can generate them to train machine learning models. Synthetic data, which AI simulations can provide, can turbocharge other AI and analytics initiatives.

• Make digital twins a platform too. To effectively use AI’s power to create business-relevant simulations, consider (as part of AI’s integration with data platforms and cloud) making digital twins a platform capability — to help make sense of your various data sets in the context of your business, your customers and your products.

• Align your specialists. Your simulation specialists very likely have an engineering background, while your data scientists will typically be more experimental scientists. Bringing these specialists together with each other and with business leaders is key to solving simulation problems. 

3. Put a number to it: Assess and predict AI’s ROI

AI has long had a grave problem for business: a lackluster ROI. Companies were building AI models that worked, but which didn’t focus on the right business problems. This problem has grown more acute as companies advance beyond “boring AI” — with its often immediately attractive returns — to more sophisticated use cases. The more complex the challenge, the more important that business leaders aim their AI initiatives at the right problem. A chief reason for this difficulty has been the challenge in measuring or even defining AI’s ROI. How, for example, do you quantify the value of a better strategic decision? Or how do you put a precise price on that supply chain disruption that never took place, because AI models gave you advance warning? 

Companies are now increasingly able to answer these and other questions, thanks to new assessment methods. These can capture not just “hard” returns, such as increased productivity, and “hard” costs, such as new hardware spending. They can also capture “soft” returns, such as an improved employee experience, and “soft” costs, such as increased demands on subject matter specialists’ time. The holistic approach to AI, by fostering scale, shared insights and shared leading practices, also makes it easier to predict the ROI of new initiatives. As you work to measure and increase AI’s ROI, you may also be able to take advantage of AI itself: its simulations can model the uncertainties that surround other AI initiatives, helping to better allocate resources.

What companies can do

• Be complete. As you use the new methods to assess and predict AI’s ROI, be sure to factor in uncertainties (such as AI models’ exact error rate), changes in model performance and maintenance needs over time, and how different AI initiatives could impact each other’s results.

• Build a portfolio. To help avoid ROI surprises, consider a portfolio approach: as you might do with financial investments or product innovation, create and assess a mix of initiatives that will raise the likelihood of delivering the overall results you need.

• Manage the life cycle. To further predict and boost AI’s ROI, aim to manage not individual AI projects, but an integrated data-AI-cloud (DAC) life cycle. That can help you continually evolve strategy, fine-tune execution and find new use cases for data — both avoiding downside surprises and identifying new value.

4. Protect yourself: Make your AI responsible

Ninety-eight percent of respondents understand the imperative: They have at least some plans to make their AI responsible in 2022. Responsible AI systems do what you ask of them — no more and no less. When well implemented, responsible AI processes can assess your models for explainability, robustness, bias, fairness and transparency. Responsible AI governance also offers checks and balances and escalation protocols when you evaluate and validate AI models. Yet, even though nearly every company has responsible AI ambitions, for each specific leading practice, fewer than half are planning action. Holistic AI leaders are doing better, but there is room for improvement.

Given the widespread understanding of responsible AI’s urgency, the lack of action likely reflects the challenges. Responsible AI requires both technology and business experience. AI professionals may, for example, miss the impact on compliance or the brand when AI makes decisions based on historical data sets, which may be rife with historically common biases. Business and risk experts may lack the technical skills to forecast how highly complex algorithms may perform as circumstances change. And since AI continuously evolves its own decision-making based on new data, it requires governance and protection that evolve too. 

What companies can do

• Govern the life cycle. To keep up with fast-changing AI models, deploy end-to-end governance of the DAC life cycle.. This governance should integrate risk, AI and business leaders, with new procedures, roles and responsibilities for each of your three lines of defense. You often can employ and enhance much of existing IT governance and controls, but many business and risk leaders may need to learn some AI and data science basics. 
  
• Assess the impact. To facilitate the work of integrated teams and life-cycle governance, consider algorithmic impact assessments. By evaluating the end-to-end AI life cycle, they can capture risk, identify governance needs, increase accountability and facilitate go/no-go decisions.
• Minimize bias. Many companies today are understandably focusing on responsible AI basics: making sure that AI is safe and does what it’s supposed to. But as AI supports ever more business-critical decision making, it will likely become increasingly important to decrease AI bias, so that your AI models treat all of your stakeholders fairly.