RBC AI Summit Workbook

What is AI, Really?

These days, buzzwords like AI, ML, and Gen AI are everywhere, and it can be challenging to understand what these terms mean. Artificial Intelligence (AI) is an overarching umbrella term describing technology that enables machines to perform tasks typically associated with human intelligence, such as recognizing patterns in data or solving problems. Simple AI models, like rule-based expert systems, have been used for decades. For instance, a biology research team might have used AI to identify animal sounds in a long recording (if a given sound overlaps with frequencies typical of canine vocalizations [e.g., 2000 Hz for barks], then classify the sound as canine).

However, the current boom is driven by a subset of AI known as machine learning (ML). Unlike traditional models, ML systems do not require explicit if-then instructions for every task. Once trained on a large enough amount of data, ML models continue to self-learn. For example, when an ML model is trained on thousands of labeled audio samples of wolf and dog calls, it learns to identify the distinguishing characteristics between the two canines - patterns and correlations of pitch, frequency, duration, and so on. After training, the model can apply these self-made rules to assess the likelihood of a new audio sample belonging to a wolf or a dog, outputting probabilities like "70% chance wolf." With every new data point, prediction accuracy improves.

Generative (Gen) AI is an exciting new step. Machine learning models analyze large amounts of data and then "knowledge" patterns together to generate new content, like text, images, and audio. For example, a Gen AI model that has learned various canine sounds like could generate a realistic dog call or create audio of a being that does not exist in reality. This creature's call could be a hybrid of a coyote with a vocalization pattern somewhere between a wolf and a dog. Therefore, Gen AI, and AI in general, have become the subject of both hype and speculation.

Why use AI?

Forget the fanfare (and the failures) - what do firms actually use artificial intelligence for today? While some uses are quite technical - predictive maintenance of manufacturing equipment, real-time optimization of energy grids, image recognition for surgery - any business can fruitfully employ AI (and not just for writing emails faster).

For example, imagine that you are running a business that is launching a new product. You first list your product on e-commerce sites, all of which now come equipped with the capability for personalized product recommendations to increase consumer engagement with your offerings. In the first weeks, you can use AI to analyze ad spend, quickly spotting patterns in how consumers react to your advertising and optimizing ad placement. During launch, you run sentiment analysis on customer feedback and social media posts to identify likes, dislikes, and issues. Post-launch, you devote your resource time to issues and build an understanding of customer preferences for future products and marketing campaigns. Then, once your product has been on the market for a bit, you can predict product demand based on customer preferences and sales data, optimizing inventory management. By doing so, you increase efficiency and reduce costs. When you want to get your product into a brick-and-mortar retailer, you use lead scoring to rank potential retailers based on their likelihood to buy your product, allowing you to prioritize the most promising prospects.

AI is good for more than just the chatbots you see as a consumer - though those do hold potential for improving customer service. AI models are immensely powerful tools with the capacity for predictive analytics (i.e., analysis aimed at identifying the likelihood of future events.) Correspondingly, a well-trained AI model could forecast sales, anticipate customer needs, and predict market trends. The strength of AI lies in its ability to process large amounts of data, uncovering complex patterns that humans and traditional statistical methods might miss. Furthermore, AI models based on machine learning are dynamic: they continue to learn and improve over time. Even the simplest NLP-driven chatbots are self-updating, refining their responses based on new data to meet customer needs better. As a result, a machine learning model is more than just a statistical package - it is a continually evolving system capable of adapting to new information and delivering increasingly accurate insights.

In sum, AI can enable faster, more precise, and more proactive decision-making. Additionally, AI can be used to automate routine tasks, quickly completing large batches of 'busy work' (e.g., personalizing email marketing content) and freeing up staff to focus on more critical functions. Hence, firms at the cutting edge of AI are already reaping the benefits of operational efficiency and innovation^[1], while stragglers risk being left behind. Still, AI is not without risks or challenges, meaning implementing these technologies must be planned and carried out carefully.

Industry Insights

Currently, the heaviest users of AI are in software and information services, banking, and retail, followed by telecommunications, healthcare, and capital markets^[2]. However, all industries are investing in AI, including high-tech ones like energy and utilities, automakers, engineering and construction, personal or professional services, and agriculture, and lower-tech ones like hospitality and education^[2].

In a McKinsey & Company survey of 1363 professionals, 65% reported using gen AI in at least one business function^[3]. Most used it for:

MARKETING AND SALES

1. Content creation.
2. Personalized marketing.
3. Lead identification.

PRODUCT DEVELOPMENT

1. Design development.
2. Literature review.
3. Early simulation.

IT

1. Help desk chatbots.
2. Data management.
3. Help desk assistants.

Participants reported that AI use resulted in substantial cost reductions when applied for human resource functions and significant revenue increases when applied in supply chain and inventory management^[3]. Other notable benefits were cost reductions when using AI for risk, legal, and compliance, and revenue increases when used for marketing and sales, product or service development, and service operations.

Correspondingly, businesses seem to be getting what they want out of AI: improvements in efficiency and productivity and reductions in costs^[1]. Additionally, 58% of participants of a Deloitte survey reported a broader range of benefits, including innovation, improved products or services, and better customer relationships^[1].

Of course, not all firms get the same results. The leaders - seeing increased earnings - adopt AI for more business functions, like risk, legal, and compliance, strategy and corporate finance, and supply chain and inventory management^[3]. Additionally, they do not simply “take” ready-made models, they “shape” them, customizing the models with their own data, or build their own models. Lastly, they are more likely to establish governance structures and proactively mitigate the risks of AI^[1][3].

Risks and Challenges

The first major challenge of implementing AI, particularly machine learning, is meeting the substantial data requirements necessary for effective performance. AI models rely heavily on having access to large volumes of high-quality and diverse data to produce accurate and reliable results. Without sufficient data, organizations risk falling into a “garbage in, garbage out” scenario, where poor-quality input data leads to equally poor outcomes from AI models. This issue not only undermines the potential benefits of AI but also wastes resources and efforts.

In response, organizations are increasingly prioritizing investments in data management practices. This includes establishing protocols for data quality assurance — such as regular data cleaning, validation, and enrichment—to ensure that AI systems are working with accurate and consistent datasets. Companies are also recognizing the importance of data privacy and security, especially in light of evolving regulatory frameworks like the General Data Protection Regulation (GDPR) in the EU and the California Consumer Privacy Act (CCPA) in the US^[4][5]. These regulations impose stringent requirements on how data is collected, stored, and used, pushing firms to strengthen their compliance efforts.

Moreover, to protect intellectual property rights, companies are implementing strict policies on how proprietary data is handled, especially in relation to public AI tools. For example, many organizations are now training employees to avoid inputting sensitive or proprietary information into publicly accessible AI models like ChatGPT or Bard, as doing so could inadvertently expose confidential data and compromise the company's competitive advantage^[6][7][8].

The second major roadblock to AI implementation is ensuring the usability of results. McKinsey & Company, for instance, reports that professionals working with AI consider inaccurate results to be the most pressing risk and the risk they focus most on mitigating^[3]. It's no wonder, as a quarter of survey participants reported experiencing negative consequences from inaccuracy^[3]. Even when results are accurate, they must be interpreted for real-life use and integrated with an organization's overarching goals.

Herein lies the third challenge of implementing AI - skill gaps. Interpreting AI results is not a simple task, requiring specialized training and hands-on experience. Without proper knowledge, employees may struggle to understand AI outputs and the implications of their insights^[9][10]. Simply deploying AI tools is not sufficient; companies need to invest in building analytical skills so that employees can effectively utilize these technologies.

Most employees, however, will also need upskilling in the basics, such as keeping proprietary data secure and structuring effective queries. Properly framing queries can significantly impact the relevance of AI responses, while training on data handling is crucial to prevent potential breaches. Investing in upskilling through workshops and practical training ensures employees can use AI tools effectively while safeguarding organizational data.

The fourth major obstacle to leveraging AI is that most firms fail to establish robust AI governance and risk management. Governance mechanisms, like dedicated advisory boards, are needed to provide standards for testing and quality control and to offer guidance on accountability (who is responsible when AI use results in negative consequences?)^[11]. Even firms that use ready-made solutions must establish criteria for selecting model providers^[3]. You do not want to end up, for example, relinquishing the rights to customer feedback to a customer service chatbot model - that data is valuable to improving your organization's operations. In short, governance is necessary for responsible and fruitful AI use.