Businesses and investors universally believe Artificial Intelligence (AI) and Machine Learning (ML) can fuel new revenue and profit growth by reinventing customer journeys, transforming the customer experience, and optimizing investments in marketing channels. Consequently, Investment in AI and ML for marketing applications is booming.
· Ninety percent of marketers are using AI to improve their customer journeys, revolutionize how they interact with customers and deliver them more compelling experiences according to a recent survey of global executives.
· Most CMOs plan to add AI and Machine Learning to their marketing toolkit and increase their investment in analytics to over 8% of their overall marketing budgets over the next three years according to the most recent Duke CMO survey.
· In parallel, investors have poured more than $5 Billion into over 1,400 AI fueled sales and technology companies to meet this demand.
Despite this frenzy of spending, the impact of these investments on growth is not very clear. 70 % of AI initiatives have shown little or no return. Most marketers still do not understand how data and analytics will contribute to profit growth according to the Forbes Marketing Accountability Report. The majority of CFOs struggle to prove the financial return on their investment in analytics. And most marketing decisions are not data driven according to the CMO Survey by Duke. This raises several obvious questions:
· If AI offers so much growth potential, why do most AI investments fail to meet expectations?
· And what will it take to realize the promise of big data and advanced analytics to accelerate growth?
It turns the problem is not really the models behind AI, but with the mindset of marketers and the business discipline of how they are using AI tools to identify ways to accelerate growth. The real issue is whether marketers are asking the right business questions so AI can help in making better decisions and create new demand?
“What most marketers call AI is about one third true analytics, one third hype, and one third something unrelated to AI and Machine learning altogether” according to Will Thompson the Publisher of Forbes AI. “AI strategies are often flawed at the concept level because business leaders confuse corporate marketing strategy with campaign optimization, demand creation with demand optimization, and A/B testing with scalable growth strategies.”
“In our experience, a primary reason for these mixed-results is a fundamental lack of understanding among marketers and their stakeholders about how and where AI and Machine learning tools are appropriate to address the macro challenges of marketing” reports Raghuram Iyengar, Miers-Busch W’1885 Professor and Professor of Marketing at Wharton who teaches executives how to use customer analytics to drive growth using AI and Machine Learning. According to Professor Iyengar, most organizations face a common set of challenges as they try to generate business value from AI:
1. They don’t apply the scientific method to growth problems. Businesses can create more value through AI by asking better questions than analyzing bigger data sets and simply letting the data “speak”. Defining a hypothesis, testing it, and proving it right or wrong, and refining it over many tests. According to Abraham Wyner, Professor of Statistics at Wharton, “AI is 50% framing the right questions, 48% about domain knowledge about the business problem, and 2% about advanced analytic methods to refine and validate that hypothesis.” A common problem is that managers tend to delegate the task of defining the question big data needs to answer to AI tools that are not well suited to that task. AI tools are excellent at mining huge data sets to test guesses, ideas or hypothesis about what will make customers buy more, stay longer, and pay higher prices. They are not useful at coming up with ideas to test and defining experiments on their own.
2. They treat AI models and tools as “black boxes”. Many executives don’t understand enough about advanced algorithms and models to question their validity or relevance to the business problem they are trying to solve. Advanced AI and Machine Learning models create value because they allow businesses to quickly and affordably analyze very large datasets and wide variety of variables to derive actionable insights about complicated (non-linear) relationships not easily described by simple regression models. Data scientists can now draw upon an arsenal of models to analyze a wider array of business dynamics. But AI only works when the right type of model is applied to the right problem in the context of the business situation. This requires significant domain expertise about the business problem and thinking about how to fit the problem with the right model. Unfortunately, analysts are generally are expert in modeling, and managers have the business acumen and judgement needed to fit the right model to the right business problem. But both are necessary to generate value from AI in marketing.
It’s important to understand what Machine Learning and AI tools are good for, and what they are not…..They are not a substitute for decision-making, rigorous hypothesis development, and test design.”Ron Cline, Head US Marketing Data and Analytics at the TD Group
3. They get the business case for AI wrong. Analytics teams struggle to explain or justify their contribution to firm value and financial performance because they lack a common economic purpose for analytics investment.A big part of the problem is finance tends to treat AI as a short-term marketing tactic instead of as a long-term growth asset. Leveraging AI technologies in marketing involves large long-term investments in digital technology, recommendation engines, data, and analytics that are difficult to translate into short term financial outcomes and sustainably create new demand. ROI models based on short-term marketing attribution measures of performance cannot capture the full value of AI investments or provide the right incentives for management leading those projects. At Aetna, we are fortunate to have a common economic purpose for advanced analytics because analytics have historically played an important role in all aspects of our business – from pricing, to medical decision-making, and marketing” says David Edelman, CMO of Aetna. “This corporate support for analytics has allowed us to build a formidable analytics team of hundreds of data scientists to support advanced analytics (including AI) at scale across the organization under a dedicated Chief Analytics Officer.
4. They get the unit of analysis wrong. Many AI initiatives fail to get traction because their unit of analysis is mismatched with the engrained Key Performance Indicators (KPI) and incentives managers use to run business. For example, AI models are excellent at analyzing data about customer characteristics, responses and behaviors to refine scalable ways to improve customer lifetime value through improved acquisition costs, churn, pricing, usage and cross sell. But most executives are organized around products, geographies, and business units. “A big problem is the structural mis-match between how customer analytics should be structured, and how they are measured, compensated and incentivized” says Professor Iyengar, who teaches a Customer Analytics for Growth Executive Education program for growth oriented executives twice a year. “Customer analytics creates value because it helps organizations maximize the lifetime value of the customer (which is longer term), but most business leaders are compensated on optimizing short term customer profits across products, divisions or business units.”
5. They focus on big bets vs. incremental gains. Because of the hype and level of investment in big data, AI initiatives are often tied to transformation initiatives with big game changing expectations. The focus on getting big returns on “big bet” change the business projects vs. the more practical goal of generating incremental learning and business model change. Taking a Kaizen approach to analytics – treating it as a process of continuous improvement and incremental learning – will build organizational knowledge, consensus and confidence to and yield big gains over time.
So, what can Chief Data Officers and Analytics leaders do to harvest greater returns from their huge investments in big data sooner?
1. Define the problem properly and concisely. This requires a breed of managers with both analytics and “storyteller” skills needed to communicate possible explanations or how customers are behaving, why they buy, and what will make them behave in ways that generate more firm profits.
2. Separate the hypothesis and the test. Focus on rigorously defining the most important business questions to be answered first, and they best models and data sets to answer those questions second. It’s important to understand what Machine Learning and AI tools are good for, and what they are not” reports Ron Cline, Head US Marketing Data and Analytics at the TD Group. “These tool and modeling techniques allow us to analyze more data sources, faster, with more statistically valid results, and less work. But they are not a substitute for decision-making, rigorous hypothesis development, and test design.”
3. Be transparent as possible. Make every effort to make the AI models being used to analyze data and make recommendations as understandable as possible to the people involved in the decision-making process to build confidence, and understanding by all the stakeholders involved in the execution of model recommendations and AI driven growth programs.
4. Make sure you create the data you need to support AI tests. A natural obstacle to progress is to use the lack of data or clean data as an excuse not to conduct data experiments and learn from the data that exists. “It’s important to create the data needed to support AI. What we’ve learned is much of that data includes marketing variables – the content, offers, pricing, channels, and timing that you believe are driving customer behavior, their response, and outcomes you are testing for”, according to David Edelman, CMO of Aetna. “In addition to dozens of data scientists, we make sure many of the people on our advanced marketing analytics are marketers – content creators, campaign managers – who need to define the hypothesis to test, and structure marketing data to support those tests”.
5. Become more customer centric. Establish organizational roles, data structures, and financial incentives centered around the customer to focus analytics on ways to maximize customer lifetime value. Ultimately, the customer needs to be the center of data aggregation, process redesign, measurements and business cases for customer analytics to deliver their full value.
6. Learn one step at time. Just like athletes need to train to become proficient, organizations need to build “analytical muscles” to move from descriptive to predictive to proscriptive insights. Effective AI strategies use a continuous process improvement approach by making small changes, achieving incremental improvements, learning, modifying to enrich models and algorithms over time. Make sure your analysis first can accurately describe what is happening in customer interactions, then add data and tests to start to predict what will happen if certain conditions exist, and ultimately build the confidence to prescribe what actions will optimize profits and growth.
To educate business leaders on ways to capitalize on the full potential of advanced analytics to drive growth, the Wharton Business School offers growth oriented executives a one week Executive Education program “Customer Analytics For Growth” where Professor Iyengar, and an elite team of faculty teach ways to use Machine Learning, AI and big data to drive profitable growth. The next session in October 24-28th in Philadelphia, PA.