AI in retail requires cause-and-effect reasoning

Doing more with less is becoming a mantra for the retail industry as 2023 progresses with more global economic challenges than it imagined even months ago.

The key questions are echoing around boardrooms of both bricks-and-mortar groups and e-commerce businesses:

• What pricing and promotional strategies will deliver the greatest ROI without compromising brand equity or value proposition?
• In revenue, category and merchandising, which strategies maximize EBITDA through higher gross profit and greater operational efficiencies?
• How can we understand our customers with less granular data, while maintaining personalization in a privacy-first world?

These are all causal questions. While AI platforms and solutions used in retail have previously been limited by correlating historical and ‘curve fitting’ to arrive at the best possible decisions, causal AI is now able to reason and make choices like humans do. Causality can ask ‘what if’ questions about the past to go beyond narrow machine learning; it can understand the impact of possible actions that might affect a desired outcome. This imperative has long been understood in fields such as economics and medicine; only recently has it begun to emerge in the enterprise space, so that a retailer can use the recommendations and insights provided by causal AI to decide on the best pricing for individual products, across specific locations, while taking into account the prevailing business environment.

Today’s dynamic shopper environment means the retail industry needs a new approach to developing pricing and promotional strategies. Competition may be driving growth through promotional investment, yet more than 70% of these investments are delivering a negative return. Optimization is made difficult by volatile macroeconomic factors: these make visibility of demand and supply, and therefore the impact of experimental pricing strategies, very difficult. Pricing soon becomes an issue of balancing goals of growth, profitability and inventory flow.

The next generation of AI will deliver to retail leaders KPI optimization platforms that look at a business as a whole, including understanding causal elements, in order to make critical business decisions. Be warned: the technology won’t be downloadable from the open source community for bending into shape by a few data scientists; it will involve ‘thinking’ at a much higher level of automation, such as ‘Do we know the root cause of our outcomes?’. After all, KPIs and ROI are outcomes to decisions, which require a causal element. The retail industry is affected by myriad external factors: relationship management; seasonal purchases; operational risk; expert judgement; budget constraints; and business context.

So how might it work in practice? Through causal discovery and inference algorithms, millions of data features are defined and connected — not just statistical, but also causal relationships. Supply chain knowledge is embedded, allowing subject matter experts to inject more accuracy to a causal graph. Since the retail industry is broad and complex, this domain element is vital. The combination of top-down human expertise and bottom-up data discovery is very powerful.

A common use of AI in retail at present is churn prediction. However, results are typically underwhelming, with churn models failing to respect the business’s constraints and goals as it struggles to perform in a fast-moving environment. Conventional AI can predict likely churners, but can’t recommend a course of action to prevent churn. Causal AI, by identifying the true drivers of churn, is uniquely capable of recommending a set of interventions to optimally allocate resources and budgets to increase customer retention, based on business goals and key metrics.

Reducing customer churn with the help of ‘explainable’ AI, which allows for intervention to produce transparent, understandable recommendations, can deliver significant ROI benefits: a 5% increase in retention is estimated to increase profits by between 25% and 95%. Retail leaders are only too aware of the cost of churn: it’s six to seven times more expensive to acquire a new customer than to retain an existing one; retained customers buy more often and spend more than those recently recruited; satisfied, loyal customers refer new ones, with no associated cost.

Causal AI models that allow for the inclusion of relevant human expertise, or ‘domain knowledge’, can empower decision makers such as marketers to interact with a causal graph to see which factors affect their target KPIs, and in which ways. They are then equipped to add in the constraints and nuances of their domains and the world their customers live in. Mapping the causal drivers of a domain also enables explanations for recommendations and current retention rates.Traditional machine-learning solutions for retention struggle to avoid biases, so a marketer may want to assess the causal AI model to ensure that it is fair when it comes to, for example, age or gender.

However, these gains are harder to achieve with machine learning that relies on past patterns and correlations to make predictions — and is therefore prone to fail amid shifts in data distribution.

Machine learning isn’t truly synonymous with Artificial Intelligence; the real AI revolution only starts when machines can learn like scientists, which requires fusing them with domain’ expertise from humans to make more sophisticated algorithms. When the causal drivers of demand or supply in the retail world change, even sophisticated, curve-fitting AI models can make worse decisions than the toss of a coin.

AI has for some time been proposed as a transformative development in retail. Nonetheless, in reality a level of Machine Learning capable of truly optimal decision making is only now emerging. Using cause and effect, causal ai, a new category of intelligent machines that reason as humans do will become a revolutionary tool for solving the challenges of retaining and growing retail customers. The long term future for AI in retail is very bright indeed.