How To Convert A Business Question Into A Causal Question: Advancing Knowledge Graphs In Early-Maturity Businesses
Thank you for subscribing and being part of the community. If you learn from or enjoy this post, give it a like and reshare. Your feedback guides my content decisions.
Get $400 off if you act now! Take advantage of early registration pricing for my next AI Product Management Certification cohort. According to Glassdoor, AI Product Managers have a higher salary than data engineers, data scientists, and analysts. Learn more and enroll here.
It’s not too late to save your spot in the next AI Strategist Certification cohort, but only one seat remains. Act now to reserve your place and move your career forward.
In this post, I will explain how to engineer knowledge graphs for more complex abstract systems, even when working in an early-maturity business. If all you get is a simple business problem or even just a KPI to improve, we can still use an iterative approach to build the knowledge graph. We must balance business needs, like quick wins and low initiative costs, with the complexity of delivering a knowledge graph.
I won’t specifically mention dynamism, but each example has a component of change that we must manage by reducing it to its simplest form.
Each business question defines a region on the knowledge graph. I will explain an intuitive way to break down questions so they provide some insights into the data contained in the region of interest. This article contains examples from my past and current work.
Casually Converting A Business Question Into A Causal Question
The business wants to grow its revenue by increasing the percentage of customers who make a second purchase. The business asks, “How do we get more people to make a second purchase?”
We must reframe this question and then break it down. Why do some customers not make second purchases? The temptation is to frame the question in the affirmative, but that can confuse data gathering. We are less interested in gathering data about customers who make a second purchase than customers who don’t.
What are the differences between customers who make a second purchase and those who don’t? We want data that contains predictive indicators about the non-buying segment. Many successful interventions (that cause a second sale) must be applied before it’s obvious that the customer hasn’t made a second purchase. Waiting 3 months may limit our options, so we must be able to predict who is more likely to fall into the non-purchasing segment.
What prevents a second purchase? We need data that contains or points to actions and events that cause some customers not to make a second purchase. This question is an extension of the last one. The differences will point out lines of investigation and further data gathering.
If women make fewer second purchases than men, we can ask experts or gather data to explore potential causes. For many men’s brands, women buy products as gifts.
What new action will cause a second purchase? We also need data on actions and events that cause customers who currently don’t make second purchases to do so. Experiments are powerful data-generating tools. We can try multiple approaches to create datasets with information about effective and ineffective actions.
What interventions are most effective at getting a woman to buy a second gift? In 2017, I helped a client analyze where new customers were coming from. We found (and this trend has been established for multiple clients since) that women often introduce men to essential items from high-end brands (socks, undershirts, shaving kits, etc.). They don’t make second purchases very often because who wants to get their father or significant other the same thing every birthday?
For some brands, a slight shift in marketing to focus on other men’s product categories instead of the same category was successful. What about brands that focus on a single category?
In the past, sharing customer lists with other luxury retailers was inconceivable. However, given the low likelihood of repeat purchases, creating a network that shares gift-givers' information makes sense. Each brand can market to people who haven’t previously gifted from their category.
Our work also revealed a parallel segment that became the true target for second purchase interventions. We needed data on who was receiving the gift, and we incentivized gift-givers to provide their names and addresses in exchange for free gift wrapping and shipping.
Frame the question well, and the next steps follow a logical sequence. This is an abstract system, so the goal isn’t a complete definition. We’re incrementally improving our understanding of customers who don’t make second purchases. Each partial segment and intervention definition delivers value against the top-level goal.
Controlling Complex Causal Question Categories
The first abstract system example was relatively straightforward. A single variable (gender) defined the cohort, and a single activity (gift-giving) defined the cause. In early-maturity businesses, these insights are incredibly valuable. Initiative costs are less than $200K, and one client realized $26M in annual revenue from second purchases by gift recipients. Returns are typically in the 50X to 100X range.
We call these low-hanging fruit because they can be delivered in three months or less. The significant insight was to target the person who gets the gift instead of the gift-giver. Once revealed, it’s fairly obvious. As the business matures and the easy wins are used up, we must evaluate more complex abstract systems with new tools.
Actors And Players
Existing customers are actors in this purchasing decision and players in this game. In decision theory, an actor is an entity with preferences, goals, and actions that affect the outcome of a decision problem. Actors can be individuals, groups, organizations, or even countries. Actors are assumed to be rational, meaning they will choose the best option for themselves based on their available information and criteria.
That’s a major problem with many decision frameworks. We can’t assume rationality, that actors have information, or that actors assess situations using information before making decisions. That’s where Game Theory and Decision Theory break down.
Behavioral Economics challenges the assumptions of rationality and informed decision-making. In business, our work more closely aligns with Behavioral Economics but with much lower scientific rigor.
Actors can also interact with each other, influencing each other’s preferences, goals, and actions. Actors can have different roles and perspectives in the decision-making process, such as decision-makers, stakeholders, experts, or analysts. In the last example, gift-givers interact with recipients to decide on the initial purchase.
Competitors are different types of actors. They work to influence customers’ decisions and influence the marketplace where decisions happen. However, their strategy aims to maximize their gains, even if that comes at other companies’ expense. The gift-giver data-sharing network is an example of indirect competitors collaborating in a mutually beneficial game.