Clients admire our use of predictive analytics. It gives them a strong and practical foundation for decision making. What were once unstructured ideas and opinions, become clear decision points with buy-in from their organizations.
We use predictive analytics differently than the more commonly known use by, e.g., Google or Facebook. To us, it is not big data. It is not short-term optimization of algorithms. We use it for strategic insights. Here is our version of predictive analytics:
Definition of Predictive Analytics
1. Predictive analytics is used to extract knowledge from quantitative data to predict trends over time, or to see patterns across cross-sections (individuals, consumer segments, countries, or other sets, at a point in time).
2. Predictive analytics can be based on big data or small data.
3. Predictive analytics can be used for tactical, operational, or strategic problems. Examples:
- Tactical: Google wants to optimize which at to present when you do a search
- Operational: McDonald’s wants to plan hamburger bun deliveries for the next week, and month by restaurant
- Strategic: IBM wants to know if Myanmar holds long-term potential from a demand perspective
4. Exploratory analytics are different from predictive analytics. Exploratory analysis looks at patterns in the past to explain what causes outcomes. Exploration is valuable, but it should not be confounded with prediction.
Our Use of Predictive Analytics
1. We are the only firm on the planet systematically leveraging predictive analytics for strategic purposes (long-term, cross-functional, C-suite)
2. We use mainly medium and small data. We invariably face lots of missing data and data quality is usually poor. This is good because if data sets are clean the knowledge in them has most likely already been extracted.
3. We combine many data sources. Our data sets never exist before a project. If they do, we shouldn’t be doing the project.
4. We always combine quantitative and qualitative analysis. A purely quantitative model is of limited value in strategy work. Clients pay us for our judgment, augmented by statistical findings. Not for the statistical findings in isolation. Even a highly technical project on price elasticities has the client saying “we retain you because you apply managerial judgment rather than just giving us the numbers.”
Our Predictive Analytics Methods
1. We keep the number of statistical techniques at a bare minimum and simple since we don’t want to discuss techniques, we want to discuss findings.
2. The techniques we use are mainly: correlation, linear regression, pooled linear regression, Armax, time series, logit, nonlinear regression, and cluster analysis. Occasionally there are others. All of these fall in the regression or cluster domains.
3. The Golder-Tellis (GT) affordability model is of particular importance to us. GT by definition gives elasticities: price, income, consumer confidence, distribution coverage, marketing spend, and other, depending on how much of GT we use.
4. In analyzing consumer surveys, we do not use cross-tabs because it is a poor technique for extracting knowledge. Instead, we use logit.
* * *
In the end, predictive analytics is a tool for us. It is not what we deliver. If we don’t have to use it, we don’t.
Nevertheless,we are masters at applying predictive analytics and it is a VRIO resource for us and our clients.
It is Valuable in problem solving; what we offer is Rare–no other top tier management consulting firm has it; it is Inimitable because others face steep entry costs in building the competence; and we have it Organized so that clients can benefit from its value no matter which project team serves them.