Dashboard to Decision Making: Why innovation is needed
Background
Data-driven organizations have business intelligence (BI) tools that give visibility to take decisions on deterministic events (largely based on historical observations). While these tools have solved the inefficiency problem of week-long delay to get a report (from multiple input sources), there still remain a number of steps before a decision maker uses of this information. For instance, users still have to download reports to perform complex analysis to identify the impact of multiple actions to a given problem.
This article compares 4 attributes of traditional analytical systems to the emerging needs of intelligence in decision making.
1. Deterministic vs. Probabilistic
Many enterprises have data warehouse for structured data or a data lake that incorporates structured or unstructured data, primarily for business intelligence. They help to provide information to the questions such as
“How was my marketing budget spent during the last product launch?”
The answer to the above question is deterministic and the analytical database, with historical data connected to tools such as Tableau, Power BI, provides the result in a nice looking dashboard or a report.
While the above insights are definitely useful, making decisions about the future requires some more analysis. For instance, a typical business question could be:
“How much marketing spend should be allocated to a certain channel for new product launch?”
Although there may be historical data from previous campaigns, the answer to this question is not deterministic and requires advanced marketing-mix modeling. The model gives a probabilistic output of a best case, worst case and range that gives a comprehensive view to take a decision.
2. One-way vs. Two-way Analysis
The BI tools perform one-way analysis because the information is derived from a processed data and the end user cannot update the data to get an alternate result.
The users apply filters to the dashboard that runs complex queries in the database and gets desired results in the UI. Furthermore, algorithms run on historical data to generate predictions that are stored as part of the processed data layer. Reports present comparison of actuals vs. forecast. While many people call this predictive analytics ready system, the reality is that end user still cannot change inputs to the query or to the algorithms that determine the impact of one variable over another for the outcome.
The limitation of one-way analysis is being addressed by Decision Intelligence. It allows business to ask questions, then the model figures out what would be relevant for the question, runs various scenarios and presents recommendations to take the next best action. Running various scenarios is the two-way analysis.
An application providing two-way analysis, takes the user input (as a parameter), processes the information (in the model) and generates the corresponding results.
A practical example would be a demand forecasting use case where the decision maker wants to see the impact on the profit margin between a buy one get one (BOGO) or 50% instant discount.
Traditionally, to respond to this new request, someone has to re-run the model, engage in back and forth communication, generate a new report — all of which leads to slow realization of value and missed opportunities.
With a two-way analysis, the decision maker inputs the discount % or promotion type as a parameter and the model programmatically runs to give an updated impact of profit margin. There is no pre-processing required to put the data into an analytical database.
3. Slicing/Dicing vs. Scenario Analysis
Many tools allow custom formula on the data to perform “what-if” scenarios and show different results of impact. The sophistication of tools includes slicing and dicing of data and on-the-fly calculation to get a new result. While these softwares have solved the compute problem with large dataset, for decision makers the formulas are predefined and they can only change the constraints to get a different output.
The biggest difference between slice/dice of data and scenario analysis is that the former is about the past or pre-computed values, but the later is to understand future impact of various options.
Continuing with the above Marketing examples, assume after the product is launched, Finance asks the question during the budget allocation, “What ROI can be expected from a certain (eg. digital and mobile) marketing campaign?”
Even with the best algorithm, the exact ROI is difficult to estimate because it involves many factor such as target audience, propensity to use that channel and so on. However, the worst and the best case scenarios can be estimated. Without the knowledge of the range of possibilities, it would be difficult to give an approximate ROI for securing the budget.
Scenario analysis is done to get a complete picture of the outcomes. The decision makers can use their functional knowledge to input appropriate parameters (business conditions) and choose the right scenario for the outcome.
The decision is as good as the completeness of the scenarios and the functional knowledge of the decision maker. The former can be solved by giving a comprehensive scenario analysis capability to the decision maker in a business friendly manner (that is, with no code interface).
4. User Interface: Metrics vs. Answers
Most of the BI systems have a dashboard that has various forms (eg. Control Tower) and it presents data through components such as charts, tables, summaries. Dashboards are powerful, classy and good at showing metrics.
While dashboard provides nice trends for questions like, month over month revenue growth, it doesn’t answer whether business was expecting more and if so, how much more?
Such questions leads the dashboard developer to provide more metrics until the dashboard becomes cognitively overloaded.
While dashboard is the starting point for analyzing data, the decision makers are seeking answers or recommendations for taking decisions. Can there be a simple interface where people can ask a question, and the system looks at past models, intra-company chats, and synthesize an answer? Google has set a high bar for public questions. Can the enterprises catch-up?
Conclusions
Dashboards are key for business and will continue to be part of data insights. The key is to strategize what is needed besides dashboard to improve decision making process. Clearly, the probabilistic nature of business problem, requirement of two-way analysis and unlimited scenarios, is encouraging organizations to evaluate new framework of Decision Intelligence.
As the enterprises are transforming to be more data-driven, it is equally important to consider a system for accelerating time to make decisions with a better outcome. It would be good to see how this space evolves in the coming years.
References:
Decision Intelligence vs Business Intelligence
Decision Intelligence in 2020: In-Depth guide for businesses
Decision Intelligence vs. Business Intelligence: What Is Your Company Running On?
Expanding physician’s reach for biopharma use case
Introduction to Decision Intelligence
What Makes Decision Intelligence a Better Framework for Decision-making Models
