AN ANALYTIC PLAN: THE KEY TO SUCCESS
By Jennifer Jones, Sentier Analytics
One day everyone is talking about getting back to the new normal. Soon after, we are talking about a COVID surge. If you are in the business of sales, what worked last month may not be effective this month. In this fluctuating environment, quick turnarounds of machine learning (ML) analytic projects are essential. The key to a fast, accurate, and successful analytic project is a well-written analytic plan.
A goal without a plan is just a wish.
-- Antoine de Saint-Exupéry
The goal of an analytic project is to answer business questions, usually about the future. It involves coordination from multiple teams, and tight deadlines are not uncommon. The analytic plan will keep everyone knowledgeable about the project from the moment the ideas are conceived to the final delivery. This document helps to make your goal a reality by providing purposeful steps within a defined scope instead of just hoping everything gets done properly and on time.
There are many components in an analytic plan, and each one serves a purpose, but not all sections will be relevant to all team members. The length of the document is not as important as ensuring that all the pertinent information is included and organized. At a minimum, the plan should contain the following:
The overview provides brief, high-level background information about the project. It helps introduce team members to the project who may not be familiar with the goal and why it is desired. The final due date should be mentioned in the overview to give readers an immediate idea of the timing.
The business questions should be stated first by the business client using their terminology. After initial discussions, the questions may need to be rephrased for better understanding or modified in scope to meet the timeline. Therefore, the plan may have the original business questions as well as the analytic plan business questions.
Multiple analytic inputs are required to answer the business questions. Some of the data inputs may be available internally, while others may need to be purchased or obtained externally. Client inputs such as optimization ceiling, floors, desired amounts, etc. are also captured here. This is the place to ensure that the business stakeholders are comfortable with any keywords that will be used for the deliverables. It is much easier to change nomenclature at the beginning of the project than to go back through the various steps of the deliverables and unify the terms.
The data assets section catalogs which database, schema, and tables will be created and/or used to supply the analytic ready data inputs to the data science team. It should also outline which role is responsible for the creation of each asset. This section is critical to the data engineering and data science teams but will most likely be ignored by the business team.
The analytic approach defines precisely how the data science team will build the model and analyze the data. The business unit must understand this step-by-step approach as they will likely be asked questions later on about how the answers to the business questions were determined. Therefore, the approach should also include any assumptions as well as how accuracy and usability are measured.
To avoid delays, the expected deliverables and their format should be agreed upon upfront. If it takes a week to run a model and the business unit cannot use the standard output, then the team will have to spend another week or more making changes to the output and re-running the model. It is also possible that deliveries in the middle of the project may determine the direction for the remainder of the project. Not all deliverables will be due at the end of the project.
Project timing and milestones should be defined at the granularity that provides confidence for all team members that their responsibilities can be completed in the expected time. With the nature of analytic projects, most tasks will be of a finish-to-start nature which means the predecessor task must be completed before the successor can start. Once the timing and milestones are laid out, it is possible to discover that the original, desired due date is not achievable. If that is the case, discussions surrounding scope, approach, resources, and timing should continue to reach an acceptable and obtainable timeline.
The analytic plan is a living document owned by the business analyst role on the team. While the U.S. Constitution has also been referred to as a living document, it is in fact, rarely changed. However, the analytic plan is going to change frequently, and like the Constitution, the business analysts must ensure all the amendments are recorded in the document. It is too difficult to keep track of multiple emails, especially across a diverse team. The analytic plan, once initially agreed to, will likely be updated at least twice a week as new details come to light and there are potential changes in direction. It should be the source of truth if team members have questions about the project. As knowledge is gained with each analytic project, the plan also helps the team to remember and avoid past difficulties, while ensuring improvements carry forward to the next project.
If you believe in analytic projects, then you believe in planning for the future. Just as the results of the analytic project are used to plan for the future, a comprehensive analytic plan document is the blueprint for successful and expeditious completion of the analytic project.
----
Sentier is an advanced analytics company delivering disruptive insights to our commercial pharma and biotech clients through our AI driven platform. The platform allows Sentier to deliver insights faster, at a lower level, and more accurately than traditional solutions.
