The Analytic Process

Creating a plan and process before starting an analytic project can save time and money.

Overview

The project management of analytical projects are not remotely as managed as how organizations would typically manage an IT Project, Capital Investment or a Development project. However, the reality is, it should if not equally as important given that the personnel is equally as expensive, time is usually not on your side, and more importantly, there are pros/cons as there is with any project.

Analytical Advantage has a three part approach to projects. The approach is slightly different whether it's a consultancy project or internally with your team.

The four major pillars are:

1. Business Project Charter

2. Modeling

3. Translation

4. Prescriptive

Let's get started.

Pillar 1: Business Project Charter

There is always the toss up of getting things done immediately vs. a thoughtful approach; there is also the saying that data mining is about digging through the data to find something that can change the business. The reality is, every project should be thoughtful and digging hasn't always proved to be telling.

A team of business leader should first sit and assess every project on whether or not it makes sense. Understand the impact of the exercise. The following questions should be asked when planning (not starting) a project.

1. What will the results tell me?

2. How will it change the business?

3. What is the gain? Revenue, cost, service, quality?

4. What is the financial gain?

5. How long will it take?

6. How much resources will be needed?

While data scientist aren't expected to run through these scenarios, the business leaders and the FP&A partners should be able to run through the numbers to make sure the project is worth the time.

It should not be a complicated process, but some basic math and financial returns should be telling. If it becomes too complex of a project, no project will ever get done, so it's important to have a balance.

Pillar 2: Modeling

The modeling process is the time consuming process, and it's not the actual building of the models, but the understanding, merging, and data prepping of the model. While there are many new solutions out there such as AutoML or AutoAI through companies like IBM or DataRobot (we do use both products) to help offset some of the require data preparation time, the customization and business knowledge incorporated in the preparation phase is crucial for improved predictability and to assure that the models are reflective of the future business operations.

Without getting into the details of modeling, below are the stages of building a model.

1. Data Acquisition The process of acquiring data is sometimes the root cause of a project. Because data is not always as accessible teams may believe or the security

2. Data Preparation

3. Modeling

4. Validation & Testing

Pillar 3: Translation

Some practitioners will suggest that the Translation stage a part of the modeling stage, however, it should be a collaborative process as