If the demand for analytics is high, both the directorates-general and the business lines, the CIOs who are generally asked to carry out these projects in good condition must also face major difficulties, the most important being certainly the lack of skills, internally and externally. With the corollary that between (high) wages and competent (rare) partners, the costs of analytics are often prohibitive.
However, companies are attracted to AI (Artificial Intelligence), and say they want to multiply predictive analysis projects. Be careful, however, to proceed with caution. Because it is particularly easy, in this matter, to be mistaken. But also particularly difficult to detect errors in time and avoid drift in decision making on a false predictive analysis.
Thus, companies that carry a predictive analysis project usually make three mistakes: they crack for buzzwords without clarifying their purpose; they rush on software while the priority must go to the team; they jump on analytics without planning their deployment strategy.
Here are 5 recognized steps in the deployment standards of predictive analytics, which should not make you forget that backtracking and iterations are important. And that we need to invest in support to understand the requirements and fundamental principles of predictive analysis.
1. Define the ‘commercial’ goal of the project
To have a positive impact on operations, the predictive model needs to be integrated in order to more effectively target customer marketing and retention campaigns.
2. Define a specific forecast goal
To serve the business purpose, a specific forecasting goal needs to be defined, with the support of stakeholders. Changing the targeting should also be possible. In practice, the goal of prediction must be precisely defined.
3. Prepare the data that will be submitted for analysis
Machine learning can be a major bottleneck, typically requiring 80% of the project workload. This is a database programming task, whereby existing data in their original form must be reorganized to meet the needs of the analysis software.
4. Apply machine learning to generate the predictive model
It is at this stage that the choice of the analysis tool counts. But before that, at first, software options need to be tested and compared to evaluation licenses before deciding which software or open source tool to buy or use.
5. Deploy the model
Last phase, integration of forecasts into existing operations.