5 steps to success in AI projects
A prerequisite for successful AI projects is the selection of the right process model. The so-called CRISP-DM method has been known for a long time and is often used, for example, in models for detecting potential customer churn. This can also be linked very well with the concept of agile development. Datasense Consulting has developed a 5-step process model from these two approaches which has already been used successfully several times in customer projects.
In detail these 5 steps are:
Data evaluation and concept development
Detailed analysis and proof of concept
Prototype development
Development of a Minimum Viable Product
Deployment/Production
Each of the steps is provided with the content, a clear effort calculation and the resulting results. The customer can decide after each step whether he wants to implement the next stage, so he has full transparency in the individual phases about costs and chances of success in a project. This should help to keep the initial hurdle for the implementation of AI projects very low.
In the implementation of the creation of a customer churn model (churn model), the individual phases include the following contents:
Data evaluation and concept creation
The central element is one or more workshops together with the customer (business unit, IT) with the purpose of evaluating the potential data for the churn model, e.g. from ERP, CRM, call center, etc.. From this, the exact use case is then defined (e.g. % rate of detection of potential churn-prone customers). In detail, these are:
Overview of data landscape
- Inventory of data sources (billing data, CRM data, call centers, etc.
(Relational databases, document repositories, image archives ...).
- Evaluate data quality and linkability of all data sources.
- Definition of next steps to consolidate all important data
Definition of the possible applications (use cases) such as different churn models (e.g. recognition rate vs. hit accuracy) for different marketing campaigns or customer target groups.
The results are then documented accordingly and are the basis for decision-making for the next steps. One outcome could be that additional data management measures are required to improve data quality, consistency and linkability prior to model building.
Detail analysis and proof of concept
In this part the concept as well as the data from the described sources will be examined in detail with the following tasks
Exploratory detail analysis of the available data
Performing the necessary PoC validations that help to assess the feasibility of the concept
Summary of the results
Clear definition of: Project goal, plan, and success criteria.
Prototype Development
The goal of this phase is to develop a limited version of the solution (often only its core algorithm) to have more clarity on the expected characteristics (performance, accuracy, ...). For churn models, in this phase different models are created and trained, compared with each other using the historical real data, and those with the best results according to the criteria defined in the first phase are identified and selected in this phase, all others are eliminated. It includes the following detailed steps:
Prototype development
Testing of the prototype against available data
Summary of the results
Building a Minimum Viable Product
After the prototype, the AI such as the selected churn models for e.g. high, medium and low hit accuracy, different target groups, is integrated into the existing systems at the customer's site as an example (e.g. automatic data loading and predictions, dashboards, etc.).
Deployment/ Production
After the integration of the MVP into the existing IT landscape (e.g. integration into CRM, call center applications, data warehouse), it is necessary to stabilize it and to implement appropriate automation mechanisms regarding extension by current or additional data and improvement of the model.
If you would like to learn more about what it takes to use AI for your company's success, please visit Predictive Analytics Datasense or arrange a non-binding consultation regarding an evaluation workshop here:Contact Datasense. We are looking forward to a conversation with you.
Comments