At a glance
- AI-supported churn model based solely on anonymised data on purchasing behaviour
- Controlling can reliably predict customer churn
- Solution could be built quickly and economically with Azure Cloud services
The business situation of our client
In order to use Artificial Intelligence (AI) to make reliable predictions about the behaviour of your customers, you don’t necessarily need vast amounts of data. Even medium-sized companies with relatively manageable data sets can tap into these exciting possibilities of digitalisation. This is illustrated by our approach to a large German bakery: the company works with anonymous customer cards that primarily offer discount benefits. In fact, the 350,000 active cards are used for about 60 per cent of all purchases. Based on this data on purchasing behaviour alone, we have now trained a so-called churn model that recognises potential customer churn at an early stage makes reliable forecasts.
The solution for our client
To build our AI-supported forecasting model, we first defined the typical “churn”: Customers who have bought from the bakery chain for three months and then not for three months. The model was then trained with this specification and the purchase history from the customer cards.
Our procedure at a glance:
- We train a model with the Advanced Analytics component of SQL Server in the Azure cloud.
- Over a longer period of time, we compare the analyses of the model with the purchasing behaviour of the customers.
- We transfer the analysis results into the existing Business Intelligence system.
- We make the results available to the controlling department via a self-service application.
- We continuously refine the classification by means of new data.
How data turns into new values
The solution structure described above makes reliable forecasts of customer behaviour possible in an economical way.
The advantages at a glance:
- Controlling can reliably predict customer churn tendencies.
- The analysis of individual customers enables individual marketing measures, e.g. special vouchers.
- Even small changes in buying behaviour are recognised at an early stage so that quick countermeasures can be taken.
- Expected sales losses due to cancellations can be analysed by region, branch and time period.
- Higher-level developments can also be recognised, such as when competition increasingly penetrates a specific region.
- By using cloud components, the solution can be set up and expanded quickly and cost-effectively.
turn your data into value.