Use the full potential of your customer data!

Step 2

Raw data

In case when millions of existing and prospective customers have already interacted with the company, the CRM and the sales data can provide a lot of valuable information that the company can leverage. 

The sales person knows that…

a) Mr. Moore purchases a second-hand car once in three years. 

b) Ms. Smith leases a new convertible once in two years. 

The marketing department at the headquarters knows that… 

a) Mr Moore has already visited the SUV sales webpage several times, looked at the car and downloaded a financing application. 

b) As they get older, ladies prefer an SUV over a convertible.

At the next step, these kind of data get bundled up (centralised), with signatures being derived from the entire data pool. 

Step 3

Customer signatures

A signature is a record of relations and connections in the data pool. Depending on the criteria related to the targets, patterns in purchasing behaviour may be identified.

In our case, combining the insights of the sales person with the insights of the marketing department, the company may conclude that Mr. Moore and Ms. Smith should be offered an SUV - just like other potential customers with a similar profile. The two sample customers are therefore part of a common customer segment ("cluster"), even though they may not have any apparent similarities.

Thanks to the Refining Smart Data process, the car manufacturer may discover the invisible similarities and connections and target them appropriately. Practically, this might be via an invite to an event (= raising awareness), or making a strategic offer as part of a recommendation (advocacy). 

Step 4

Statistics

Once all available data have been statistically evaluated according to signatures, our car manufacturer may realise that 30% of their potential customers have a profile similar to Mr. Moore and Ms. Smith. Therefore, they may wish to launch an advertising campaign in order to boost the SUV sales in this particular customer segment. 

The statistical data are simple in principle, but they may not be used for the purpose of creating models or controlling campaigns. However, a statistical overview may indicate the order of magnitude and distribution. 

Step 5

Modeling

The larger the amount of data available, the more difficult it is to grasp them and make appropriate decisions. Our custom models provide automated calculation, which isolates Ms. Smith and Mr. Moore from the common cluster. This new segmentation is created with a statistics app. 

Our car manufacturer thus gets to know its customers better and may target their SUV campaign to potentially interested parties from the Moore-Smith cluster. 

Step 6

Implementation

With the new segmentation, tens of thousands of A/B tests rather than a single one are being carried out in the campaign, using a variety of criteria. The Refining Smart Data process allows to divide the data into smaller blocks - the so-called clusters, as well as to continuously refine and automatically evaluate them using artificial intelligence algorithms. 

For their pilot project, the car manufacturer may decide to launch a campaign on Facebook, which in the course of the project will be monitored and optimised. 

Step 7

Visualisation

The AI-based advertising campaign provides insights that go much further than merely identifying whether an interested party clicked on Option A or Option B. Our car manufacturer may now not only determine the specific video ads that are performant for a specific customer profile, but also see the exact criteria of the customers highly likely to buy an SUV. 

The visualisation dashboards do not provide only the "conversion rates" or the “target group sales during the campaign”, the reason being that the campaign modeling revealed the consideration of the "sales of the control groups" as well as the "over-performance of the participants compared to standard segments" to also strongly impact on the success of the campaign. The company's marketing department may now easily compare individual activities (e.g. of the test group, the control group or the holdout group). 

Step 8

Monitoring

For the modeling, the minimum requirement is to calculate ranks within the potential target group - or, as in the case of our car manufacturer, to calculate the conversion probability within the newly defined customer segment. 

As part of the Facebook campaign, a target list with 100,000 customers has been created and divided into target groups according to scores. 

During the campaign, it may be established that the interested parties who - like Ms. Smith - are particularly interested in insurance benefits, go on to achieve a significantly higher score. 

Step 9

Check

Now is the moment to analyse and evaluate the model, i.e. the advertising campaign in our case. The scoring of the different target groups gets adjusted and refined again, according to the campaign results. 

What new insights may our car manufacturer gain from the campaign? 

For people like Ms. Smith, the interest in an SUV arises from an increased need for safety. People want to be safe on the roads as they grow older. With this knowledge, the automobile manufacturer is planning another campaign in cooperation with an insurance company. 

The Refining Smart Data process enables the company to automatically automatically generate and run campaigns faster and more precisely, as well as better tailored to the customer. The data-based modeling and detailed control of the campaigns give the company precise, measurable results”. The car manufacturer now has a profound insight into its own customer base and market dynamics. With the consistent continuation of the Refining Smart Data process, the company can Generate "knowledge from data" on an ongoing basis and thus secure competitive advantages. 

Step 2

Raw data

The insurance company has 4 million poste code records from the UK and Ireland with a granular risk potential for frost, wind damage, flood or burglary. 

Step 3

Customer signatures

At the next step, signatures may be generated related to the building type, age, value or cost of repairs. 

Until the present moment, the risk analysis at Savesta had been based on the experience of individual insurance brokers and their data management ability. Although Mr. Harrison attaches great importance to continuous training of his sales team, problems with new customers have been arising time and again, with the damage risk having been estimated as too low by the customer advisor, resulting in higher costs for the insurance company. 

Step 4

Statistics

Using a matrix combining the signatures with the postcode areas, policies can now be calculated according to the regional risks. 

With these fully automated calculations, Mr. Harrison may be able to minimise the error rate in the risk analyses conducted by his sales team. 

Step 5

Modeling

The modeling is carried out by the insurance company's actuary as standard. Supported by the data-driven risk analysis, the actuary can adapt the calculation of the policies to match the company’s desired results as well as optimise them sustainably. 

Step 6

Implementation

From the refined data, based on the data such as location and building type, a web-based premium-calculating app may now be developed. In the event of an increased risk potential, policies can automatically become rejected or subject to further individual approval. 

Step 7

Visualisation

The weekly reporting gives Mr. Harrison an overview of the number of app users in specific postcode areas, as well as the ability to compare an application filed by a customer against the policy being applied for. 

Step 8

Monitoring

Savesta’s in-house actuary is responsible for the monitoring of the campaign, as well as defining models. 

Step 9

Check

Project results get checked and, if necessary, further adjusted according to the risk database, in order to promote an optimisation of the premium calculation. Adjustments with the new data from the risk database may take place annually. 

Step 2

Raw data

Stock levels, sales (per week), forecasts and target prices are used as the basis for data analysis. In addition, prices from comparable products by competitors are being integrated into the database. 

Step 3

Customer signatures

After the raw data has been centralised (i.e. all relevant date linked and bundled up), an overview of the individual products and product groups may be created. These product signatures indicate trends across various corporate divisions, showing the way to specific solutions for sales optimisation.

Step 4

Statistics

Using model calculations, the company may now make assumptions about the impact of their price adjustments - relative to the market shares and margins. 

Step 5

Modeling

The model calculation reveals the company’s greatest potential harboured in the sale of product bundles. By sponsoring these bundles, high-demand products may be combined with high-stock products. 

Step 6

Implementation

The company’s sales partners may use bonuses to make the sale of the product bundles more attractive. By means of a targeted adjustment of the inventory, the company may become more profitable as well as meet the requirements of every party involved in the sales process. 

Step 7

Visualisation

Dashboards with a traffic light system display the model as well as its results, allowing an easy and simple analysis. These dashboards connect the sales to the price level (relative to the minimum price), the stocks to the manufacturer’s central warehouse as well as enable the product groups assessment.

Step 8

Monitoring

Monitoring based on a traffic light system is crucial in balancing of all relevant business areas. Taking individual parameters of product groups to account, it allows to accurately determine a specific parameter becoming leading, e.g. If a parameter enters the red traffic light zone. Every other parameter, e.g. price increase or decrease, an increase or decrease in the stocks, is subsequently monitored and managed according to these criteria.

Step 9

Check

In order to identify any acute or potential bottlenecks in advance, the modeling parameters may be adjusted quarterly. 

Step 2

Raw data

First, all relevant available data are being identified, e.g. the demographic data (i.e. the industry, location, employees), the sales figures for the last six years as well as the relevant contact details. In addition, current leads and the most recent sales and telemarketing contacts are being excluded. 

The data gets cleaned up so that any incorrectly entered or recorded data do not get included in any of the subsequent calculations, thus eliminating the potential errors and miscalculations.

Step 3

Customer signatures

Based on the raw data available, trends and market shares may now be determined in combination with internal segmentations (Class A, B, C). 

Step 4

Statistics

Using statistical methods and calculations, potentially successful models may be identified. 

Step 5

Modeling

A response model may be created for the telemarketing agency based on the database signatures. From the initial 250,000 potential interested parties’ data, approximately 150,000 addresses remain.

For each of the data on record, a score may now be calculated, allowing the client Ito focus their sales activities on high purchase probability clients.

Step 6

Implementation

The telemarketing agency performs the calls according to the latest response model available, reporting on the results and feeding them back into the database. 

Step 7

Visualisation

The results of the pilot project are now getting visualised in order to get a comprehensive overview of the key figures. The visualisation makes it easy for those key figures to be compared and evaluated.

Step 8

Monitoring

Modeling can be further improved by control groups assessment. This is how a simple response model can become a predictive incremental response model.

This extended model helps predict the likelihood of the success of future actions. For example, it is possible to precisely determine the moment when a cross-selling campaign has passed its peak. 

Step 9

Check

Once the campaign finishes, it may be examined and refined using the Refining Smart Data process, suggesting alternatives where necessary, based on the current campaign outcomes.