Data Played Better

implementation complexity level low

Need / Considerations

The client identified that their loyalty levels were dropping and that they were missing repeat business sales on existing customers.  So how could they be pro-active towards this problem asap

The client had limited internal resources within insights & prediction.  Furthermore, they had a CRM system, that is difficult to extract data from and implement changes to their customer flows

Solution

The client provided us with the customer file that included all existing customers, along with a flag on identified churn Y/N; this data was run through our Predict, that analysed the data sets and scored the non-churned customers against the likehood that they will churn in the near future. This output was then fed directly into the CRM system and used to segment the customers so that optimised pro-active loyalty based campaigns could be run

implementation complexity level high

Need / Considerations

The client had a running application risk scorecard that was in need of a review and update, due to the fact that the performance was dropping

This current scorecard utilised only internal based variables, that were collected during the application process

Solution

We ran a performance review on the existing model, and then followed this by running all customer data against our Predict ML suite, with the results showing a number of significant improvements in the risk prediction on the portfolio when we utilised our 3^rd party data. The results proved that the external data improved the model by 40%

implementation complexity level low

Need / Considerations

Based on standard DM selection criteria, the client was struggling to locate all available consumers in their market areas across Denmark. The client has a very specific target market, due to the nature of their product offerings – notably older generation

The client wanted to optimise their 1-2-1 DM communication, by increasing their potential DM selection, but maintain the target fulfillment goal of hitting a minimum of 80% likelihood

Solution

We utilised our ML engine to create an expert model, that increased their potential DM universe from 260.000 addresses to 451.000 addresses; whilst maintaining the minimum fulfillment of 80%.  The ML models blended our data with the customer's internal data and then mirrored this onto all Danish addresses

implementation complexity level medium

Need / Considerations

The client had created a comprehensive architype segmentation model, however, they were struggling to match their four architypes back to their full consumer base.  They had tried to recreate the arhcitype algoritms on their customers but the match rates were not satisifcatory (approx. match rates 35%)

Solution

The client provided us with their customer data, the survey results used to build the acrhitypes, the respondants' addresses, and the architype algorithms; using ML techniques (notably random forest models), blending our data, we were able to recreate the algorithm results and mirror this directly to all their customers; with an overall  match rate and assignment liklehood = 75%

Furthermore, the models are continuously reviewed each year to ensure that this assignment level is maintained and, where possible, increased

implementation complexity level medium

Need / Considerations

Previous, the client was trying to cross-sell consumer loans to their whole customer base, using a one-size fits all model with a focus around the product and not the customer’s need – this was resulting in low conversation rates

Solution

Through predictive modelling (blending the internal customer data with the 3^rd party data in our data lake) we created a recomendation code that was focused on cross-selling the product. The code was created by analysing the customers' needs and current situation linked back to the optimised model-driven selection. Combining the code seelction with the customer segment characteristics in  dynamic communication process, the new appraoch increased the conversation rates from 0,5% to 3,0%, resulting in a 1,2m DKK uplift in new sales, compared to the old approach

implementation complexity level medium

Need / Considerations

The client wanted to optimise their pricing model on one of their products for new customer sales, factoring in levels of risk related to claims likelihood. However, they only had limited data points and claim volumes internal to work with, so it was extremely difficult to use this information alone to create the new pricing model that would service the whole of their market

Solution

In collaboration with the client we created a shared project that combined the customer’s risk model and data; with our data components that in turn was fed into our ML engines. The model outputs were then further tested on the actual internal claims and then fine-tuned accordingly.  These ML models were then implemented onto a full market view, so that each time the client engages in a new customer sale, they are able to clearly predict the potential risk at point of sale and assign the relevant pricing

implementation complexity level low

Need / Considerations

The client wanted to check that their internal churn models (based on internal data only) was fit for purpose – i.e. was predicting the right cases

Solution

We took the client’s customer base and ran it against our data lake and ML prediction tools – to segment on their churn level. The result was that we identified five external variables that would improve the client’s model by 8% prediction

implementation complexity level low

Need / Considerations

The client had a wish to start using predictive modelling in their business activities; however, they need to know if the business case would hold up in advance; they were struggling to get started as access to their internal transactional data was a challenge, without the need of involving their IT team in a large level project.  The client reached out to us to see if they could get started by using a limited subset of internal data and blending this with external 3^rd party data

Solution

By accessing our data lake, we were able to fasttrack the client in their predictive modelling journey; meaning that by blending the limited internal data with the 3^rd party data sets, we could create a predictive model that could be implemented quickly – the uplift in the deployment helped to prove the internal business case was very strong; and as such allow for the next level of investment – i.e. extract and utilise more internally data in new models and areas across the organisation

implementation complexity level high

Need / Considerations

The client had a wish to implement a completely new segmentation model on their whole customer base; the model that was in place at the time was based solely around the customer’s existing engagement but did not factor in their customers’ needs (current and future); it was a one-size-fits-all model and was not allowing for customer specific interactions, plans and campaigning

Solution

By engaging in a detailed segmentation and predictive modelling, we created a complex micro segmentation model that utilise both the customer's internal data, blending with our 3^rd party data from our data lake. On top of this model, we created a suite of recommendation codes, utilising our ML engines that are assigned to the customer base and refreshed monthly, allowing for ongoing slicing and individual selections based on prioritisations driven by the various prediction models. This whole approach has enabled the client to implement customer journeys that are specific to each segment, which has increased loyalty, and enabled significant increases in cross-selling

implementation complexity level medium

Need / Considerations

The e-commerce client was experiencing a low response rate on their campaigns; as the current campaigning concept was to run one to many approach sending the same offer to all – i.e. a scattergun approach.  They identified that a change was needed but where not sure on the best approach

Solution

By utilising RFM prediction models, we were able to identify best value customers within the existing base, and generate an RFM score. This new score and insight was fed into a market twinning tool (machine learning algorithm), that segmented the potential in the market based on the client's ‘best customers’.  Based on this, a new optimised ‘new biz’ campaign was run that resulted in a decrease in ”cost per lead” (CPL), compared to previous campaigns

Implementation complexity level medium

Need / Considerations

The client (a retailer) identified that their existing re-activation strategy was not resulting in the levels of re-activation they wanted.  They had a level of segmentation approach but this was focussed solely on e-mail communication channels and a frequency driven by the average re-purchase interval for all their customers

Solution

We analysed the client’s customer data; blending it with our 3^rd party data, the result was used to give key insights into the re-purchase trends on the clients' customers.  Based on this insight, we created a predictive model that factored in both the customer’s potential ($) and the frequency of re-purchasing; the score outcomes from this were then ranked and fed into their marketing automation system.  The ranking was then used to drive personalised customer specific communication – which in turn helped increase the overall win-back levels, and also helped identifying churn