Where can Data Science be used in Customer Decisioning?

In the field of Customer Decisioning, a Data Scientist can support with analytics, modelling and the implementation of AI solutions. In a previous article, (Customer Decisioning: Why it’s Crucial to an Unbound Experience) we touched upon the benefits of treating each customer as a ‘segment of one’ and briefly mentioned how Data Science capabilities can further enhance the process. That’s absolutely the case; the power of the Data Scientist is the value added on top of existing solutions. In this section, we’ll expand on the above-mentioned point.

Normally, there would be a way for a business to establish a next-best-action:

1. Rules-based approach: This covers scorecards, segmentation or any rules-defined approach used to assess values and produce a result. It’s a simple to implement and transparent to communicate approach, however, it is hard to maintain in the future since there is a need for manual updates with every major change in the rules.
2. Adaptive modelling approach: Adaptive models, especially in an out-of-the-box environment, are incredibly useful because of their ease of setup and use. Based on the underlying algorithm, some don’t even require historical data and automatically update to adapt to new user actions. This is particularly useful for when new offers/products are launched and there is no data available to train the model. The scale and automation of these models makes them easy to build and maintain.

   For example, the adaptive models used by the decisioning brain mentioned above, use a Naïve-Bayes approach. This allows the model to assess the probability of a target event taking place by incorporating real-time decisions from the customer. When several hundred customers with a similar background do not engage with a certain type of ad, the models will then determine that product would not work well with that population and would direct a more suitable ad in future.

   Adaptive models are incredibly useful, as they help reflect the realities of content engagement; however, on the flip side, they will not be well suited for tasks where historical data or a more bespoke build is required i.e. a customer churn model. Adaptive accuracy is good enough to provide an indication of what the main leavers are to provide incremental results. Nevertheless, a more bespoke model will be more accurate, but, where appropriate, can work in conjunction with their adaptive cousins.

Both points are important tools in a decisioning framework. However, sometimes you’ll want a bespoke add-on - something that requires a broader understanding of your underlying customer data and that is more involved than the out-of-the-box solution.

For example, it could be a propensity model to predict which customers are likely to purchase an item. Perhaps that model needs to be a cog in an existing system and its output will feed into a different model. A simple example for the above case is a messaging prioritisation solution (adaptive model) that takes a propensity to purchase model (custom built predictive model) as a predictor. With that predictor it can decide whether to send a promotional email on said item or present specific discount to a customer to incentivise a purchase.

What’s the day-to-day role of a Data Scientist within Customer Decisioning?

Typically, a Data Scientist will build a model based on a certain specification and through the model lifecycle, they will perform monitoring and retraining activities. A model can be built using any major language in a widely transferable format locally or remotely on a cloud service (e.g. AWS, Google Cloud). A versatile Data Scientist can use a variety of tools and have quality checking in place to monitor and retrain a model when its performance drops.

It’s that focus on model governance, combined with the passion to see their models drive business activities and influence customer engagement, that make a Data Scientist stand out from their peers.

Another point briefly touched upon was AI. These days, AI feels less like a choice and more like a rite of passage to be considered tech-savvy. Nevertheless, the unknown nature of AI algorithms can pose a threat to data security and needs to be used with proper guidance when implemented, otherwise the output from models may produce pointless results and waste resources.

A Data Scientist can provide a clear picture of which sides of AI can be useful and how they fit with the current infrastructure. Is machine learning necessary when a simpler model can be more cost-effective and transparent? Do we need manual checks on incoming communications or can a Natural language model help sort comms onto the right people and save time. A Data Scientist can guide you through this journey.

Why is Data Science not more readily used?

Following the section above and with all the praise and patting on the back I’ve given to Data Scientists, job’s done. What else is there to say? It appears Data Science and sliced bread are on the same plane of existence. Happy days, we found the solution to all our analytic woes! It sounds too good to be true, doesn’t it?

Of course, the reality is a bit more complicated, so let’s not dethrone sliced bread just yet. If Data Science is that amazing, why isn’t it more prevalent when making decisions? We touched upon some of the reasons above.

Here are a few to consider:

1. Reliance on out-of-box solutions:
   Out-of-box solutions offer many useful resources and can make the decisioning journey simpler. Even Data Scientists benefit from them, since they can simplify model monitoring and retraining. However, there is only so much a platform can do and sometimes a bespoke solution is required. Remember the example above with the propensity model? There can be reluctance to go away from the platform because of invested interest, cost concerns or general unwillingness. All fair points but they do remove some of the possible solutions that can be developed. At Merkle we can ease the discussion of Data Science and work to see how it can complement any existing solutions.
2. “Blackbox” models are a red flag for some:
   A “blackbox” model is any model that uses an algorithm to solve a problem that is not easily explainable. Blackbox models have their pros: high precision, ease of set up and use. The problem arises when transparency comes into play. Not knowing how a certain solution was generated is an issue that turns some people away from Data Science. At Merkle we understand this concern and have experience approaching the creation of models by testing different versions and comparing performance and transparency.
3. Lack of subject matter expertise:
   All the definitions of Data Science and the areas of expertise that make it up contribute to the difficulty of forming an internal team. There are a range of abilities that are required, and a good Data Scientist is undoubtedly an asset, especially in a decisioning environment. There not only is the base expectation of modelling and analytics required, but also the knowledge of how decisioning works and what contributes to a next-best-action.

There are valid reasons to be cautious when approaching modelling and AI. We’ve all seen the headlines with data safety and transparency being on the top of the list for many clients. We understand that it’s hard to get out of established practices, especially when technology and guidelines are changing at a never-before-seen pace. We understand that a reliable partner is required on this journey to build trust and remove the complexity and fluff that’s currently surrounding Decisioning and Data Science.

Pulling everything together, what does it all mean?

In summary, brands are leaving commercial value and business benefit on the table by not utilising more Data Science capabilities within their customer decisioning platform. Data Science techniques are a key component when it comes to maximising the power of the single customer decisioning brain. Brands need to take advantage of modern advanced analytical capabilities to drive the very best omni-channel customer experience and create a long-term competitive advantage over their competitors.