To recap, we have covered three types of analytics already, each of which has its own characteristics, methodologies, and best practices:
- Descriptive analytics – summarizes what happened in the past
- Diagnostic analytics – attempts to determine why important events happened
- Predictive analytics – predicts new data points from exists data which would be difficult or impossible to get otherwise.
In some ways, prescriptive analytics achieves what marketers might expect from predictive analytics. That is, prescriptive analytics recommends what actions we should take in the future.
Before we begin, though, we’d like to let you know that Econsultancy runs an Advanced data analytics training course.
What is analytics?
In our previous posts on analytics, we defined analytics as a practice, a process, and a discipline whose purpose is to turn data into actionable insight.
With prescriptive analytics, the focus is as much on the action as on the insight.
Prescriptive analytics overview
Paradoxically, understanding the complex area of prescriptive analytics starts with a simple question. In marketing, why do we do what we do?
Dispensing with trivial answers to that question, most marketers do what they do because, relying on experience and reflection, they believe it is the right thing to do. For example, a food delivery service may notice that their customers tend to buy more often on weekends. A seasoned marketer would then recommend that the company offers discounts mid-week to boost business.
With the right data, marketers can automate the process of noticing a consumer’s behaviour and sending an offer. Additionally, they can look for other behavioural patterns which can also be influenced with appropriate actions.
Finding the right data and devising the rules to recommend appropriate actions is the essence of prescriptive analytics.
An example of prescriptive analytics
Perhaps the most well-known example of prescriptive analytics is a recommendation engine.
Recommendation engines use personal data sourced through descriptive analytics (e.g. pages viewed and items purchased) and predictive analytics (in-market segments or inferred demographics) to suggest products that a particular individual might be interested in.
For recommendation engines, prescriptive analytics means deciding which data to use and how to process that data using decision logic to produce a tangible action. For a recommendation engine that action is recommending the products.
The distinguishing features of prescriptive analytics
Analytics experts disagree about what distinguishes predictive from prescriptive. Both methods use past data and algorithms in order to make an educated guess about something which is not known.
For marketers, however, differences are arguable more apparent.
For example, one notable difference is the desired result of the practice. Predictive analytics may only discover new data whereas prescriptive analytics must produce a recommended action.
Prescriptive analytics may also deliver more than one recommended action. Should it do so, the algorithm should also be able to rank them, so the most appropriate is considered or viewed first.
Also, unlike predictive analytics data output, prescriptive recommendations must be tested against a desired business outcome. The results of a prescriptive algorithm cannot be ‘eyeballed’ like predictive data to determine whether it is correct. Recommendations, for example, can only be validated by analysing whether the consumer clicked on the recommended product.
Because of this condition, prescriptive analytics also require a feedback mechanism which validates recommendations and improves them over time. How to do this is beyond the scope of this post, but interested readers can find more info in one of the many books on ‘recommender systems’ or read: What’s the difference between AI-powered personalisation and more basic segmentation?
So, to make a clear distinction between predictive and prescriptive analytics, if a marketer is characterizing data, then they are doing predictive analytics and should focus on producing correct data.
If they are trying to deliver business value through recommendations, then they are doing prescriptive analytics and need to be more concerned with actual results such as conversions or revenue.
How to do prescriptive analytics
Engineer defining events
To get started with prescriptive analytics, review your website, app, or product and look for opportunities to leverage the platform to learn more about your customers.
For example, an ecommerce site which has multiple product categories could create mini-sites which not only featured related products, but some relevant content as well. So you now assume that someone visiting a toy-themed section of the site is clearly in-market for toys. Should they also read a post about infant care, then it is quite likely that they are a new parent.
Those who trigger the defining events then need to be categorized into segments so that they can receive the right recommendations.
Devise recommendations with a measurable action
Then, you need to identify an action which, if taken, will validate the predicted category. This could be making an offer, requesting personal information or suggesting content (which requires a click).
For the new parent mentioned above, the measurable action may be asking for personal data in return for a free sample of a product. For someone in market for a package holiday, offering various travel-related content, and measuring clicks, may be sufficient.
Implement feedback mechanisms
The results of the suggested measurable actions should then be reviewed to see whether your defining event is useful at categorising consumers.
That is, if relevant offers are not taken up, then your defining event is probably not creating an in-market segment for the product.
Measure business impact
In addition to correctly identifying prospects, prescriptive analytics should also drive profitable action.
For a publisher, those who are identified as in a particular segment should click on relevant ads more. Shoppers who are supposedly in-market should, ultimately, not just click on offers but should buy more.
While delivering business value may be a longer-term goal, thinking of how to measure return on investment (ROI) in your analytics at the start of the project is always recommended.
Review data at each stage to improve results
With prescriptive analytics, there are three points where you will have data to help you improve:
- Categorisation – are people engaging with your defining event?
- Validation – did those who engage also validate their category with the measurable action?
- Monetisation – are you able to encourage your prospects to do more business with you?
While there are no fixed rules for how effective each stage must be, using a control group which has not been exposed to the prescriptive analytics applied will give you a relative indication whether your programme has been successful.
Prescriptive analytics best practices
Start with a commercial tool
As prescriptive analytics is quite complex, it is best to start with a vendor so that you become familiar will the options available. With guidance, you will also be likely to have a better chance of producing successful recommendations and personalisations.
Adobe, IBM, and now Google have industry-leading products for brands with a sizable budget. For smaller marketing departments, there are countless independent personalisation and recommendation engine vendors who will typically provide services along with their products.
Provide the next-best-action, not just the next-best-offer
Marketers traditionally focus on the moment when customer is about to buy and, consequently, only deliver offers to encourage a purchase.
With digital marketing, however, marketers can do much more than that. As discussed in a previous post, data from digital marketing can also identify behaviour which indicates that a customer has hit an ‘inflection point’ in the buyer’s journey. That is, they are poised to move from, say, just being aware of a need to being interested in a particular product.
The wider opportunity, then, for prescriptive analytics is to find these inflection points and provide information which moves a consumer along the journey in the brand’s favor, also known as next-best-action marketing.
The next-best-action may be additional product information, an opportunity to speak with a subject matter expert, or even a strategically-placed ‘buy now’ button to circumvent a normally long buying cycle.
Try many strategies
One issue which marketers have with prescriptive analytics is that they are motivated to use for a particular product line or customer journey. Which recommendations will work for customers, however, is difficult to determine from the outset.
For that reason, many different approaches should be tried before deciding on prescriptive analytics as a strategy. For example, a brand may find that website personalisation and recommendations don’t work, but that dynamic audience segmenting delivers superior return on ad spend (ROAS).
Through this four-part series, we have covered a broad range of analytics. We started with the simplest and most commonly-used analytics, descriptive, and ended here with perhaps one of the most advanced topics in marketing, prescriptive analytics.
Through numerous examples, how-to guides, and best practices we hope that readers now have a better idea of what is meant by the term ‘analytics’ and the goals of each methodology.
The most important takeaway, though, is that marketers should now to be able to distinguish each type of analytics and ensure that they follow the appropriate guidelines.
Descriptive analytics needs to provide clear and simple representation of complex data. Diagnostic analytics should produce a root cause or short list of contributing factors to an issue. Predictive analytics aims to deliver an algorithm which produces high-quality new data points. And prescriptive analytics must produce both a recommendation which drives profitable action and validation method.
Regardless of which analytics you use, though, there are countless of additional books, reference guides, and blog posts to help you on your way. A few which were used in writing these posts are listed below.
- Business Analytics A Practitioner’s Guide
- A User’s Guide to Business Analytics
- The Four Realms of Analytics, Tim Vlamis
- Big Data Reduction (Parts 1,2, and 3), Michael Wu, Chief Scientist at Lithium Technologies