A few years ago (six to be precise), Amazon patented something called “anticipatory shipping.” The idea was to move products in the direction of specific consumers, based on a specific set of business rules, even before the consumers purchased them. The benefit to Amazon: reduced shipping and fulfillment costs. The benefit to the consumer: shorter wait times between order and purchase.
Amazon planned to do this through machine learning (ML), whereby it would predict what the customer is most likely to buy and when, and then ship the item to them even before they buy it. Of course, if they shipped an item that the customer did not buy or want, the customer would simply refuse the package. The lost revenue on those items would be more than made up through the purchases that customers ended up keeping, as well as through increased customer satisfaction and lowered shipping costs. Win-win. This, in a nutshell, is prescriptive analytics.
For a long time, the field of data and analytics was focused on describing what happened — how many customers bought the product, what they looked like, how many came back, etc. It was like looking in the rear-view mirror — interesting, but limited in effectiveness. This evolved into predictive analytics, where the focus was on predicting what consumers would do next. This, of course, had a much higher business value.
You could anticipate what consumers were likely to do, and take action to influence the outcome. This is what teams of analysts, paired with their business counterparts, have spent the better part of the last decade doing. An action was chosen from a set of options, based on business rules. Effects of that action were observed and fed back into the data to make the next prediction more accurate. And so on.
With the advent of advanced ML algorithms, analytics has now entered the prescriptive phase. This means taking a set of data and predicting the most likely outcomes based on a given set of (future) actions and then recommending the most beneficial course. The machine doesn’t just predict what will happen. It tells you what to do about it.
Imagine that you are a credit card portfolio manager. In the past, you ran a predictive analysis on your customers to understand which ones are most likely to churn, and based on the results, you put in place retention strategies. The choice of those strategies was left to you, and as a human, you had a limited set to choose from, with a rough understanding of the overall business outcome.
Today, prescriptive analytics can tell you not only which customers are likely to attrite, but also the best course of action for each attriting customer. Some may stay because of a zero percent interest offer, others may likely be enticed by a balance transfer offer, and some others are best left alone to attrite. The machine runs the analysis for you and comes back with the best course of action for each customer (or cohort of customers) with the overall objective of maximizing portfolio value.
While prescriptive analytics can be thought of as an evolution of predictive analytics — which it is — it holds a much bigger potential to transform how businesses interact with their customers. While predictions are about doing “business @ speed of thought” (thanks, Bill Gates), prescriptions are about doing business at faster than the speed of thought! A report from Global Industry Analysts projects that the market for prescriptive analytics will reach $1.6 billion by 2022.
Nine actions you can take to reap the benefits of prescriptive analytics:
1. Ensure organizational readiness
Prescriptive analytics is about a complete approach to doing business. As decision-making shifts further away from traditional organizational silos, line owners see it as an erosion of their decision-making power. And while that is true to an extent, it is imperative that they see prescriptive analytics merely as a step up in the value chain rather than as a replacement for what they do. When machines take care of decision-making for certain aspects, humans are freed up to focus on tasks that are of greater value. And that is a shift that requires a mindset change and significant training and commitment from the team.
2. Secure executive buy-in
As with any major initiative, executive buy-in is key. Your organization’s top execs need to be convinced that this is a good idea and be ready to assume some of the risks that come along with it. Executive buy-in is also critical to prioritizing the areas of the business where you would like to pilot prescriptive analytics initiatives based on risk and return.
3. Identify your priorities
Ask yourself, and your team -what are the key business priorities? What areas will benefit from some quick decision-making, leading to immediate business impact? Contrast this with the risk to the organization of making incorrect decisions (no technology is perfect, and there is a definite learning curve). Ideally, a 2×2 matrix between risk and reward can make understanding the trade-offs much easier.
Perhaps the low point in your business process is lead quality, which makes sales development your current weakest link but your greatest future opportunity. Extend that opportunity to a business case: Accurately identifying which prospects are likely to convert in the next 90 days will increase sales development efficiency and overall sales. Then using prescriptive analytics techniques, identify the key actions that you need to take over the set of customers, holding out a group of customers as a control.
On the customer side, let’s say your CX priority is quicker customer service resolution. Here, a potential use case for prescriptive analytics would be anticipating a known maintenance issue that will likely arise in the next 90 days, which would provide opportunities to train support staff or order necessary parts in advance to shorten the time needed to resolve it.
4. Test, learn, retest, relearn
Having identified your key priorities and having run the test, generate a library of learnings. How well did the machine predict the outcomes of the prescribed actions? How often did it get it right? How often was it wrong? Feedback is critical to learning and making improvements, so you need to make sure you are capturing what you learn in a disciplined fashion and implementing the insights the next time around.
5. Ensure execution and operational capabilities
Having the operational and executional capabilities to implement the output of a prescriptive analytics program is fundamental. Let’s say you identify five segments of varying spend over the next 12 months and need to deliver variable offers through direct mail, site personalization, and email campaigns that reflect prospects’ site engagement and trade show participation. Do you have the channel management capabilities and expertise to deploy such programs? Focus on what you can do today in goal setting, priorities, and execution capabilities, and plan for increasing these abilities in the future.
6. Remember that data comes before science in “data science”
You wouldn’t hire a master chef, deprive her of ingredients, and expect a five-star meal to materialize — so don’t think that hiring data scientists is the golden ticket to prescriptive analytics. Prescriptive analytics is dependent on data: clean, accurate, integrated, complete, and accessible data.
Before discussing science and algorithms, ensure you have a solid foundation of data: firmographic and demographic data at a minimum, plus digital engagement data (including content consumption and behavior). Are those data sources up to date? Is the format and taxonomy consistent? Are there any gaps in the data set? Can you integrate them to create robust customer and prospect profiles?
Anonymous, disjointed behavioral data doesn’t help identify causal relationships. Choose your data sources and evaluate their readiness to create an action plan to remediate issues and/or obtain missing data. Lay a solid foundation of robust enterprise data management and governance upon which the science can evolve and advance.
7. Implement a data-layer-centric martech architecture
Simply put, a data-layer-centric architecture is one where disparate marketing platforms — like web analytics, marketing automation, email, CRM, and ecommerce — expose and share data with one another via JavaScript code.
This approach allows real-time measurement and personalization across marketing apps. And because it joins disparate user IDs across platforms, the approach also becomes central to creating a single unified customer profile across paid, owned, and earned media. This data architecture can fuel smart CX personalization to provide a smoother, individualized customer journey.
Another benefit: Data joins are made and managed in the data layer rather than in custom integrations, minimizing the effort and disruption in replacing or supplementing martech platforms and creating a much more flexible, future-proof architecture. This allows you to stay ahead of new technology, prevent lapses in personalization or data retrieval, and optimize for the customer experience.
8. Define and implement a data storage and hosting strategy
The companion to a data-layer-centric architecture is big data cloud storage. All that digital data being integrated via the data layer needs to be captured and stored for reporting, analysis, and modeling, with the joins and relationships intact. Cloud and private cloud solutions provide a scalable, cost-effective storage and management solution, often in the form of data lakes. They also offer the opportunity to integrate offline data sources (such as POS, trade shows, telematics, service and support, call center, and so on) with your unified customer profile.
9. Identify and choose the right analytics platforms
With your data strategy in place, choose an operational model or solution that matches your needs, use cases, and ability to execute. The market offers a variety of modeling solutions to serve enterprise, midsize, and smaller organizations. These options can leverage programming and non-programming data science teams, citizen data scientists, non-data scientists, automation, and customization to bring the capabilities and promise of prescriptive analytics to everyone.
Bottom line
Reducing risky “what ifs” is just the start for big data. Prescriptive analytics is already a promising frontier in big data, but even more exciting is the potential that dynamic, AI-powered decisions have to streamline the customer journey, create meaningful moments, and boost overall business performance.
