After a deep dive into the inner workings of one class of prescriptive analytics—recommender system (personalization engines), it’s time to step back and explore how we can leverage prescriptive analytics in general. Today, I am going to outline 3 relevant use cases:
These are very distinctive use cases, where the objectives being optimized are drastically different. I hope these examples will not only demonstrate the power and versatility of prescriptive analytics, but also give you a better understanding of how they work. (If they don’t please feel free to let me know!)
Optimization of Business Process and Operation
I’ve discussed before that the simplest example of prescriptive analytics is a GPS, which operates in the geospatial domain. In general, prescriptive analytics are not limited to geospatial optimization. We can optimize processes and other parts of business operations. One of the most common use cases of prescriptive analytics is business optimization. Since the underlying computation of prescriptive analytics is already an optimization of some objective, this application is very natural. In this case, the objective we optimize is typically the efficiency or throughput of the process.
Whether you are optimizing a business process in marketing, sales, or customer service, you must tell the prescriptive analytics system what you are trying to achieve. This is like telling the GPS where you are going. For example, increase conversion by 10%, increase sales by 20%, or increase your net promoter score (NPS) by 5 points. These are the goals, or “destinations” you are trying to reach in a non-geospatial domain.
Subsequently, the prescriptive analytics system would prescribe a sequence of actions that lead to the corresponding business outcomes you want (i.e. increase conversion by 10%, sales by 20%, or NPS by 5 points). For example, to achieve 10% conversion lift, the system may prescribe reducing the frequency of your email marketing by 35%; simultaneously increase your real-time social media engagement by 30%; and when your real-time engagement reaches 15%, start directing people to your customer community for peer-to-peer engagement and recommendation. These are like the turns that your GPS system advises you during the journey, except these directions are not in the geospatial domain.
Optimizing Business Impact of Your Social Media Engagement
Now you know what prescriptive analytics can do for your business, let’s dive deeper on how it prescribes actions. The key lies in the objective that’s being optimized. Let’s examine a couple of examples from social media.
These days, social media is not new to business, but many enterprises still struggle to figure out how best to leverage it. Because there are so many different things you can do on social media, it’s hard to determine where to best allocate your limited resources (e.g. time, money, energy, etc.). Should you create more YouTube videos or should you use Snapchat? Should you publish more blogs, or participate more in the Q&A section of your community? The social media landscape is complex with thousands of social channels. Even within a single social channels, there are probably a handful of actions you can take. Take Twitter for example. It’s probably one of the simplest social channels out there, but you can already engage in many different ways: tweet a message, reply to one, retweet it, favorite it, follow someone, or simply read the tweets coming out of the firehose. This gives rise to many different social metrics that quantify how you engaged on social media.
Prescriptive analytics can help you focus on what you should do to achieve the biggest impact, but you must tell it what kind of impact you are looking for. Whether it’s increasing marketing conversion, sales, or customer satisfaction (CSAT), you should be able to measure the impact you want to drive. This is typically a key performance indicator (KPI) for your business. Once you have the KPI that you are trying to drive, and the social media metrics that describe how you engage, it’s relatively trivial to perform a time-lagged cross correlation analysis to see which social metrics have the strongest correlation with your KPI. Although correlation is better than flying blind, it would be better if we could establish causation as well. But this would require less trivial methods in statistics or econometrics (e.g. instrumental variables). If you are comfortable with these advanced techniques, you can even establish causal relationship and identify the strongest causal predictors for your KPI.
The correlation strength (causal or not) between the social metrics and KPI is the objective. Upon maximizing this objective, the prescriptive analytics system would be able to prescribe the actions that are most effective at driving your KPI. This helps you focus your effort on a few actions that will give you greatest impact (as measured by the KPI). And if you choose a different KPI, the system will prescribe a different set of predictors that maximize this objective.
When to Post on Social Media—Engagement Time Optimization
Because social media engagement is voluntary, people can participate anytime they want. But when is the best time to participate? The answer really depends on what you are trying to achieve. The goal of most social media participation is usually to reach the widest audience (whether that’s for a brand or for us personally). Even when you are posting a question and want the fastest answer or the most accurate answer, these goals can often be achieved indirectly by maximizing your reach. By reaching the widest audience, you increase your odds of reaching someone who can respond immediately; and by reaching the widest audience, you also increase your odds of finding someone who has the expertise to address your question accurately.
There have been numerous studies on when is the optimal time to post on social media, and many infographics provide general guidance on when to post on various social channels. In general, these studies are rather limited because the data is highly aggregated, the sample size small, and the methodology not rigorous enough. We recognized that there really isn’t a universal best time to post on social media, because the best time to post is ultimately specific and unique to the individual.
Our brilliant data science team (humble brag!) have analyzed over a billion posted messages and observed reactions and found that the best time to post depends strongly on your specific audience’s engagement profile. Meaning when your specific audience is most actively participating on a particular social channel. Keep in mind that it is your audience’s behavior (i.e. how they choose to use social media) that determines the optimal time to post for you. Since we all have different audience on social media (e.g. different followers, friends, connections, etc.), the best-time-to-post for you may be totally different from the best-time-to-post for me. This is precisely how Lithium Reach is able to recommend the best-time-to-post that is hyper-personalized to a person or a brand.
Now, I could certainly tell you more about the product, but that’s not my style. However, if you have specific product related questions, I’d be happy to discuss, or invite more qualified product management staff to chime in.
We further validated the optimality of our hyper-personalized recommendation on a sample of half million active users and more than 25 million messages observed over a 56-days period. We found that our individually optimized post-time leads to an average of +17% engagement lift on Facebook and +4% on Twitter. This is a VERY conservative average. In practice, we have seen more than +50% engagement lift at times from brands using Lithium Reach’s recommended time-to-post feature.
We discussed 3 business use cases of prescriptive analytics. In each case some objectives are optimized, so the system can prescribe a few actions (or sequence of actions) out of infinite number of possible action you can take.
While optimization is something that computers can do very efficiently, doing this at the individual level is still a challenging big data problem. Because people’s social interactions change continuously, we must constantly re-optimize as the new data arrive from the respective social streams. Fortunately, our data scientists have done all the heavy lifting and built this hyper-personalized post time recommendation algorithm into Lithium Reach. Now, you can get the benefit of personalized recommendation on when to post without needing to routinely analyze billions of message and their reactions.
As you can see, prescriptive analytics is versatile and powerful. It has just as many applications in science and engineering as in business and social media. Next time I’ll share a somewhat esoteric use case of prescriptive analytics in data science. Yes, data scientists like me also use prescriptive analytics. Come back next time to find out more!
Michael Wu, Ph.D. is Lithium's Chief Scientist. His research includes: deriving insights from big data, understanding the behavioral economics of gamification, engaging + finding true social media influencers, developing predictive + actionable social analytics algorithms, social CRM, and using cyber anthropology + social network analysis to unravel the collective dynamics of communities + social networks.
Michael was voted a 2010 Influential Leader by CRM Magazine for his work on predictive social analytics + its application to Social CRM. He's a blogger on Lithosphere, and you can follow him @mich8elwu or Google+.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.