Data infrastructure is a core element of any ecommerce business that often gets stuck at its bottom level: descriptive analytics. Advanced predictive and perspective analytics are instead the two directions you need to adopt in order to truly prepare for the future of your online store.
Unfortunately, many decision-makers still rely on simple pie and bar charts to make their own predictions. They see a decrease in sales, they automatically assume they’re not doing marketing correctly. They notice their competitors are raising their prices, they do the very same thing.
The problem with these decisions is that they don’t take the future into account. Market trends and customer demands change so often that simply looking at past reports won’t provide the best solution to your business problems.
That’s exactly why you need to understand the difference between descriptive, predictive, and prescriptive analytics and why you want to go through all three steps before making any kind of business change.
Moving beyond descriptive analytics
Data goes through our decision-making systems through a pipeline. That’s why all organizations have some form of descriptive analytics in their databases.
Descriptive analytics refers to the raw data you use to interpret past results and offer as input for a predictive analytics technique.
All types of advanced analytics that help you predict the future and prepare an action plan start with this very basic descriptive form. In other words, without descriptive analytics, you can’t progress through every step of your data infrastructure until you eventually reach the prescriptive level.
Descriptive analytics include charts, sales figures, percents, summary metrics, trends, intervals and tests, even clustering. While all this is incredibly useful if you want a better grasp of stats on business pillars such as sales, finance, or customers, you don’t want to rely solely on descriptive analytics to make a decision.
Take any item’s price that’s been steadily rising over the past few years. Is this a clear indicator of its future pricing?
With descriptive analytics not taking into account any potential risks or trends, the answer is certain: No. Past performance simply can’t predict what the future will bring. That’s where advanced data analytics come in.
What is Predictive Analytics?
While we naturally make future predictions based on what happened in the past, predictive analytics focuses on what could happen in the future. From forecasting and regression models with an explicit prediction role, predictive analytics creates a type of digital simulation that reflects the real world.
Taking this into account, it’s clear to see why descriptive analytics is a step you can’t miss. You can’t predict the future if you have no idea what happened in the past. Doing this would mean you’re essentially making a decision without any back data to begin with.
Within this predictive model, you can simulate future behavior, trends, and changes that could affect your decisions in the long run. Kind of like getting access to data you don’t really have and making a very good guess. Most supervised machine and deep learning algorithms fall under this category due to their ability to create parallel systems that mimic reality.
For these reasons, predictive analytics is a go-to option if you want to test systems and decisions safely, without them affecting your business in real life. In general, it’s currently used for designing cities, roads, space travel, and even flight training. From a business standpoint, and what’s of interest to you in ecommerce, you can use predictive analytics to predict shopper behavior, identify sales patterns, and so many more use cases we’ll get to later on in the article.
The results of predictive analytics literally tells you what’s going to happen in the future. You can see the progress of your sales, profits, and clients in time. This way you’ll no longer make unfounded decisions that can take your ecommerce business down the wrong path.
What is Prescriptive Analytics?
Prescriptive analytics tells you what you should do after the data analysis with the information you got during the predictive stage. This is done through a variety of scenarios you can use to control risks and choose the best action plan. Note this is still an emerging data analytics technique as it encompasses the predictive nature of its previous stage.
The future of how we use data lies in prescriptive analytics as it allows us to turn to machines for “advice” or “prescriptions”, as the name implies. Simply put, a prescriptive system replaces part of the decision-making process by providing a series of potential action paths.
The human input within a prescriptive analytics system is limited and can be left aside if the decision process is programmed beforehand. Just like with the predictive system, a prescriptive analytics method doesn’t fully act without human intervention. You have to add in the criteria that guide the machine towards concluding what a good outcome looks like compared to an unfavorable one.
Even more advanced prescriptive-analytics machines take it to the next level and execute upon our instructions based on the courses of action they identify. The current most advanced prescriptive-based systems are self-driving cars that are able to pick appropriate speeds and even avoid obstacles.
While they’re not yet able to make complex decisions in retail or ecommerce, they can choose where to send information, how to store results, and other similar administrative tasks you’d want to automate.
Prescriptive analytics also ensures the knowledge sharing between an online store and its brick-and-mortar counterpart is done effectively. By doing this, both instances will be aware of each other’s trends and sales stats. If an upselling hack works for the ecommerce store, it will likely bring in similar positive results for the physical store as well.
Predictive vs. Prescriptive Analytics: Core Differences
To make sure you never confuse the two terms despite their common forward orientation and use of historical data, remember the following fundamental differences between predictive and prescriptive analytics:
|Predictive Analytics||Prescriptive Analytics|
|Tells you what will happen in the future||Advises you on how to make things happen|
|Focused on the result||Focused on an end action|
|Uses historical data||Uses a predictive model|
|Makes use of statistical models||Is based on optimization models|
|Creates non-actionable outputs||Creates actionable outputs|
How Prescriptive and Predictive Analytics Change Ecommerce
What if I told you that using prescriptive and predictive analytics during the decision-making process can tell you just how your customers think?
This is not far from the truth.
You’ll know how potential clients will behave on your online store, what kinds of promotion they’ll be most responsive to, and even what needs they’ll have in the upcoming months. No wonder that companies such as Macy’s have been using predictive analytics as early as 2014 and have seen results from the very first months.
Briefly, all of this data [or rather “these answers”] can help you handle other potential issues such as how you’re going to manage production, profits, or delivery.
Here are a couple of extra benefits of advanced analytics to keep in mind:
- Improve the value of a shopping cart and upsell items. A predictive system takes into account past purchase behavior and categorizes customers based on similar segments. Once a potential buyer who is part of a defined shopper category, the predictive system tells you which products you should recommend to a person in order to sell more. Ecommerce corporations like Amazon and eBay are already using predictive analytics to find out how shoppers think and are capitalizing on this technology to recommend the right products to them.
- Boost your sales and revenue by offering better pricing and selling the right products. In many instances, two people belonging to a shared customer segment can pay a different price for the same item. Smart analytics platforms let you tweak your pricing strategy by offering the proper price and discounts. This is done based on how often a shopper views a product and even if they leave it in their cart for a couple of days. Some stores also ask for a zip code before you browse so they can show higher prices to people based in high-income areas.
- Help you make better and more informed decisions. Whether we’re talking about a marketing, supply, logistic, or customer level, predictive analytics, in particular, can help you answer questions such as: How many products are you going to sell? What’s the best social media network to promote your items on? Where do your customers spend most of their time? The suggested solutions will also help you reduce the trial and error occurrences by opting for the right solutions straight away.
- Speed up your work. From creating better reports to tracking KPIs and improving overall team efficiency, predictive and prescriptive analytics methods help you save time. And going through thousands of data points that are always changing in a timely manner is impossible for humans anyway. You’ll spend less time and effort on manual data analysis and can even have a smart system create the reports in your place. Through this, you can either reduce the number of your employees or redirect their workload towards tasks that require their creativity, a trait no robot will match up to.
Typical Uses of Predictive and Prescriptive Analytics
If you get creative, predictive and prescriptive analytics in ecommerce can be used to fix most issues you might have or worries regarding future sales development. Here are the most common uses:
Forecasting Sales and Pricing
Without a proper sales forecast plan in place, you’re bound to distribute your budget unevenly and incorrectly. Not to mention you’re putting your revenue at risk.
Predictive analytics makes use of existing data to estimate future demands and simulate how your sales will look like given a certain pricing strategy. This way, you’ll know exactly how to price your products, when to give out a discount, and when it’s okay to look into cross-selling and upselling without affecting your prospective revenue.
In addition to this, you’ll be able to apply smart pricing to your store, optimizing prices in real-time. This is done as the system takes into account customer preferences and actions, as well as the prices of your competitors and market demand.
Determining Market Needs and Customer Behavior
One of the core things any online store owner wants to know is what people will want to buy in the future and how the market is going to change. Predictive analytics can actually help you figure out if a product is worth selling, what will determine people to buy an item over another, and if there’s any upcoming trend you can catch on early on.
Besides, you can use all this information to craft better marketing strategies to grow your store and retain more customers. Email marketing, content publishing, and ad campaigns are just three of the digital marketing areas you’ll have to optimize towards an increased ROI and lower churn rate.
Automating Customer Service
In the past, everything a potential or existing customer would write to your support team had to be read by another human. This meant loads of hours the customer would have to wait until their inquiry got sent and answered by the right person.
Customer complaints and inquiries are instantly read by a prescriptive-based system and categorized according to its class. Then, the ticket can be automatically sent to a department or professional who’s best suited to fix the problem.
Using predictive analytics you’ll be able to predict transportation rates and even potential future constraints related to shipping, fleet management, procurements, and so much more. Time limitations and restrictions at a governmental or environmental level included.
Knowing this information beforehand allows you to prepare backup plans and reduce your costs while also avoiding unexpected loss of cash. Rough estimations you’d set in the past are no longer feasible as you’ll want to know exactly what’s going to happen next in order to keep your budgets in check.
Tools like Transmetrics are already helping logistic businesses take their cargo and fleet management data analysis beyond the basics. This is done through demand forecasting, data cleansing, and predictive optimization that relies on data science and AI.
Online shopping fraud is far from being something that’s easily avoidable. Cybersecurity too will always be an issue that could impact your ecommerce website at any given point. Predictive analytics, in particular, is used to predict criminal behavior, keep threats at bay, and ensure you’re prepared for any potential attacks by fixing vulnerabilities in time.
Tools like Signifyd are also making use of machine learning to help you manage chargebacks and get back any lost revenue that resulted from fraud attempts. The guarantee is a profit and conversion increase as shoppers will no longer put an end to their buying process when an order gets falsely declined.
MorphL Case Study and Other Advanced Analytics Applications in Ecommerce
One of the most frequently asked questions we get at MorphL is: now that we have these predictions, what do we do with them?
Indeed, just looking at those predictions in a dashboard doesn’t add any value to the business. They need to be integrated within the e-commerce product. They need to do something and trigger an action.
A lot of marketing platforms nowadays support automation – the basic goal is to deliver personalized messages to customers and leads and various platforms allow you to create a dynamic series of messages to send to your contacts.
The message a person receives is decided by factors you specify, like what their spending habits are, where they are in the buying process, and past interactions they’ve had with your site (descriptive analytics).
In the broad scope of things, marketing automation incorporates several different aspects of marketing and business development, including email marketing, content development, conversion rate optimization, and lead generation. Automation can be accomplished within a single platform (Hubspot, Klaviyo, Marketo) or across multiple apps (Zapier, IFTTT).
Most of these marketing automation tools rely on deterministic triggers (decisions and outcomes after the fact):
IF cart-abandoned = true THEN SEND email message
However, you can define an intelligent automation process by using probabilistic triggers to make decisions on any time horizon, from immediate to long term:
IF cart-abandoned-probability > 80% THEN SEND 10-percent-discount-code
This trigger clearly factors information about possible situations or scenarios, available resources, past performance, and current performance, and suggests a course of action or strategy. This forms the basis of a prescriptive process and all of our machine learning models can be used to orchestrate intelligent process automation for e-commerce.
A machine learning model, such as the shopping stage prediction model, takes into account dozens or even hundreds of features – from page views and session duration to days since the last session, search depth, and device category.
Once the shopping activity of an ecommerce website has been modeled, one could simulate the possible outcomes (conversion rate, average order value, etc.) by varying the input parameters and we might notice that by increasing the number of sessions with 5%, we get purchase probability of more than 75%.
And you can get access to this kind of insight prior to running a marketing campaign or investing in building a new search feature, making it more likely to be a good decision when presenting this idea to the upper management.
However, this decision was assisted by the predictive model that was previously built and not generated or recommended by it. If we reverse the logic however and start from the outcome: we want to increase the average order value by 25%, the prescriptive system might provide us with a few scenarios that lead to this outcome, one of which is increasing the number of searches with 5%.
The previous example doesn’t take into consideration the cost/benefit ratio which is mostly encountered when trying to identify the optimum price of a certain product – also known as price elasticity. Of course, this can be applied to other optimization problems such as stock optimization or budget optimization.
The hypothesis is simple:
Given the history of transactions for a given product that has been sold at various pricing points, identify the best price that yields the most revenue.
Without going into a lot of technical details, simply put, the optimization problem is solved by simulating thousands of scenarios on the predictive model that was generated based on the historical data.
This is where all of it comes together and the true nature of prescriptive analytics can be seen at work.
Where to Get Started with Predictive and Prescriptive Analytics
Descriptive analytics can no longer help you make an impact. In fact, they never did. We just didn’t have access to the advanced analytics we can make use of today.
The predictive vs. prescriptive analytics debate is no issue either. Simply because the two of them are complimentary. Yes, prescriptive analytics is a more advanced version of predictive analytics, relying on it just like the latter can’t exist without descriptive data. So the three types of analytics will always coexist:
- If you want to find out what’s currently happening to your data, turn to descriptive analytics.
- For seeing what the future holds, implement predictive analytics.
- Take it to the next point and have the machines tell you what an appropriate strategy looks like and opt for prescriptive analytics.
In all cases, you’ll need to move through each of the stages to get to the latter one and it all starts with a solid data set in place. The good news is you probably already have this data if you’ve launched your ecommerce shop a while ago either in a classic spreadsheet or cloud database, your Shopify analytics dashboard, or in Google Analytics.
While a couple of thousand data entries should be good enough to begin with, the more data you have, the more accurate your predictions get. Also consider getting access to external data sets on factors such as market trends and demands so you can benchmark your own data against these.