Have you ever dreamed about learning which products your customers are most likely to buy in advance?
How great would it be if you could determine the highest price a customer would pay for a product? What if you could optimize customer service to resolve concerns proactively before they become issues?
Do you think this advanced insight would allow you to make even more money from your ecommerce efforts?
Predictive analytics is making all these dreams a reality by offering solutions for seven key areas of ecommerce.
1. Improve customer engagement and increase revenue.
Different customers engage with a retail site in different ways. Predictive analytics looks at all the variables to generate the desired engagement from the customer. This could mean signing up for a newsletter, clicking on a promotion, or some other form of engagement.
Lattice has researched how leading companies like Amazon and Netflix are using predictive analytics to better understand customer behavior to help sales professionals better qualify their leads.
Investors in this space understand the market opportunity as predictive tools help inside sales reps better score leads by searching all publicly available information, then matching that against the qualities of a company’s existing customer base.
It’s because of predictive tools that Dell was able to send 50% fewer leads but get nearly double the results.
In another example, an undisclosed education portal used by one-third of college-bound seniors used a predictive advertising system to better match promotional offers to their existing traffic. The result was “a 25% increase in response rate beyond that generated by the existing system, which translates into an estimated $1 million of ad revenue every 19 months.”
2. Launch promotions that better targeted your customers.
Promotions are a must-have for a retail business to succeed, but it’s not easy to get them right.
According to a study by Oracle, 98% of fast-growing merchants feel that segmentation and targeting are important for their online merchandising strategy, yet more than half aren’t satisfied with the tools they have for promotions.
Predictive analytics changes that by correlating data from multiple sources to determine a personalized promotion that works for a customer or segment.
Macy’s has seen the benefits of predictive analytics by deploying a solution from SAP that better targets registered users of their website. Within three months, Macy’s saw an 8–12% increase in online sales by mining browsing behavior within product categories and sending targeted emails to each customer segment.
StitchFix is another retailer that has a unique sales model, asking users to take a style survey then using predictive analytics to match customers with the clothes they might like. If the customer doesn’t like the clothes they receive, they can return them for free return shipping.
Another example (outside the retail world) is Turkcell, Turkey’s largest mobile carrier. Using more than 150 data points on their customers, such as usage patterns, device preferences, and location data, they deliver more relevant promotions in real-time and reduce their churn.
However, it’s important to realize that predictive analytics tools are not a plug-and-play solution in which you dump data in and revenue spills out. According to a research report by Ventana, only 13% of the 2,600 businesses they studied considered predictive analytics a critical element of their business intelligence strategy.
David Menninger, former director of research for Ventana, theorizes why this might be:
Predictive analytics remain a specialist tool […] Personally, I think it is very hard, and the math involved is beyond the capabilities of many people or their training today.
In another article, author Robert T. Mitchell interviewed experts from three analytics consulting firms to find the most common mistakes when deploying predictive analytics solutions:
- Not having end goals in mind;
- Not removing junk data (due to lack of understanding);
- Having a company culture that’s unwilling to test changes.
Dean Abbot of Abbot Analytics shared the story of a debt collection agency that wanted to determine the optimal sequence of actions for collecting a debt but adhered to a strict set of rules required of their collectors.
Data mining is the art of making comparisons; you need historical examples.
Fortunately, most of the experts interviewed agree that even though flawed prediction models are rarely fatal and can always be improved upon, the consensus is that building good models is a lot of work and can take a significant amount of time to get right.
For the client, this means investing a significant amount of time and money without an immediate payoff, or worse—a completely wasted investment. John Elder says that it can take up to a year to get everything right, and for that reason, even though 90% of the models built for his clients are a technical success, only 65% of them ever get deployed.
3. Optimize pricing to maximize profits.
Traditionally, retailers have used A/B tests or bandit tests to set prices for different products and come up with the optimal price to maximize profits. The problem is that each price is set manually and is prone to human error.
Predictive analytics takes a different approach, building a model to support real-time pricing that uses input from various sources like:
- Historical product pricing;
- Customer activity;
- Preferences and order history;
- Competitor pricing;
- Desired margins on the product;
- Available inventory;
- and more.
This video shows how Uber & AirBnB have cracked the code of setting prices with different variables in the supply-and-demand equation, and how predictive analytics has helped.
Pricing Management is an ongoing process and needs to be monitored closely to avoid automated price changes that cause issues in the retail environment.
Accenture has emphasized the benefits of predictive analytics for pricing management for quite some time. They argue that implementing predictive pricing gradually allows retailers to build their analytics talent in-house.
Based on Google Trends data, many companies have followed that advice:
4. Stay properly stocked and reduce overstock.
Walmart revolutionized the field of inventory management by asking its suppliers to support real-time inventory management, also called Vendor Managed Inventory (VMI).
Predictive analytics takes it to the next level by minimizing the required/threshold inventory for a product if the predictive model sees no immediate big sales. This helps the retailers buy products that have a higher demand and a greater profit potential.
In an excellent, if dated, research paper on big data and predictive analytics for inventory, researchers identified that lack of technical skills as the biggest roadblock to widespread adoption. The rise of several vendors that offer packaged and easy-to-deploy solutions for inventory management with predictive analytics has changed that assessment somewhat.
One such example, shared by Southern States below, showed how the farming cooperative used predictive analytics to maintain sales, even with 31% less inventory on hand.
For farmers, this meant 31% less spoilage, but for a typical ecommerce retailer, it means less money spent for storage and fewer things in the clearance section or otherwise wasted.
5. Minimize fraud by proactively detecting it.
Fraud is a reality in retail. Billions of dollars are lost every year.
Any technology that can reduce losses from fraud is a breath of fresh air. Predictive analytics solutions allow retailers to analyze browsing patterns, payment methods, and purchasing patterns to detect and reduce fraud. Some retailers are even using predictive analytics with machine learning to automatically define rules to detect and prevent fraud.
Walmart demonstrated that they’re serious about using predictive analytics for fraud management, among other things, when they acquired Inkiru in 2013.
Other retailers have also used scoring algorithms for fraud management. Predictive analytics has enhanced the ability of the algorithms to catch fraud before it happens.
6. Provide better customer service at lower costs.
Customer Service is another area in which retailers have too many questions and few definite answers. Some of these questions are:
- Should the site have only chat/email customer service or is phone service mandatory?
- If phone service is required, how many reps are needed?
- Does the site also need live chat if it has phone service?
- What is the optimal hold time when someone calls?
- How do you prioritize questions from a loyal or high-value customer?
All these questions (and more) can be predicted by building a model specific to the customer service needs of the retailer. The model will gets refined over a period of time (just like other models) and starts providing more accurate predictions to improve customer service.
Linux distributor Red Hat uses predictive analytics to improve its customer service, increasing what they call “subscriber stickiness.” According to this study, they were able to preemptively provide customers with answers to problems they didn’t even know they had yet.
Hotel chains, like Marriott, are another great example of businesses that are heavy users of predictive analytics to exceed customer expectations before, during, and after a stay.
Luxury properties like Four Seasons and Ritz Carlton are always trying to exceed customer expectations, and predictive analytics helps.
7. Analyze data and make decisions in real time.
Streaming analytics is the capability to generate insights—and make decisions—in real time.
This is important. Predictive analytics that uses historical data doesn’t help retailers in a fast-paced environment. Real-time decisions influence the best day to launch a promotion, identify products that generate more sales, target segments with specific campaigns, etc.
Netflix is a well-known example of a company that uses streaming analytics to capture and analyze all customer interactions, including when a customer hit “Pause,” how many times they did so, the color of the movie title that attracted the customer, etc.
Technology platform Granify allows for this level of predictive analysis, using historical browsing data to adjust a website in real time. For example, if a visitor is showing behavior that suggests anxiety about sizing, Granify will emphasize the size chart; however if they’re more concerned about shipping, it will adjust the site accordingly.
So, how can I deploy predictive analytics?
Simply having a predictive analytics platform doesn’t guarantee success. Like John Elder said earlier in the article, accurate predictive models can be incredibly difficult to build and can take a lot of time and money.
To make sure your investment in predictive analytics isn’t wasted, work with a skilled data scientist to build your predictive models and a skilled developer to integrate your platform.
With those components in place, you have three options:
Option 1: Use predictive tools to integrate with your ecommerce platform.
Given the rise of predictive analytics, several ecommerce platform vendors offer predictive tools and plugins. These should be the first choice if your business is using one of these platforms as it’s the easiest way to start using predictive analytics without the headache of painful integrations.
Some examples are:
- Springbot on Magento is a pretty good place to start, as plans start for companies with 25,000 customers or less.
- Custora is a more robust tool set that helps increase customer lifetime value and integrates with Shopify.
Option 2: Use an open-source predictive analytics product.
If you already have the technical talent in-house, there are several open-source predictive analytics platforms, like KNIME, that allow you to create more custom solutions. Admittedly, several have been bought out by larger companies, like Salesforce (Prediction.io) and Microsoft (Revolution Analytics).
This option requires a retailer to do the dirty work of implementing the open-source solution in their environment, which also means hiring the right resources to implement solutions. Given that these are open-source products, there could be a few stepping stones (i.e. failures) before success.
Option 3: Buy a full-featured suite.
This is easily the most expensive option available—a single-user license for most predictive analytics suites is a five-figure monthly investment. However, these end-to-end platforms also provide the most functionality.
The benefit of these offerings is that they have pre-built models for different areas like fraud, pricing management, etc., and require only minor tweaks to work in a retailer environment.
Additionally, most vendors offer consulting services to help deploy the tools, saving you from hiring or contracting IT resources to do the work.
Predictive analytics is critical for retailers to succeed in today’s environment and shouldn’t be ignored.
You don’t have to enable every use case for predictive analytics, but you should pick the areas that will create maximum impact. Reviewing your desired targets for revenue uplift, fraud-loss prevention, optimized customer service, cost savings, and better insights.
The benefits will be seen after a period of time, so it’s important to deploy the models and continue to monitor and refine them until the business benefits become clear.