The e-commerce revolution was built on the back of digital advances and automation, but is the next generation of machine learning and AI friend or foe?
Machine learning is the method today’s systems use to learn about you and the fundamental things that make you who you are. It looks beyond generalizations and characterizations on your age or gender and realizes that you’re a huge fan of cartoon dogs and emojis, but only the ones that have the same hair as you.
In the e-commerce space, machine learning is being used to slowly learn what customers prefer and how they want to see information to get them to make a purchase. It tests and adapts, using new options and information to slowly refine the best way to reach your customers.
It’s an amazing tool with plenty of opportunities, but hiding within that potential are some risks if you’re not paying attention. Let’s look at what can help or hurt.
Intuitive Search and Display
One of the more exciting applications of machine learning is going to be the ability for you to search for the things you want and need. Right now, to find a product on an e-commerce store, you have to search for it using the words you prefer and the e-commerce store owner must hope they used the same words for the search results to match what you want and what they offer.
Machine learning cranks that up by providing support for a broader set of synonyms at the most basic level. Smart machine learning looks for synonyms in the nouns you provide as well as similar phrases people use for the same type of problem. A leaking faucet and a dripping sink should pull the same results, but they don’t always in today’s search.
The extra layer that machine learning adds on top of all this comes from its capability to learn your site and metrics. So, smarter search engines can prioritize click rates and existing conversions, put products with higher customer ratings at the top, or even restrict results based on what’s currently in stock.
It’s possible to deliver dynamically different page content using search results and even display items and containers, from ads to sections on what customers also liked, based on each individual customer. The downside, or potential threat to your business, is that systems that are not closely monitored could limit the display of some products and search results.
If you let your system fine-tune continually without checking, it could stop promoting a product that isn’t selling, and the continued lack of sales would push any promotions or links down even further. Don’t let a runaway search hide any products.
The Potential of Chatbots
Recall any of the times you’ve recently had to call your bank, credit agency, or someone’s whose robot voice told you to pay attention because the menu options may have changed. On the Internet, whether it’s your e-commerce product pages, FAQs, carts, or any other location customer’s visit, you can ensure your customers never have to go through that frustration thanks to smart chatbots.
By understanding unstructured data — answers to questions like “how was your day?” instead of “what’s ten minus two?” — chatbots are now able to provide a realistic conversation that doesn’t feel fake or frustrate people with non-answers. Your customer service representative can now answer questions 24/7 with a wide range of data thanks to smart chatbots.
Chatbots mean small and mid-sized companies can provide round-the-clock support without needing to pay someone those hours. Machine learning comes into play because it allows a chatbot to be programmed with general information and respond to customer queries. As it interacts with more people, machine learning will enable your chatbot to learn the specifics of your online store and service.
Initially, chatbots can empower you to direct customers to basic information like shipping options, colors, size charts, and other formulaic options. Eventually, as you employ more sophisticated machine learning, you can have chatbots that identify potential upsells, ask customers questions to deliver coupons, or address long-term needs of your customers.
Unless you work in customer service, it’s hard to imagine chatbots being a threat. The tricky issue here is the system that learns your customers and their preferences. A significant amount of chatbots come from third-party services and they may be collecting or using some of your customer data for the purpose of answering questions or generating social media posts automatically.
Make agreements airtight when it comes to how data can be used and ensure that it is anonymized, or you may have upset customers who realize that they got an email for a service or one that includes their address after they’ve been on your site.
Pricing and AB Testing
Beyond chatbots, machine learning creates opportunities for a variety of virtual buying assistances that can do anything from email you when airfare drops below a certain price or reminds you to re-order your water filter every 30 days. E-commerce brands can take this data review a step further to look for opportunities such as coupons to drive down prices to a certain point or for savings ahead of a specific holiday or holiday season where your goods are purchased in order to encourage increased spending this year.
If I know that you buy t-shirts online when you can get two for under $25, I can send you a deal that gets shirts to that threshold right now. And next month, I can try with an offer that gets the price down to $27 and see if you’ll bite.
Machine learning removes a lot of the guesswork and manual labor required to identify those thresholds and send offers accordingly. Using profiles and purchasing habit data, systems can teach themselves to create a potential sale that meets a variety of threshold criteria around margins, existing inventory, and repeat business.
Higher levels of personalization require more customer avatars or marketing personas and ultimately this could be a large set that because it is hard to generalize new customers. So, there’s a need to balance cost ranges on products as well as how long an offer lasts. Customers may be upset if they see one price, don’t immediately click and then cannot find the reduced price again.
Value-Adds: Fulfillment Improvements
Machine learning has a significant potential to provide customers with something a little extra. We believe that shipping often makes the biggest difference and machine learning can not only determine preferred shipping options but test when a free bump to something faster will increase your sales.
Plus, smart systems can use your existing data to ensure that those promises match inventory levels and shipping time based on what your suppliers or distributors can match. You also have the opportunity to build in scarcity and promote sales — you might’ve seen something like this when an e-commerce giant says that if you order in the next 20 minutes you can get it tomorrow.
Bots also can provide consistent updates on package status and do trend analysis to help show the commerce brand when those messages are received, desired, and effective.
The true value-add of machine learning is that it can prevent over-promising and ensure you’re working on what best delivers an improvement for your website and the shopping experience you offer. Smart use of data and machine learning can improve your operations at every point of delivery, from when customers arrive and search your site to prompting them for purchase decisions and following up with reliable service and information.
Implementing Machine Learning in Your e-commerce Offerings
Now that you’ve seen the possibilities of machine learning, you’re ready to sign up and get it rolling for your company. Great! That’s the easy part. Adding it to your site is the hard part. There’s a lot to learn and review so that you can make your products, services, and site smarter.
Your IT team or partner will need a few things to get started on a machine learning implementation for you:
- A database or access to your existing database where the end-application can access the information it needs to know about your brand.
- The programming language, which can impact which services and partners are available based on their API.
- A description of the algorithm you want that is as complete as possible, including the problems or needs you want addressed.
- Examples of services or elements that you liked or operated similarly to what you want so your developer has a yardstick for comparison.
- The size of your audience. For example, if you need to work with a large set of customers or are planning a distributed application, you’ll most likely end up with a Hadoop ecosystem, while smaller sets of data and uses can rely on Java and C++ for the machine learning engine.
Creating a list of definitions, goals, needs, and uses will help you build a plan to bring to a developer or to give to your internal team to figure out next steps.
The good news in all that is more and more companies are lowering the bar to entry thanks to APIs you can start using immediately. So, there’s a chance someone has put together a package that you can use to achieve the results you want.
For example, Google offers a Mobile Vision API that allows an Android app to use the device’s camera to scan barcodes, recognize text, and detect faces as well as basic emotions or attributes.
There are also a variety of machine learning programs that have broader uses and you can define them as you go. Choosing an engine that operates like a decision tree will support A/B testing because you can feed it the ads or emails you’re using, information about the interactions or audience, and what you consider successful or unsuccessful results.
In this use case, an engine that works like a decision tree will eventually be able to learn what ads get people to click. By varying the coupon in your ad, for example, it can learn what dollar amounts or percentage off deals are more likely to lead to a complete sale. Eventually, you can use it to learn which deals are most likely to generate the highest relative sales value, and the engine can be used to set parameters for your marketing campaigns while adjusting coupons based on real-time results.
That’s a long road, but it is doable with technology available to you today.
Start with Chatbots and a Smart Partner
If you want to wade into machine learning slowly, we’d recommend you begin with chat options on your website because there are a variety of existing projects you can build on for a more immediate turnaround.
Head over to GitHub and you can find a host of chat bot engines that you can quickly power up and train. Training data is also available for free and purchase in order to assist you. Use a training walkthrough to see how you can generate responses based on what are essentially collections of known word strings and conversations.
By using or purchasing training and a language basis, you can have someone program the chatbot to answer common questions and then slowly begin to learn more about your customers. Your team or your development partner can also provide the chatbot with information on what to share and promote.
Chatbots with a machine learning basis are written in languages like Python so you’ll be hiring someone to help you unless you’re familiar with the language and code. Your professional partner will also be able to make recommendations on technologies like Google’s TensorFlow, which is useful if you’re building from scratch.
Today’s data world involves a variety of complex systems, and machine learning is one possible catalyst that will propel some e-commerce brands farther into the technology realm than they initially expected, while boosting their profitability too.
Realizing your potential will boil down to choosing the right technology and the right partner or team.
About the Author: Jake Rheude is the Director of Business Development for Red Stag Fulfillment, an ecommerce fulfillment warehouse that was born out of ecommerce. He has years of experience in ecommerce and business development. In his free time, Jake enjoys reading about business and sharing his own experience with others.