An unprecedented Black Friday is coming. Will consumers run masklessly back into stores and clean out the shelves? Only if you have a sale on Clorox toilet wand refills. Will they swarm your app or website? Maybe.
As one data scientist put it, “All predictions are wrong. Some are lucky.”
Some retailers will likely continue to see diminished demand even on Black Friday—such as those specializing in men’s slacks. Most retailers have already seen their traffic move online. For this Black Friday, they need to be effective, ready, resilient, and cost-effective.
Effectiveness is the conversion rate of visitors to buyers. Ready is measured by throughput and latency, even when lots of traffic comes at once. Resilient is handling whatever 2020 throws at it. Finally, in the cloud era, cost-effectiveness is eating the world. (I felt that eye roll.)
Effective front page
The front page needs to sell to be effective. Any website analytics will tell you the most visited page is the homepage. Analytics will also tell you the homepage has the highest bounce rate. For the front page to be useful, its content needs to be informed by context.
Return customers hitting your front page are usually the easiest to convert. They did not come to your site through a pay-per-click ad, a search, or a referral. They came directly to your site because they know you. When they arrive at the homepage, it should not be just any promotion or set of items. It should be offers tailored specifically for that customer.
For first-time visitors coming in for an offer or via other referrals, the homepage should also be customized. AI, machine learning, or just plain old counting should calculate what other items were bought by people who visited that offer. If the offer is brand new (a cold start), use data from existing customers who clicked on that item.
For first-time customers coming to the homepage without context, the page should also not be static. As the day wears on, the content should rotate based on what people are buying or searching. Additionally, the front page should make offers to these first-time shoppers to encourage them to identify themselves.
The straight keyword search is dead. As a repeat customer, if I type “shoes,” the search should know that I do not mean heels. My customer profile and my past searches should inform my search results. I should have a way out of this personalization by being more specific, but if I type “shoes” I should see running shoes and flip flops, based on my past purchases.
While the site may not know a first-time shopper’s preferences, the context should inform future searches. Meaning if I type “skis” and then “goggles” and then “boots” I should see snow boots and ski boots and not cowboy boots.
There are plentiful tools for personalizing search and homepages. Some commerce platforms such as BigCommerce or Shopify have much of this built-in. Third-party vendors such as Coveo and Lucidworks offer solutions for this sort of signal capture and machine learning personalization.
Just for you, we are offering 20% off on de-icing equipment. Seriously, I have had vendors offer me de-icing equipment for all of those frozen winters we get here in Savannah, GA. Context and personalization make offers compelling. If I searched on something, clicked on it, and started to leave the site, the smart play is to offer me something similar or a discount on what I was viewing.
If I have searched several related items such as a glass carboy, tubing, malt, and hops, as a result, perhaps I should be offered a bundle such as a homebrew starter kit or the most popular homebrew item.
No matter what, offers should be personal (Who am I? What do I like?) and contextual (What else did I search on? Am I leaving, searching for more things, buying something?). Many tools do this kind of personalized offer, including the tools built into commerce suites. There are other third-party offerings such as those by Justuno or Privy.
Ready and resilient
In 2020, retailers should use cloud computing, have an elastic architecture, and live on the edge. Most retailers have already moved their core infrastructure to the cloud. Ideally, this should either be a commerce platform or a combination of SaaS and cloud-native solutions.
Content delivery networks (CDNs) like Akamai, Cloudflare, or Netlify have now existed for decades. Images, static content, and other resources can be easily cached at the edge of the network to reduce the load on your web server. Even five years ago, CDNs were the luxury of kings or at least the Fortune 500. Today, they are affordable for the rest of us, and there is no good reason not to use them for a modern website.
The next step is to ensure the site’s search and database architecture are elastic and resilient. While most ecommerce platforms include search, these search tools may not scale when all personalization tools are enabled. It may be necessary to investigate other solutions. Those solutions should replicate and have some way to scale and to fail gracefully. In short, they should have a modern cluster architecture. There are still many ecommerce sites that use decades-old search technology, especially Endeca. These tools were not designed for modern traffic or the cloud.
As for databases, retail is often “overserved.” A NoSQL database that can relax consistency will meet the needs of most retail use cases. Meaning you do not need to pay for a distributed transaction unless the specific use case requires it. Almost any of the NoSQL databases can replicate to multiple data centers and provide sufficient high availability.
My recommendation is to choose a multi-model database that allows for quickly capturing low-latency signal data, retrieving product data, and running real-time analytical queries. There are cloud service provider offerings such as Microsoft’s CosmosDB and third-party vendor products such as Couchbase. Barring that, you could cobble a few different databases together, but that involves paying more for ETL and dealing with greater complexity.
Even for retailers with more traffic, margin is strained, so no one wants to pay for capacity they are not using. The term “autoscaling” is overused and stretched out. Ask any vendor (be they database, search, or otherwise) and they will tell you that they can scale up. But what happens when you want to reduce the number of instances you are using? Can that happen without an outage or excessive performance degradation? A lot of technology goes into scaling up. A lot more goes into scaling down.
The questions to ask are, “Can I easily remove nodes?” and “How long does it typically take to rebalance for my type of data and load?” You should also ask, “Can I turn off components that I am not using?” and “What is the idle cost of the software just running?”
Boom fizzle pop
It is 2020. Assume that anything can happen. Things seem to be going boom, fizzle, or pop every day. The stakes are higher this year. They require investing more imagination into “What could go wrong?”
What if your cloud service provider fails in a target country? Multicloud is not just a buzzword. For global retailers, it is a necessity. There are geographies where a preferred cloud service provider does not reach or where there is only one data center. A global retailer needs alternatives. Nor can you assume your provider will never fail. While cloud providers fail rarely compared to the self-hosted data centers of old, and failures of multiple availability zones are rare, it does happen. Having it happen on Black Friday in 2020 could be catastrophic for some retailers.
Does the site’s architecture easily allow components or services to be shut off without the user noticing? What if your whizbang personalization component or third-party service chokes? It is one thing if the homepage becomes more generic or does not show an offer on exit. It is a bigger problem if visitors see a 404 inside an iframe or get an hourglass or spinning disc.
Keep it together
The great news about 2020 is that it is almost over. Staying up and converting customers this Black Friday is not fundamentally different than in years past. If anything, there are more options. It is just a matter of deploying the right personalization, search, and offers to maximize conversion rate while ensuring an adequate cloud-native architecture that can handle multiple failures while scaling to meet demand.
Copyright © 2020 IDG Communications, Inc.