Personalization is an accepted and expected mainstay in digital marketing. Everybody understands the conceptual idea that personalization is about providing your consumers with the right message at the right time and location. However, having worked with many companies on outlining their personalization strategies, I can easily say that, while each company claims to provide its customers with a personalized experience, in digging a little deeper, they admit that they have a long way to go before they can make that claim with any real proof. It’s important to move the discussion away from the heady conceptual stage and toward developing a roadmap for actual boots-on-the-ground tactics.
Broadly, I have identified three categories that fall under the personalization tactical umbrella: automated decisioning, campaigns, and A/B testing. Only after you’ve implemented all three categories will you have bragging rights to claim that you truly personalize your customer’s experiences. These categories are media and channel agnostic, and their deployment depends purely on the technology, data, and analytical capabilities available in each of them.
Algorithmic Decisioning – They’re slow but comprehensive
I like to think of this category as optimization that happens during the resting state. This type of personalization is continuous, affects 100% of your traffic and is always on throughout the year. However, this is also the slowest to react to changes and should be used for products and offers that are considered evergreen. The end result of the algorithmic decisioning process is to come up with an algorithm for a statistical model that helps decide the best content (product, creative, etc.) to show to each visitor based on results from testing.
An example of this is a bank with multiple products, like checking, savings, credit cards, home equity loans, etc. The brand needs to decide what product to market to each visitor every time they come to the bank’s website. Since these products are going to be around for a long time (evergreen), the bank can invest in a long-term testing program that results in utilizing machine learning algorithms to figure out the best product for each visitor. Notice here that I am purposely not referring to any offers (these are usually short-lived), just broad products.
This personalization category has the highest reward-to-effort ratio, but is most often ignored by marketers. The most commonly cited reason for not implementing this is the length of time taken to reach optimum state. Machine learning algorithms and differentiated treatments require a fair amount of conversions per piece of content. This, in turn necessitates testing with no modifications in content for longer periods of time at the start of the testing process. However, companies that can tolerate an initially extensive testing period will reap large dividends with low maintenance going forward.
Campaigns – They’re exciting but short-lived
Unlike algorithmic decisioning, campaigns are moments of inspired genius. A well-crafted campaign can drive huge lifts for the targeted population. But there is a caveat: Campaigns are usually run for short periods of time, affect only a small percentage of your visitors, and suffer from diminishing returns.
An example of this would be something like a bank offering a $100 reward for opening a checking account in the next 30 days. Only a certain section of visitors are eligible for this campaign (those who do not already have a checking account), and it will expire after 30 days. In the meantime, the bank is still left to grapple with deciding what to show visitors who are not eligible for this campaign.
Wouldn’t it be great if the brand could show these remaining visitors a product that they are most likely to purchase?
The exact boundary between running an offer in a campaign vs. making it a part of the algorithmic decisioning mix is not always clear. In our example above, let’s assume that only 10 percent of site visitors qualify (very different from likely to buy) for the offer, and the offer is valid for 30 days. This time frame is too short to test and build a predictive model, so we decide to be practical and show this offer to all of those who qualify. Therefore, we categorize this as a true campaign.
However, what if 60 percent of visitors are eligible for this offer, and we expect it to run for six months? Does this still qualify as a true campaign, or do we want to consider it as part of the algorithmic decisioning mix? Either answer could have a valid justification here.
The way I would recommend separating them out is this: If an offer runs for such a long time that it needs to be targeted in order to free other visitors to be exposed to other campaigns, then this offer should become a part of the algorithmic decisioning mix. If the timeframe for the offer is short enough that we do not have enough time to test, gather data, build models, and implement and validate them, then it is considered a true campaign.
This personalization category is more unstable, and it can see high lifts as well as big drops. It is also effort intensive, as planning for new campaigns is a large creative and analytical exercise. However, this aspect is crucial in keeping things fresh and exciting for your company.
A/B testing – Buttons everything up
A/B testing is a vast topic, but it is an excellent way to get started on personalizing content for your visitors due to its relatively low cost of entry and high initial returns. While algorithmic decisioning and campaigns cover showing different experiences to different visitors, A/B testing is more about optimizing all aspects of an experience to ensure the best outcome.
Going back to our bank analogy, while algorithmic decisioning and/or campaigns may decide that “visitor A” needs to be shown the checking product, A/B testing will help us decide if the creative should feature an emotional message or a fact-based message, if the image should contain a family at the beach or at home watching TV, whether the call to action (CTA) button should be a hard rectangle or a rectangle with circular corners, whether the click-through page for the CTA button should lead to the product page or to the product category page etc. While algorithmic decisioning and campaigns optimize at a macro level, A/B testing optimizes at a micro level and ensures that all the nuts and bolts of the ship are tightened up.
Here’s a real life analogy to understand these concepts better. Algorithmic decisioning is similar to creating a wardrobe for office wear. It takes a while to craft your style, but once you settle on wearing checked shirts and dress pants from Monday to Thursday, and Polo shirts and jeans on Friday, you are 100 percent covered on all days. You don’t need to do much maintenance except adding new ones from time to time and eliminating old ones when they don’t fit you anymore or the colors fade.
Campaigns are like that costume you painstakingly handstitched for Halloween or that really impressive suit you bought to wear at your wedding. Your friends are going to remember these clothes much more than what you wore for the remaining 364 days of the year – not something you fall back on for everyday wear.
A/B testing is a special case, and can be applied for both of these clothing scenarios. This category of personalization is about understanding whether that red shirt works better for you than the blue one, whether the jeans should be a slim fit or relaxed, whether your jacket should have two buttons or three, etc. Therefore, under the constraint of the style of clothing that you have already chosen based on algorithmic decisioning or campaigns, A/B testing ensures that you get the best possible fit and style.
As you can see, all three categories of personalization drive lift in their own ways, and an evolved personalization program will utilize each of these in varying proportions to achieve continuous improvement. Can you think of any other categories that you would want to add to the list?