One of the most lucrative ways to use Artificial Intelligence/Machine Learning (AI/ML) in your marketing efforts is to customize your special offers. Offers can be discounts and deals, of course. Offers can also be special messaging for automatic delivery, buying in bulk, and loyalty clubs.
Today, let’s talk about the deals and discounts kind of offers – your coupons and promo codes.*
For years, many companies have used one-size-fits-all coupons. Things like 10% Off if you give me your email address or Buy One Get One Free if you order before Easter. But now, with the help of AI, we can personalize offers to every individual on your list. How? You can do this in many ways, using all sorts of different behavior variables. If you’re first starting with personalized offers, two of the best places to begin are (1) separating your users by device; and (2) separating your users by channel, especially segregating your direct/branded traffic from everyone else. If you’re new to AI/ML projects, I’d recommend choosing one of the two. If you have several projects under your belt, consider doing both at once.
Then, look at things within your group such as new vs. repeat customers; lifetime value; what products/services they looked at; AAUS (average active user session – time spent on site actively); the number of pages they viewed; whether they’re sentiment-sensitive; location/geotargeting; how deep they were into a particular category; and so on. One of the easiest places to start is to look at Carted-Not-Converted (CNC) visitors. When you’re looking at CNC, you want to look at all starts regardless of whether they ever got to the cart/checkout. (Incidentally, this can be a jackpot if you aren’t already doing it.)
Example: Mobile devices vs. desktop vs. tablet. If you have an app or have built-out tools for Voice, create separate buckets for them too. Then, within each of those areas, look at the number of users with carts that have not been converted. Do some research – what items are carted? What does your gut say about why they haven’t converted? Price? Shipping? Availability? Something else? If you had to bet your house on it, what do you think would get them to convert? (No, you can’t answer free goods and services. Nice try, though.) Come up with 3-5 offers you’d like to test and implement them using a tool or a homegrown system. (More cart and checkout conversion tips can be found here.)
What kind of offers should you test? A good rule of thumb is to develop a conservative offer, a mid-range one, and an offer so juicy it makes you dizzy. (The intent is not to go out of business here – it’s to figure out your floor and ceiling numbers.) If you’re not comfortable with that, then develop five mid-range offers. Things like discounts, BOGO offers, free/discounted shipping, rebates, dollars off, bulk buying optimal pricing, free/discount upgrades, bonus accessories, and so on. If you already have an established offer program, use the controls as part/all of your foundation.
When you’re looking at personalizing offers, some of the most successful things to test – after you’ve built a bit of structure – are timing (when they should be presented); location (where they should be presented); and what creative and messaging you are using (how they should be presented.) This is where AI really comes in handy.
Do you need AI to do this? No. If you’re feeling conservative, kick it old school. Once you get the results and you want to break it down further, will you want an AI-enabled tool? Quite likely. AI will allow you to add more variables to the example above quickly. You’ll be able to test more and learn faster. Even better, it will continue to process and analyze all the data – including your real-time results – and recalculate/change based on its learnings.
A word of caution: start small. Build a tight foundation and then grow. It’s very tempting to do ALL THE THINGS at once, but it won’t end well. (I say this from firsthand experience.) You end up losing valuable margin dollars. You don’t need to use the example above but keep it simple. This type of project is deeper than one might think, and as you get more sophisticated, things can run amok easily if you don’t have control over your variables and your model.
What about the ~50% of your visitors who don’t use promo codes at all? Should you care about creating offers for them? Companies who have built solid personalized offer programs often find that there’s a big chunk of visitors you can get to increase their AOV (Average Order Value) with bespoke offers. The right offer(s) can also reduce this group’s number of days to sale. Definitely worth testing. On the other hand, you can also put the offer-resistant folks in a bucket for separate examination. What things do they best respond to? Sneak peeks or early access to new items? Additional support? The more you know about the user, the more efficiently you can market to them.
Please Note: I like offers with end-dates because deadlines create urgency and cause people to focus. However, I find this whole “let’s blanket everyone with 10%” to be lazy marketing. There are so many more effective choices. Whether you do any discounting depends on your overall stance about offers, but if you’re into using incentives, there’s nothing quite like the right offer at the right time on the right day.
Have questions about personalized offers? Tweet @amyafrica or write info@eightbyeight.com.
*This is an acknowledgment that I’m aware this sounds like the start of “an SEO walked into a bar” joke.
Catalogers and other offline marketers have many advantages when it comes to E-Commerce. 
