Of your first-time buyers, on average only 20% of them will return and make another purchase. Increase that by just 10% and the cumulative effects mean you’ll double your revenue. But how do you get someone who has bought just once from you to buy more items?
Yes, you can hope they’ll buy the same thing again but actually you also need to promote your breadth of offering. You can do that in a generic fashion easily – but that will have limited impact and all the problems of generic emails of list fatigue and unsubscribes kick in.
Then you have to ask how do you personalise those emails when you only know about the one item. How does that item in isolation tell you enough about their overall taste to recommend other categories and products?
Here are how 2 modern data techniques help you group customers to do just this.
AI Product Recommendations
You’ve all seen on Amazon the ‘People who bought this also bought this’. This is the type of algorithm that can help find the statistically most probable items the customer is likely to buy next based upon their first purchase.
Not all of these product recommendation algorithms are equal though – there are a few pitfalls you need to avoid.
Firstly, many AI solutions are designed for converting customers in the same session. Using this type of algorithm, any subsequent post purchase emails you send are only going to include more of the same type of items as they just bought. The example we always go to here is our work with Jones Bootmaker where if the customer has just bought a pair of black work shoes, the last thing they will want to see is more black work shoes.
The next pitfall is around how you control the results to make them work in your campaigns. For example, can you ensure there is a diversity of product types in each email rather than all products from one category? Can you ensure that the same products aren’t repeated again and again for the customer? Can you add merchandising rules to work alongside your business priorities such as upweighting own brand products that have a higher margin?
There has been an explosion in AI martech solutions over the last few years. It only takes a matter of minutes for an ESP developer to go to cloud service such as Amazon Web Services or Google Cloud and plugin with a generic product recommendations service just so that feature can be added to the list. That AI algorithm is then used for all their clients, from high end luxury fashion through to low-cost travel agencies.
The same algorithm simply cannot be optimum – marketers need to be able to control the data and rules used to align with their specific business, just like the ‘Flavours’, our AI Product Recommendations algorithm, and ‘Recipes’, our create your own personalisation algorithm tool allows.
Category Affinity Analysis
Personalising a subset of products is one thing, but what about deciding which top level departments, categories, brands or travel destinations are likely to be relevant to send?
Most brands will have buying and merchandising teams wanting to push specific product lines, or perhaps brand managers that want to launch a new range. All of them want dedicated email activity, not something buried in the middle of another email. But if you gave everyone the email they desired to the whole database you would be sending subscribers 3 emails a day!
You might target these emails by only sending them to customers who have bought from that topic area before. But that will be a tiny volume - most people have only ever bought one item so the volume won’t fulfil the revenue requirements for the campaign.
Instead, we can use affinity analysis to score each customers likelihood of buying from this category. This prediction is done in a similar way to the AI Product Recommendations of based upon what other people who also bought or browsed the same items, what else did they do.
These prediction scores allow us to broaden who we can reach and predict the likelihood to buy. By ranking potential customers by their prediction score we can then work out how many we need to send to achieve what possible percentage of the revenue compared to sending to everyone.
For example, it might show that to achieve 90% of the revenue that a whole database send would generate, we actually only need to send to the top 15% of prediction scores.
You might be thinking that isn’t increasing product breadth as only 90% of the potential will buy. But this approach allows you to send many different emails across a whole range of brands, departments, categories or destinations in a week, each to a small percentage of the potential list so you don’t overwhelm your database.
If you send 6 emails that generate 90% of a full blast to 30% of your list each time, instead of 3 full blasts in a week this will deliver an 80% increase in weekly revenue.
Does that sound like hard work? That is where experts with 20 years in email personalisation can help. Drop us a line if you want to find out how we can make this easy for you.