We predict predictions - part 2: What to look for in AI-powered prediction tools
In my last blog I put forward my prediction for 2021; it’s going to the year that AI-powered Prediction Technologies take off in a big way.
Up until this point, product recommendations in email have been relatively scarce, particularly at the individual recommendation level (even if they are dressed up as such).
If my predictions are correct, marketers first need to understand what recommendations are (I recommend reading the articles posted by our CTO & Co-founder Ricky here).
The concepts at work here are fairly intuitive but understanding what goes on in a recommendation engine is really important for Marketers to understand so they can understand what to look out for, and . Here’s some more important things to look out for (and avoid):
- Black Box Solutions: I often hear that existing technologies offer little to no transparency into the algorithms or logic that provide their predictions. This is particularly problematic for fashion & luxury brands which have heavy brand controls on what can be sent out the door. It's hard to sign off on an email when you can't visualise the end results. Make sure to really drill down into the level of flexibility and control you’ll have when you’re in the driving seat.
- Prescriptive Templates: A lot of recommendation tools have pre-defined templates that offer little to no customisation. The recommendations are typically served as low-resolution images. This limited customization / one-size-fits-all "recommendation bar" can look blocky, blurred and off brand. Lookout for a powerful editor that gives you the control to create branded, pixel-perfect, high resolution designs.
- Lack of Merchandising: Marketers need a level of control to promote or filter out specific products (at a granular level e.g. remove specific SKUs from recommendation). Recommendations need to be available for a number of use cases and scenarios such as recommending products that are in the latest sale.
- Lack of Depth: Marketers need more than the basics. Recommendations that consider all data are necessary. It is important to understand whether the technology can take data from external sources (such as CRM data) or if it's limited to specific 'pools' to make recommendation.
- Speed: If the prediction isn't fluid and up to date with each interaction, its value is significantly reduced. Simply, data processing delays are problematic. Lags e.g. overnight computation leads to experience gaps as the data is likely 'past it's sell by' at the time the customer sees it. Say for example, a customer is browsing and then returns 3 hours later to complete the purchase. The data from the initial engagement is processed and recommends the product purchased in their second interaction. This one has happened to me...
- Real-time: Live everything is imperative! Ingesting site interactions, computing the data, using live product data in that moment and serving the recommendation in real-time is the difference between a recommendation and a great recommendation. Using live product data means only available products are recommended, if the recommended product sells out, it’s swapped for another product selected for the customer that is available. Serving the recommendations in email live, means products are always updating and changing based on the latest interactions and product data.
Customers engage with recommendations & predictive content at higher rates, they keep coming back because consistently being shown content and products that they like. As a result, their lifetime value increases as they’re better served.
If you’re seriously considering the inclusion of recommendations in your strategy for 2021, you need to know about our product recommendations for email. We’ve done our homework and have created a revolutionary product recommendation for email marketers serious about serving true recommendations to their customers. We’ve focused on the shortfalls and created a product that is set to change email forever.
Our recommendations are quick to implement and users have full control over the recommendation configuration (down to individual SKU). On top of that we have a one of a kind ‘playground’ that allows you to see the recommendations being processed and made in real-time, and an advanced design studio that allows you to create your own recommendations template. You have full design control over the look & feel of the AI-powered product recommendations with the added oomph of our live data processing and visibility layering capabilities.
The platform observes the unique patterns of behaviours your customers display, as it gains a better understanding, it evolves to provide better predictions. Furthermore, this is an automated process, you design your template, set the business rules you want to follow, and paste html into your ESP. It automatically displays the best content to each individual, this goes way beyond segments or other high-level profiling, targeting predictions at a much more granular level.
What to look for in the future of AI Prediction Technologies?
I think that the next level AI Prediction technologies will provide end users with transparency into the recommendations. There is an understandable wariness of AI Tech, but I believe this comes down to a lack of understanding rather than genuine long term distrust. For mass adoption and maximisation of this technology, I feel transparency is a great ethos to uphold.
What does transparency mean in this context?
I imagine it as displaying the logic that influences the prediction to the customer when the recommendations are served. Potentially as a little tag line above or below the product or content. The idea is to highlight exactly why they’re being shown X product and the data that has influenced.
For example; People with similar preferences to you, bought products in this colour & style recently.
If you’re interested in taking the AI plunge get in touch and with us and we can discuss the best applications of this tech for your brands emails.