What is a Recommendation Engine?
In a nutshell, a recommendation engine /recommender system uses data to generate tailored recommendations to an individual. The aim is often to increase cross/up-selling or encourage users to consume more content. Showing your customers products they will like without having to go searching for them is a great customer experience. This can generate extra income for your organisation and improve customer satisfaction. It’s reported that around 35% of Amazon’s purchases come through recommendations. Additionally, around 75% of content watched on Netflix are recommendations by their system. This reportedly saves $1Billion per year in marketing costs.
With the increasing availability of data, coupled with cheaper, more powerful computers, we are seeing recommendation engines becoming commonplace. They are no longer restricted to “giants” such as Amazon, Netflix, or even Tesco with their successful Clubcard scheme introduced in the late 90’s. We now see recommendation systems within smartphones, serving “News” items you may be interested in. On your smart TV under the banner of “People who watched X, also watched Y”. And even on your favourite shopping website checkout basket “Items often brought together”.
Recommendation Engines can be deployed in a variety of places to help you innovate and provide marketing sparkle. For example, here are several typical use cases for Recommendation engines.
Some Examples of Our Recommenders
Identify the Next Best Action / Ask. Planning donor journeys, i.e. this donor looks good for Regular Giving & Legacy.
Used to serve dynamic content within email campaigns.
Jobs you might be interested in, (see LinkedIn or Glassdoor).
Recommending products for a specific customer. For example, similar products to those bought/interested in. Products similar users bought. Products often bought together.
Recommending content to users, people who read X, also read Y. Because you watched X you may also want to watch Y.
Including recommendations with basket abandonment emails, giving customers more options to buy.
Because you swiped on “Nigel”, you might also like “David”.
Applying Human Intelligence to Artificial Intelligence
At Adroit we firmly believe that Artificial Intelligence is extremely powerful. However, it relies on Human Intelligence to guide it. It is this human intelligence applied by our team of data scientists and marketing professions which make us different. Whilst some are happy to sell “off the shelf” recommendation tools, we understand that it’s human knowledge, intuition and intelligence which helps us to provide beneficial and innovative solutions. Indeed, a computer can’t tell the difference between a spurious correlation and a valid one. Or whether a recommendation is appropriate, even if it would be blatantly obvious to a human.
An obvious example are cases where recommendations are based on products previously purchased or “also viewed products”. Especially if they have quite a long repeat buying cycle. For example, if someone has recently purchased an espresso maker. A product which should last a few years. Serving the same customer other espresso makers based on the context of “People who viewed this also brought, probably isn’t the best idea. But, serving them related products such as milk frothers, mugs and coffee pods would be.
In a similar vein, we need to ensure the recommendations are relevant. Given enough data, the problem isn’t about finding relationships, it’s about finding meaningful relationships. Just because the data shows high correlation between people purchasing a particular brand/model of Espresso maker and purchasing a ride on lawnmower. It doesn’t mean that making this recommendation would be appropriate.
It’s also likely that at certain times you would like to push certain products. For instance, you have too much stock, the season is ending, or there is a particular opportunity to sell that product? In the example where a key influencer has been seen wearing a garment or using a certain product.
What are the Challenges with Artificial Intelligence?
There are also data challenges which need to be addressed, such as how do we deal with new prospects / customers who we know nothing about? How do we stop recommendations becoming too tight for individuals, giving them no variety? How do we prevent the recommendation engine recommending out-of-stock items? And how do we optimise the recommendation engine to increase value and margins? How do we keep recommendations fresh for a customer?
Adroit can help you ensure that the recommendation systems can be flexible and adapted to changing needs.