The AI experience: Road to spot-on product recommendations
Top-notch customer experience, improving customer lifetime value, personalized and timely: these are all hot topics, and thanks to the era of AI we are living in, more and more achievable. In this blog post, we will focus on how an extraordinary product recommendation engine can boost your business and simultaneously deliver an excellent customer experience.
In a customer-centric business, a well-timed and accurate personal touch is the golden ticket. However, if you have a large customer base, tailoring your product recommendations to each individual customer isn’t a walk in the park.
So, how to tackle this? Thanks to algorithms, you can now create highly personalized product recommendations for your customers in a timely manner.
Why is this intriguing?
At their very best, good product recommendations can be a true win-win. You as a marketer can boost your customer satisfaction and customer lifetime value for your existing customer base, while simultaneously providing excellent customer experience. There are surely some pitfalls in implementing and using a recommendation engine that must be taken into account, but the opportunities are intriguing. So, instead of focusing on often-costly customer acquisition, how about finding growth from those who already are familiar with your products?
When is it relevant to invest in a product recommendation engine?
Product recommendations get trickier when your product portfolio is large, and/or your customer base is diverse. With just a few products it can be relatively easy to make heuristic rules on which products have strong connections and are most likely relevant to specific customers. In this case, AI would likely just get in the way and take more time to construct something you already knew. Do not get me wrong, I still strongly recommend applying some advanced data analysis on your data to verify your assumptions. This also helps to notice when your data has grown so large that you need more advanced technology to keep up and stay relevant.
The larger your business grows, the harder it becomes to stay on top of it all. The connections behind purchase decisions are complex, but simultaneously it gets more and more important to help your customers find what they actually need. After all, it is more fruitful to automatically help the customer find what they are looking for, instead of making them spend hours scrolling through the internet. The good news is if you have a ton of purchase and customer behavior data, you already have the first key to success, now you only need to know how to use it. In short: if you really want to excel at product recommendations, but managing your product portfolio demands majestic juggling skills, the situation calls for some analysis and AI.
Knowing your goal is the first step
The very first thing to ponder and decide is “what do I want from this?”. There are some different aspects of the AI-driven product recommendations, and it is important to head in the right direction straight from the start. Defining which actions you’re planning on taking based on the recommendation can even affect the choice of a suitable methodology. For example, an online clothing store might want to recommend items matching the customers’ style on their front page, but also lift these products in their Facebook ads. In addition to this, they would also like to predict the most likely product categories to personalize the quick access links on the front page. These requirements help define the necessary data and select the proper models for predictions.
Here are a couple of common examples of possible aims for product recommendation engines:
- Lifting customer lifetime value: For example, finding the top 5 most interesting new products for each customer and displaying them in advertising and website hotspots to lift conversion rate. Even better, instead of just looking at the likelihood of conversion, you may also factor in the profit margins to maximize expected revenue.
- Lifting product sales: Finding the most likely buyers for different products and boosting the sales of key-value products with product recommendations. Maybe you have a low volume but lucrative product, and difficulties in recognizing the potential clients?
- Insight for future: Gaining a better understanding of your product portfolio: what products could you discard, might there be the potential for some new products? The insights gained from understanding your current product portfolio and customer base can be of the essence when making large investment decisions.
Defining the goal helps you to choose the best method and have a goal-oriented mindset. It also helps to clarify what data is needed. Data quality and quantity is highly important in any AI application, and sometimes data deficiencies might set you back from what is actually achievable with the data you have. For example, in the former online clothing store example, one must have customer-specific transactional data from different products. To make predictions better and work for also those website viewers who are yet to make their first purchase, having at least session-specific product page view information would take your model to the next level.
So, always aim for high standards when it comes to data. For complex AI implementations, data should:
have excellent quality
have high granularity
Deficiencies in any one of the three might force you to start on a smaller scale until the data issues are fixed. The good news is, that there are plenty of handy and useful analyses that can be done with data even when it does not tick all the boxes, but it really is worth it to put your data into shape.
As mentioned earlier, the main goal might even affect the method selection, and there are plenty of different ways of doing product recommendations. In general, good places to start could be a market basket analysis and/or collaborative filtering. With plentiful, high-quality data a carefully constructed neural network can overperform the simpler models. In the end, it comes down to finding the most suitable method for the defined goal with the available data.
What could go wrong?
One part of building a great AI-based system is making sure it is trustworthy. As with any analysis or automation, considering the most common caveats is important. To be fair, all cases are different, and there can be plenty of things to consider, but here are some more common issues that might arise with diving too deep into product recommendations:
1 filter-bubbles: too strict recommendations might not appreciate exploration and all recommendations are not always be correct
2 customer anxiety: too timely and accurate recommendations can be even spooky for some consumers and thus bring negative backlash
3 undetected data issues: every AI reflects the data it has been trained on, making the data quality a top priority. With poor data you get poor recommendations and not recognizing the problems with your data can even be harmful.
To some extent, the above-mentioned issues can be prevented with careful data analysis and model testing. However, to make sure the model works in practice as well, start small and for example, A/B testing with pre-existing practices is a good way to start off.
Wrap-up with few key takeaways
The gains from timely and good product recommendations can be massive and a true win-win. This requires a clear view of the goal and great data quality, but also time: building an intelligent recommendation system does not happen overnight, but often is an iterative process filled with learnings, setbacks, and improvements. As with everything, the first step is just to get started, and the rest follows.
Are you interested to discuss more on how intelligent product recommendations could boost your customer lifetime value? Get in touch!