Unboxing a recommendation engine

If you had a beautiful user experience on any modern internet platform such as e-commerce, social network, SaaS, etc.., then it’s highly likely that the company delivered that beautiful experience using a great UI design powered by a recommendation engine.

Consider recommendation engine -RE- a smart business advisor who understands and can predict customers’ behavior and needs with certain precision and error. Therefore it has a unique view of the customer -“what do you want”.

REs also brings the possibility of serendipity finds on the internet. You may have noticed recommendations online that are in line with your preferences and are aware of the events happening around you, resulting in impulse purchases. For example, last month, I bought running shoes on Amazon, which were advertised to me during a popular marathon event in the city.

After designing a recommendation engine for a product that generates a multitude of offers for international travelers — from booking their first flight to booking local experiences to taking the cab back home, I’m explaining the entire magic trick by recommendation engines in 4 simple steps

• Identify the persona of your customers
• Learn from their experiences to establish contexts
• Build offers using the context
• Explain to the customer why was the offer made

Of course, each step is a wide area in itself so I’ll share introductory details with you.

Identify the Persona

Persona refers to characterizing behaviors of similar-looking of people and events into one. e.g: premium business travellers ( 35–45-year-old people who take business trips and have high spending power), deal seekers (18–27-year-old people who buy only when offered a discount), etc. Recommendation engines start by identifying patterns among customers to identify customers that look similar in their behavior and motivations to form a persona.

Establish Context

Recommendation engines establish contexts by learning from our experiences of the past and present, and good ones predict the future. Experiences are learned over long-term, short-term and real-time.

Long-term experiences establish behaviors that repeat over a long time such as months, years, or sometimes decades. Events such as people taking annual vacations, people buying during the Black Friday sale, millennials buying a new car every 4th year, etc are long-term experiences. I choose 6–18 months as a “long-term” time frame in my RE design since within 6–18 months I saw repeatable patterns. One can choose the time window depending on the business you are designing the RE for.

Short-term experiences establish behavior in the recent past. Events such as an increase or decrease in the number of people traveling to Japan for leisure, an increase or decrease in people reserving hotels and flight tickets to Amsterdam, an increase in the number of cars sold by Audi, etc are short-term experiences. I choose 0–3 months to notice recent shifts in travel and preference patterns of international travelers.

Future predictions are delivered by a RE by drawing inspiration from long-term and short-term experiences. For example, a RE may predict who will travel in the next 30 days, where will he/she travel, and enrich the prediction of travel with predicted preferences during the upcoming travel. Future predictions enable marketers to understand the expected behavior and therefore craft a contextually aware message. Future predictions also give a good time window to marketers in becoming part of the customer’s consideration set in the buying cycle.

Current contexts help a RE to keep a very close watch on the experiences people are having in real-time at the moment of truth. The current context monitors customer moments to validate or tweak the offers it prepared. It also monitors how accurately its predictions match the current experience and therefore improves the accuracy of future predictions

Build the offer

Building an offer for a customer involves first understanding the products being offered by the company. REs build objective evaluation of products across dimensions such as price, quantity, color, etc.

REs thereafter correlate products with customer contexts to evaluate how closely they match the behavior, persona, and current needs of the customer. This evaluation yields an offer that is presented to the customer.

Explain the offer to the customer

This is by far the most important area that requires the utmost skill and customer empathy.

You will be amazed to know the kind of correlations machines can build and therefore predict with good accuracy what you will buy. For example a series of Google search, over a period of time, for a “Local gynecologist”, “Blood tests”, “Ultrasound” and “maternity hospitals” when combined with your persona, easily establish the context that you are about to have a child. The context, therefore, sets up the advertising opportunities for “mother and baby care products”, “creches” etc.

While this is fascinating, using such contexts without informing customers appropriately, will be infringing on privacy and outright unfair. Therefore, using the context in an ethical way requires the company to be exceptionally empathetic towards customers. To do so a good recommendation engine should always explain why it is advertising an offer. Amazon does this beautifully by categorizing recommendations under these heads

• “Customers who viewed this item also viewed”
• “What other items do customers buy after viewing this item?”
• “ Sponsored products related to this item (What’s this?)”

I used the above four steps, a series of machine learning algorithms, and multiple times iterated UI design to create the recommendation engine. The UI design rests on the concept of evidence packs that translate complex algorithm’s output into as simple explanations of “Why was this offer made”.

So next time you see a highly relevant offer made to you online, remember there is a recommendation engine doing its job well !!