Site search engines in the e-commerce world have a tendency to operate in isolation. They receive a query and return results. This works well till the site reaches a couple of thresholds. These are:
- when you cannot improve the accuracy or quality of search results anymore through rule changes.
- when the human effort required to manage promotions and merchandizing is enormous.
- when the buzz around certain products is so high and variable from day to day and week to week, that by the time business rules change, the popularity has ebbed or the pattern has changed.
Search working in isolation is a problem
When
site search works without feedback from the website, we are basically assuming a static product catalog and an unchanging buyer behavior. This is certainly not the case in most real-world stores. Merchandisers work around this by tracking stats such as:
- Top selling products
- Latest deals
- Highest margin products
- Highest discounted products, etc.
Then they manually create promotion rules to cross- and up-sell these. The underlying assumption here is that the search engine at least allows the category manager or the merchandizer to:
- Track and classify the products based on the above criteria.
- Create and manage such business rules for each of the above.
- And if you’re lucky, provide a management dashboard that enhances merchandiser productivity, accuracy and turn-around time for such requests.
However, this approach does not scale if you have to manage such rules across many many stores and categories. The process also breaks down across an even larger number of products and SKUs. At scale, the dynamism in buying behavior and the trends in the
product browsing and sale are huge. Transactions across catalog vary daily, weekly or monthly and a product relevant for a query today, is not necessarily relevant for the same query next week. In this scenario, it is humanly impossible to create a rule for every conceivable trend or product grouping. Even if you could, it will take a significant effort on part of the category manager or merchandiser.
Achieving accuracy at scale is better off automated
Lets say you put in the human effort needed to build the merchandising rules for promotions, cross-selling and up-selling based on the data you’re collecting. At the scale I am talking about, the risk of a human creating an incorrect rule is pretty high. Furthermore, with the variable patterns in buying and browsing behavior, keeping the rules updated is an even tougher task. This is where e-commerce search needs to automate:
- the collection of behavioral data
- quantifying the data into metrics
- then, feeding the metrics into the search system for creating merchandizing rules, on-the-fly.
Behavioral data in search
Generally, behavioral data is usually thought of in relation to
recommendation engines or personalized email marketing campaigns. These are truly fascinating applications but they are also secondary solutions to the larger problem of e-commerce conversions. This is mainly because search remains the primary means of looking for products on the site. Therefore, even when a
recommendation system aims to sell accessories to a product or sell similar products other users have purchased, it is an aside to the primary product that the visitor is searching for or had searched for in the past.
Similarly, while a personalized email marketing campaign can increase stickiness and bring back a user to make a purchase, it needs a critical mass in terms of traffic to generate the kind of personalized data that will make an email marketing campaign effective. With behavioral targeting, the search can be used as a subliminal channel for cross-sell, up-sell and running other marketing campaigns. In a future post, I will go into the mechanics of using behavioral data in search. Please let me know your thoughts in the comments section below or please email me directly at: shehjart@unbxd.com
