11436 SSO

Five Reasons why Bolt-on Analytics is a Bad Idea

Apr 29, 2015

Bolt-on analytics are products that only solve one particular use case, and only pay lip service to integration with other enterprise analytical systems. As an analytics product builder, I cringe whenever an enterprise buys yet another analytics bolt-on product.

Bolt-on analytics marketing pitches go something like this: “Just install our SDK/tag in your website or mobile app, and get XYZ benefits in 10 minutes.” The main buyers and users of bolt-on analytics products are online marketing departments and, grudgingly, IT departments.

Before I started building analytics products for a living, I worked at Amazon.com, where almost all analytics, email targeting, push notification targeting, and recommendation systems were built from the ground up, so there was very tight integration. Then I moved to a large enterprise company where almost everything consisted of bolt-on analytics products with little regard to integration whatsoever.

Here are five reasons why I believe that using bolt-on analytics products is a bad idea:

1. It forces you to upload private data to your vendors.

While working at that large enterprise I mentioned, we wanted to correlate website traffic to order performance. Because our web analytics vendor made it so difficult, if not impossible, to download data into a (Teradata) data warehouse, the only solution was to upload the enterprise’s order history to a web analytics vendor to perform the correlations. The security team’s response was something along the lines of “You want to upload $50 billion worth of order transactions outside our firewalls? Are you nuts?”

2. It only uses data that the bolt-on solution collects.

If you're trying to do predictive analytics, getting as much data as possible for making a prediction is important. However, to avoid the pain of data formatting, many bolt-on solutions only work with the data that their SDKs collect directly. This is particularly terrible when you have multiple channels, such as web, mobile, and store. For example, If you buy a bolt-on product for mobile recommendations, you end up losing transaction information from the web and in-store, which is vital to providing much more relevant recommendations.

3. You end up with a spaghetti of analytics products.

A tangle of disparate analytics products slows down your website or app, costs lots of money to maintain, and prevents you from truly understanding what your customers are doing. Worse yet, at some point, enterprises wise up and want to get the data into their data warehouse or Hadoop system. Bolt-on analytics products often have poor data export tools, and, worse yet, there’s an incentive for them not to give you the data (especially the raw data).

4. Omnichannel analytics becomes impossible.

Most enterprises have multiple digital channels: web, email, mobile, and in-store, for example. To analyze the data across all digital channels, there must be a way to integrate all analytical results. For example, you should be able to integrate a universal customer ID across all interactions. Poorly designed bolt-on analytics products don’t let you tag the customer event datasets with common identifiers such as a customer ID, making it impossible to analyze the data afterwards.

5. It makes for a poor customer experience.

At the end of the day, it’s all about the customer. Especially for targeting and recommendations, having different solutions for email targeting, push notification targeting, web recommendations, and mobile recommendations means that there is no way to easily coordinate the customer experience across different digital channels. There isn’t even a way to coordinate the customer experience across major functions like marketing and customer service.

Five years ago, bolt-on analytics solutions were inevitable because data warehouses simply were not equipped to handle web and mobile data. Today, with the rapid adoption of Hadoop and the commodity hardware to run Hadoop, there is absolutely no reason for enterprises to forsake control of their own data. In addition, not all analytics vendors are evil, and lots of them do want to integrate with your data warehouse. Coming up, we’ll discuss technologies and strategies to help you make your own data warehouse or Hadoop data lake work together with other analytics products.

Photo: Peter Castleton/Flickr

Scaling Microservices