To say Amazon screwed up big time with their disastrous product, the Amazon Fire Phone, is a bit of an understatement…. they took a $170 million write-down largely attributable to unsold inventory. But why did it go so wrong? And how can product managers avoid being the next product disaster using data science?
We’ll never know for certain why the Amazon fire phone failed to resonate with consumers. But, we can try to help other product developers avoid making the same mistake. In this article we will to explore some of the modern data analytics and business intelligence techniques from the world of data science that should help you minimize the risk of producing a product flop.
Sticking with our theme of the failed Amazon Fire Phone, we’ll look at how this process might work in the cell phone market.
Deciding what product to build
You could rely solely on your intuition and experience to make an educated guess about what will sell. But history is written only by the victor, and for every Steve Jobs, there are thousands of people producing products you never end up hearing about. Relying on your gut instinct is dangerous. Especially when in today’s “Age of Data” you can combine data insight with domain knowledge to come up with a strategy based on facts and figures.
The cell phone market is a relatively saturated one. There are lots of options for consumers to choose from, and many products only offer slight variations against competitors. Because phones essentially all do the same thing, for a product to be a success, it has to appeal to a particular niche (preferably an underserved one).
Take stock of your data
As a major cell phone manufacturer; you probably already have a plethora of data at your fingertips, such as which of your phones have sold well, in which regions and with which demographics. That information is good, but it’s not enough. To make a good decision you also need to consider what’s happening in the wider market.
The good news is that now, because of the internet, there is more data at hand to add to your data stack and execute decisions on. It may be hard to access, but companies like import.io have begun to make that data available to business on a commercial scale.
Supplementing your data stack
There are several different methods for gathering external data on your competitors, customers and markets. You could collect this information manually into a spreadsheet, but that would take a very long time and would likely be incredibly boring. If you are technical, you could write a web scraper to do this task automatically, but scrapers can be quite complex and they’re not very reliable. If you’re not technical, you could use one of a number of scraping tools to collect this data automatically without actually having to write any code.
All of these methods, however, are pretty time consuming and require you to do all your own data cleansing on the other end. Which is why we’ve started offering Data as a Service, which gives you up-to-date, accurate and pre-cleansed data direct from the source.
But lets get back to how you can specifically use data both internal and external (once you’ve got it) to make a better product.
Text and sentiment analysis
Tweets and product reviews are a great way to gain insight into what your potential customers think about the current products on the market. Knowing how your target audience feels about the current products can help you identify gaps in the market and features that are loved as well as ones that they could do without.
But, text isn’t easy to process. Text analysis tools like MonkeyLearn use machine learning and natural language processing to help you identify the relationships between keywords and concepts.
“Sentiment analysis is a powerful example of how text analysis automatization can help companies build better products with unique features. One of the most common use cases of sentiment analysis is business intelligence; companies use sentiment analysis to discover insights about their products and services.”
– Raúl Garreta, CEO and Co-Founder at MonkeyLearn
Visualise data for insights
Representing your data in a visual way helps you to identify trends and outliers in a much simpler way. This allows you to easily evaluate new product opportunities and prioritize product features as well as understand product trends.
Visualising data sets, particularly when you combine internal sales data, with external data, sourced from the web, often reveals new and exciting insights.
“The best data visualizations are ones that expose something new about the underlying patterns and relationships contained within the data. Understanding those relationships — and being able to observe them — is key to good decision making.”
– Julie Steele, O’Reilly Media
Predictive Analytics and Machine Learning
Possibly the most exciting thing you can do with data, is predict the future. Machine learning services like BigML allow you to build models that can predict what will happen if you change a variable. You can predict the effects of different pricing, adding features or certain markets.
Once you’ve built a model of your market, including as many market variables as you can, like pricing, product features, sentiment, geography etc, you can run as many simulations as you want until you get a set of features, pricing and a geography that is statistically likely to do well in your target market.
“With a few clicks and no coding whatsoever you can build a very visual decision tree to predict your target variable. One of the cool things about the model is that it will tell you what features/traits are most useful in predicting your target variable and which variables can be practically ignored.”
– Atakan Cetinsoy, VP of Predictive Applications BigML
Don’t underestimate the human factor
Of course there is always going to be room in the world for intuition, outright cool factor and traditional market understanding in developing products. Ultimately, the success (or failure) of a product is reliant are far more factors than data science can predict (yet).
However, when you are making a $170 million gamble, you need all the information you can get to mitigate as much risk as you can. Data might not be all you need to build a killer product, but rest assured you won’t be able to build one without it.