Attention Cannibalization
If you come from a business or marketing background, revenue cannibalization is something you are familiar with. It’s the concept of a business launching a new product line that is creating revenue by taking revenue away from another product line in the same business but not growing the revenue/profit pie. If you are an engineer or designer and you haven’t heard about revenue cannibalization, you just did.
There are good reasons to do this. You can do it because you know one type of business is dying and a new model is becoming more prevalent (e.g., people are subscribing to Office 365 in lieu of buying a full license of MS Office). Another reason is to convert users from an older version of a product to a newer version (e.g., Gillette blades), either to prevent customers jumping ship or because the new product will present newer revenue opportunities in the future.
There are bad reasons to do it, and that’s most common than not. It’s when a company is too big and lacks upper management coordination, so many units are building similar or substitutes products and competing against each other, or the same product has multiple add-ons that compete for the same dollar. In those cases no new value is being created for the company as a whole, just revenue is moved from one product to another.
What if we think of the products as business, and the features as products?
Let’s take this a level down. Imagine now that you are talking about a single product and instead of revenue you change the metric to attention. This is attention cannibalization. The idea that a new feature might be cannibalizing the attention of an existing feature, without creating any new attention to the product as a whole.
This is dangerous and I’m going to give away the conclusion first! It’s dangerous because you are led to believe what you are doing is improving the product and the overall experience and it to “validate” decisions made months ago, at the cost of not moving the product forward at all — you are actually moving it sideways.
Product Managers, Designers or Developers, might talk about their success in terms of usage of their feature. Maybe there was a feature that didn’t exist before, and after they launch they see that 30% of users use it right after sign up. Pop the champagne! Or, they change how a feature work, and they see that the usage for that feature went from 2 times per visit to 3 times per visit. Holy crap, a 50% improvement in usage!
The reality is that in a system, looking at the production value of a single component is worthless to understand the value of the whole system. But wait, it could be worse than that. It could be a new feature took attention away from a higher value feature. Maybe you added a new way for users to customize their avatar and that choice was enough to justify some folks from not upgrading to the premium version of the product. Revenue starts to dip, slowly and given the noise in the revenue signal and the fast speed in which new features are released, it can become quite harder to figure out the root cause of a drop in revenue, or a drop in overall conversion, engagement, virality or retention.
Another problem to pay attention to is when the attention cannibalization has a delayed impact. You add new feature X and you see a 25% usage on that feature on a daily basis and no drop in revenue or any other feature usage, so you might say to yourself that was worth it, but then a couple of weeks later you start experiencing a drop in retention. For example, when Twitter decided to auto-follow accounts for new users, and it seemed like it was a good idea because there was a jump in initial engagement with the product, but overall was a bad idea since retention was affected by it. This is the short-term gain for long-term loss strategy that you might not even know about it.
At this point you might be asking how to identify this problem. First and foremost, you must look at the product as whole. Did conversion, engagement, virality and retention went up or down for the product as a whole? Second, you must look at cohorts of data and see what’s the longitudinal impact of each feature launch? Third — and this might go against all the new trends in Lean Startup — don’t ship too fast. Making dramatic changes to your product without understanding the impact of previous changes is the sure way to lose track of what’s working and what’s not — clearly this is much easier said than done in this day and age.
