A very simple product philosophy guides my thinking on any new web product: develop products where the interactions of every user improve the site for the next user. This simple guiding principle can have profound implications for how you build products. But first, a little history…
The experience curve (this section from Wikipedia).
[The experience curve] states that the more often a task is performed the lower will be the cost of doing it. The task can be the production of any good or service. Each time cumulative volume doubles, value added costs (including administration, marketing, distribution, and manufacturing) fall by a constant and predictable percentage.
In the late 1960s Bruce Henderson of the Boston Consulting Group (BCG) began to emphasize the implications of the experience curve for strategy. [3] Research by BCG in the 1970s observed experience curve effects for various industries that ranged from 10 to 25 percent.
These effects are often expressed graphically. The curve is plotted with cumulative units produced on the horizontal axis and unit cost on the vertical axis. A curve that depicts a 15% cost reduction for every doubling of output is called an “85% experience curve”, indicating that unit costs drop to 85% of their original level.
The interaction curve.
How can you improve user experience by taking every action and inaction as a signal to improve your product? While many products have put things in place (e.g., Digg) which have greatly leveraged user interactions, I am suggesting something that benefits from every user’s every interaction.
What I am proposing here is that instead of developing viral marketing, that entrepreneurs should fundamentally change the way they think about their product. How can you turn every click into insight? How can you turn inaction into insight?
Examples of the interaction curve at work.
1. Search spell checkers.
This example has been overused, so I will keep it very short. Google and Yahoo! propose spelling alternatives (as in the example below) based on a statistical analysis of previous queries by other users. While many people have talked about how Google has mined historic queries to suggest alternatives, I would guess (and hope, but I don’t know) that they tune these recommendations based on every click AND based on every non-click.
2. Content filtering.
In this example of a recent Digg comments page, two of the top three comments are hidden because they have received so many negative votes. For example, in one of my previous projects we used 2 flags from independent IP addresses to hide a comment and then sent it to customer service to be reviewed. However, like Digg, we allowed users to click “show.” What I have not yet seen, but have always wanted to do is to show trusted users a piece of content and then scored it when it was viewed and *didn’t* get flagged as inappropriate. Again, the most sophisticated solutions today typically focus on user actions. User inaction often speaks as loudly as action.
3. Content targeting.
I read in a blog post somewhere that 1 in 500 events that could be surfaced in your Facebook News Feed are surfaced. With the proliferation of the feed, the biggest problem is not going to be getting content. It’s going to be finding the right content for the right person at the right time. How could we apply the interaction curve to content discovery? Content surfacing logic should be built from the get-go to do A/B testing at an atomic level — even if you lack the content and users in the early days, retrofitting a system once you do have content and/or users is hard.
What are the best examples you have seen of the interaction curve?
2 responses so far ↓
Jeremy Pepper // August 7, 2008 at 7:54 am |
One word: Madden Football.
There was a profile in USA Today about the game – it’s 20th anniversary is this year, and they started on the Apple platform.
The game continually improves and updates, partially from feedback from the gamers.
Not sure if that’s a right fit, but I think for your post, the gaming industry probably has a bunch of examples that would work.
mtrifiro // August 7, 2008 at 2:51 pm |
This is a great post. I am a big fan of using statistical data to surface intelligent options to users, but be careful when using popularity as an indicator.
The caveat is what I call the “Olive Garden” syndrome. Olive Garden restaurants are hugely popular but their food is spectacularly bland. This is by design. Back when the chain was owned by General Mills, a business writer interviewed one of their executives and asked the secret to their success. The executive said (I paraphrase), “Relentless product testing with thousands of customers.” But how do you pick the sauces that decide to place on the menu? “We test hundreds of recipes. We pick the ones that are least-offensive.”
Least-offensive? What restaurant would admit this is how they select their menu items?
For Olive Garden, the formula works. You can bring an entire family to the restaurant and everybody will find something they, uh, don’t hate.
There is a lesson here. If you develop a product and you need to appeal to the widest range of customers, offering “least offensive” might work.
But don’t confuse “least offensive” with “best.”
Least offensive might look like the right answer (for instance, you are a mass market restaurant with unskilled labor in the kitchen, you need the same dish to appeal to parents and kids, and you don’t have the ability to adjust your recipes on the fly for each diner). Frankly, it *is* the right answer if you don’t know much about your customer, but it will rarely be the right answer for *me* or any other individual.
The trick is to know your customers. If you know your customers, and can segment them meaningfully, then popularity analysis within each segment can yield great results. As you slice your segments narrower and narrower, and as you get better at knowing what each segment values, then a strange through-the-looking-glass inversion occurs: Least offensive starts to converge on best because popularity can be ranked on dimensions that are important to each individual.