Surfline’s LOLA dashboard is a great example of pretty ‘advanced’ visualization in the consumer internet space that works well because it matches the user’s cognitive model.
Every surfer holds a conceptual model in his/her head for forecasting the quality of waves at their local break involving swell period and height, tides, the ocean bottom, wind, weather, water temperature and more. Before sites like www.surfline.com, dedicated surfers would devote entire rooms to collecting maps and data on geography and the weather, so they could be on the water when the best waves rolled in.
When I lived in Florida, where there are no waves all summer except those created by tropical cyclones, we would pray for hurricanes all summer… checking the National Hurricane Center site every few days. Ike and Josephine made my 2008 (while Ike ruined 2008 for many others). Having trained hard for six months, I paddled out into double overhead waves from Ike at St. Andrew’s Park. I wasn’t ready. The water was crystal clear, schools of dolphins were playing inside multi-story waves and I was on an elevator I could not get off. Terrified, I fought the current for an hour, repeating to myself, “I just wanna go home, I just wanna go home.” I didn’t get a single wave. I drove home.
Fortunately, when Josephine’s swell hit my home break two weeks later, the merely overhead waves seemed small compared to those of Ike. A friend (and witness) talked me into a hundred-yard, overhead right. Best wave of my life, complete with bragging rights.
Surfers take waves pretty seriously, so Surfline developed a predictive model for surfable waves called LOLA. It is informed by buoys (swell period, direction and amplitude), weather stations, satellite imagery, user ratings and more. It is unmatched in quality for forecasting good waves.
A free forecast created by this model is available to all users, but the dashboard costs extra. The LOLA model, and its interface are fairly complex. Because the data model and the interface match the user’s cognitive model, Surfline is able to expose the complexities of the LOLA model as an up-sell under a freemium model. I’ve been a paying customer on and off for several years. It is a great product.
Predictive models are powerful and compelling. They seem magical. They tell the future! They are essential for canning complex data into simpler forms for mass-consumable web pages, to address a core need or to answer an important question. In this case: ‘Are there good waves coming? Should I plan to go surfing?’ Most users want only a fast, at-a-glance answer to this question, a rating from Poor to Epic. But some users… Surfline’s core audience, want the deeper insight that exposing the raw data gives them.
This is an opportunity, and it brings us to the point: If a predictive model is likely to address a problem… you should provide end-users with interactive access to its inputs before you sink a lot of time building the model in the first place. If you can’t build a dashboard that nails the first five questions a user would ask of your data through interactive visualization… odds are you are building the wrong model to solve their actual problem. That is because unlike academia or other industries, building predictive models for the consumer internet is an inherently iterative process. At the intersection of data and product, iterative validation isn’t optional.
How do you get that? Get out of the building, talk to users (or in surfline’s case, be your user), get in their heads, build a small, interactive dashboard fast, that lets your users touch the data. Show it to everyone, get the end-user involved in product development until you nail their first 5 questions in a row with descriptive statistics in simple charts, and then get it on the public internet. Collect feedback, and iterate more.
Now you’re ready to build your model. And odds are… its the right one. The one that will answer the right question and create the most value. And your users will help clean your signals for you ;)