For my first tests, I wanted to check if people would give recommendations when asked by a friend. Therefore, I decided to ask my friends on facebook to give me their best recommendations for different categories of cafés and restaurants in Zurich (9 categories in total).
The questions were the following:
- What is your favorite Asian restaurant in Zurich?
- Where can we get the best sushi in Zurich?
- .. the best pizzas… ?
- .. the best steak… ?
- .. the best kebab.. ?
- What is the best bar?
- What is the best café?
- What is the best place for a brunch?
- What is the best restaurant for a special occasion?
Within less than 24h, I received 255 qualified recommendations from 34 people.
I sorted out all the replies and cumulated the same answers to finally create a ranking of each category. I then made a cross-check with what TripAdvisor has to offer. Interestingly enough, in all our lists of answers, the most voted recommendation never appeared in the TOP 5 of TripAdvisor(!). This is for me a key moment, it already says that my “algorithm” provides different results than the current online alternatives.
So, what are the key findings:
- My network has a different opinion than TripAdvisor.
- Limiting the number of inputs per user increases the quality of the results.
- The more specific the question is, the more valuable the list gets.
- People give recommendations in exchange for valuable insights.
- The added-value of these lists are perceived as low-value by locals (expected).
I finished my experiment with a blind test opposing my network’s results VS TripAdvisor. Here is a sample of the survey:
There are a bunch of people who took this survey. Unfortunately, I did not split the participants based if yes or no they have given recommendations in the first place so I am not sure if the data is not distorted (my first fuckup). I will need to do the test again with people in my network who have NOT given recommendations EXCLUSIVELY.
The participants voted that my network had better results 73.5% of the time VS 26.5% for TripAdvisor.
How to move forward:
- I want to have some qualitative interviews with some people who gave me recommendations. I need to better understand the motivation behind their action.
- Send the blind-test to 30 people who had nothing to do with the recommendations in the first place to confirm the results.
- Define the first version of my product based on the key findings and customer interviews.