Sean Cook
Sean has chased the best tech from San Diego to NYC to SF working in GIS, data science and marketing. You can find him alternatively creating dazzling new analyses and driving content strategy at Periscope Data.

How ZeroCater Made Our Lunch Data-Driven: Survey Analysis and the IoT

May 23, 2017

Here at Periscope we absolutely love data, so I couldn’t help but notice when our catering company, ZeroCater, brought in this interesting 'internet of things' survey collector.

ZeroCater has always delivered great meals and responded quickly to our feedback, but for a data-driven analyst like myself this this was too interesting to ignore. By using this device they were empowering us to respond in real time to our meals. What would data like that look like? I had to know!

Collecting data like this presents unique challenges. Compared to a more traditional survey method, this will increase response rate but reduce granularity. Net Promoter Score surveys often have the same trade off: companies will collect large amounts of low-specificity data and have to decide how to make it actionable. This can be through circling back with notable responses for detail, connecting responses to product usage, or using it to target and reduce potential churn. For ZeroCater, it serves as a way to review vendors as well as collect info about the best meals.

ZeroCater was kind enough to send us our own data so I could see what they were learning about us. As a first look, I examined lunches that were rated with a very negative or very positive score. This revealed some interesting facts right off the bat. Thai food is almost universally loved, with >80% of respondents giving it the highest possible rating! Mexican, Mediterranean, and Japanese were also very high scorers.   

select
  tag
  , very_positive / surveys as very_positive
  , very_negative / surveys as very_negative
from
 zerocater_reviews
where
 primary_tag = true
group by
  1

Looking at just positive reactions, pizza and Korean food look equivalently well received. However, once negative scores are examined it becomes apparent that Korean food is a much more controversial choice. I understand that kimchi is more controversial than pizza but it poses an interesting question: To what degree must you favor the non-controversial choice? (Luckily, the answer is easy. Continue to send Periscope Data Korean food as often as possible.)

This data reveals why some type of net ratio score comparing both happy and unhappy reviewers is so important.In NPS, a 0 - 6 is a Detractor, a 7 - 8 is a Passive, and a 9 - 10 is a Promoter. A simple ratio here will work as well. We will ignore the concept of a Passive and judge our meals on the ratio of thumbs up to thumbs down.    

select
  tag
  , very_positive  / surveys as very_positive
  , very_negative / surveys as very_negative
  , ((very_positive + positive) - (very_negative - negative)) / surveys as net_ration
from
  zerocater_reviews     
where
  primary_tag = true
group by
 1

With our ratios shown in green, everything's different. Now those safe choices like pizza and American shoot up! In most industries this would definitely be a better metric for reducing unhappy customers, decreasing churn and increasing retention. In food, it is probably still the best metric, but has an additional caveat: people get bored. Were we to receive pizza every day, I suspect we’d be happy for a couple of days — but the net ratio would probably crater within a week.

For one last interesting chart, here are our net ratio scores by day of week. With only six weeks of data this might be anomalous, but it’s a number I am going to keep an eye on!

ZeroCater has only just begun to experiment with this device. We received ours about six weeks ago, and no doubt as more data rolls in our lunches will continue to improve. At the very least, I suspect we may get more drunken noodles and chicken satay in the future!

Haven't tried Periscope Data yet?
Start a trial and we’ll send you one of our famous coffee mugs.
Read More
Haven’t tried Periscope Data yet?