In this post, we share the creation process of Metaview Wrapped, as an example of how Metaview experiments with ideas, and how we ‘Do Things That Don’t Scale’.
At Metaview, we help companies run amazing interviews. We achieve this in a number of ways including training interviewers with automated interview shadowing and coaching them with personalized, contextual feedback.
To do that, we are building a SaaS product, at the cutting edge of conversational AI. The Product Operations team at Metaview often runs experiments and does things that don’t scale, but more importantly does things that will delight our customers. In this blog post we discuss the creation of ‘Metaview Wrapped’, a video we generated for our customers to share their interview highlights over the year.
One of our operating principles is to ‘Optimize for rate-of-learning’. This means that we treat learnings as the highest form of currency for delivering future impact. We run experiments, question thoroughly, and talk to customers over and over.
One of the most common types of advice we give at Y Combinator is to do things that don't scale
Last December, as the festive period approached, we ran one such experiment. Inspired by Spotify Wrapped, we decided we would attempt to create a short video for each of our users, summarizing interview data from interviews they had run during the year. After all, Mariah Carey did once say something along the lines of, ‘All I want for Christmas is personalized data-driven interview insights’ - I think - although I’ve not heard that song in while. We called this video ‘Metaview Wrapped’.
The Creation Process
Step one was coming up with a storyline for our video. This involved some quick querying in SQL to determine:
- Who are most frequent interviewers on the Metaview platform?
- Which conversational data points from interviews might be interesting to share?
- How could we transform that data into a fun story?
Once we had a list of our recipients, their key conversational interview metrics, and the storyline for the video, we needed a vessel for delivery. Google slides was ruled out early due a lack of flexibility in animation. Keynote it was (for non-Mac users, Apple’s answer to PowerPoint.)
We next began designing what the video animation would look like within Keynote.
Once the full video was fleshed out, we needed to find a way to populate each video for each individual with their own interview data. There were 1000+ videos to populate so manually updating each Keynote presentation would have been very laborious. Here’s what we did:
- We created a main parameterized Keynote deck. That essentially meant replacing values that needed to be filled in for each customer with
- We created an AppleScript which would take as input a CSV file with each recipient’s name and their conversational metrics. This script would then create a copy of the main parameterized Keynote deck, and fill in each recipient’s parameters with the datapoints from the CSV file. This AppleScript took a 2-3 hours to get right, and it saved us 80 human hours at the very least!
We now had 1000+ videos created with each recipient's unique data. Now the question was how would we send them to the recipients?
But before that - we found a royalty-free banger to add as a backing track!
Back to the problem of sending the emails to recipients:
- The first problem here was that the file sizes were too big. Each video was about 1 minute long and had lots of animation.
- To solve this we piped all of the videos through FFmpeg to convert m4v to mp4. This helped us to reduce each video’s file size from 49.2MB to 9.4MB.
- Our next step was to then figure out the best way to automate the sending of an email to each recipient with their own video attached.
- We used YAMM (Yet Another Mail Merge, a GSheet add-on) to achieve this. By setting up a Google Sheet with the recipient's name, email address and a link to their video (this link was automatically added to the GSheet using a script), we could bulk send the emails with the correct attachment on each.
And voila, we had sent 1000+ personalized videos with interviewer data, with less than a couple of days work involved. See the end result below:
We didn’t know if we would be able to run this experiment ‘at scale’ when we first decided to play with the idea of Metaview Wrapped. In the end, our experiment worked out, and we had some fun in the process. Some of our customers quite liked their videos too!