Using Data Science to Design Human Connection

With so many newbies wandering the studio, and more joining the ranks every week, it was becoming hard to keep all the new faces straight. Our colleague Annette, one of our co-experience directors, saw an opportunity for us to revive a studio-wide program that set up any two IDEOers to go out to lunch on the studio’s dime once a month. The idea was to help establish new connections, but busy colleagues would often cancel lunch plans and the task of scheduling (and rescheduling) soon fell to one of our beloved leadership coordinators, Biz. As a result, this benefit had become severely underutilized.

Our mission: to use data science to (mostly) automate the scheduling process so that newbies and veterans alike could easily participate in the program. The tool would need to assign “optimal” groups of Chicago IDEOers, search their calendars for a mutually agreeable time, and send them an invitation to go to lunch together. Here’s how we did it:

1. Designing the perfect group

We started by asking a seemingly simple question: What makes a good group? To find out, we practiced our design research skills in some working sessions with Annette and Biz.

2. Creating custom datasets

To build the groups, we needed some data on our coworkers. We turned to IDEO’s intranet to get a list of everyone who worked at a given IDEO location, their disciplines, and projects they’d worked on. We combined that with manually-created datasets that contained each person’s level of seniority and discipline. Using this data and the earlier criteria, we created an algorithm that randomly samples employees, generating one group at a time. Once a person is in one group, their likelihood of being chosen for another decreases; to spare their calendars, no employee can be in more than two groups.

3. Addressing the scheduling conflict

Once we have a group, all that’s left is to find a time for them to meet. Through the Google Calendar API, we were able to get a list of available times on people’s calendars. Since these times aren’t uniform, this data structure isn’t easy to work with. After many frustrating attempts to create a functional system, we realized that fellow data scientist (and Datascope co-founder) Mike Stringer had already solved this problem (albeit for a different reason). We used the Python library he created to perform unevenly spaced time series analysis, saved ourselves heaps of wasted time and energy, and learned a lesson: When surrounded by world-class nerds, phone a friend before trying to reinvent the wheel.

4. Putting the Meetbot to work

As designers, we knew that creating the algorithm was only half the battle—we also had to share our work to make it successful. To give the bot some personality, we created a mascot we dubbed Meaty the Meetbot, a friendly humanoid meatball on a mission to bring people together. We introduced Meaty (and his functionality) during an office-wide lunch meeting. The newly-minted Meetbot was better suited to the IDEO Chicago of 2018 than the old program: it scaled to accommodate the 45% increase in employees, and infused a little data science flavor. For the big finale, we sent out the first batch of invites all at once, so everyone received them at the same time.

Since Meaty’s initial rollout in late March, there have been 112 group lunches in Chicago (according to Google calendar), bringing together everyone from support staff to directors to designers. Word spread to other studios, and the Palo Alto and San Francisco offices are on deck to start their own group lunches. Meaty has also virtually connected IDEOers around the world, scheduling calls between design researchers in Cambridge and New York, and handling the time-zone-juggling challenge of scheduling global calls between Munich, Shanghai, and Tokyo.

Although Meaty the Meetbot was born as a silly thing to tinker with until we had “real work” to do, it’s had real impact. For us, it’s been a reminder that even in seemingly inane projects, data and algorithms can be used in service of human interactions. By letting a computer take over the grunt work, we freed up some (way more valuable) human time, making room for people to connect and create. Maybe there is such a thing as a free lunch.

Annette suggested that we create groups of three, rather than two, based on social science research that finds triads are more conducive to sparking creative collaborations and community than dyads. That fact inspired us to call the program “Meet ‘n Three,” a play on the traditional Southern combo meal offering of a protein and three sides.

Together, we decided that groups should comprise three people whose paths don’t cross day-to-day, while also striking a balance of design disciplines—not all interaction designers or all data scientists. And an ideal triad would mix people who were new to the studio with those who had been around longer. With these principles ironed out, we got to building the algorithm.

The group is then assigned a score for its “novelty,” which includes the number of disciplines represented, the variety of seniority, and the number of projects shared between group members. The algorithm optimizes the score for an entire month’s worth of groups using a Monte Carlo approach. (In other words, we create many more groups than we actually need, and throw away all but those with the highest scores.)